There is, in a sense, nothing more to say about this latest attempt on Trump. We wrote about political violence in America after Trump was shot, saying that no matter your politics, it would have been worse if he had been killed. We wrote about it again after Charlie Kirk was killed, collecting data to show that only a tiny minority actually approve of political murder.

Yet misperceptions persist, and the best takes I’ve seen so far are those that seek to correct them. Counting left vs. right attacks misses a general loss of trust in the political system. Attackers tend to be more educated not because we’re teaching people violence, but because the more educated are also more politically involved. And the many who called the shooting a hoax — despite hundreds of journalists in the room — underscore the need for trusted media.

Look to the system

Rachel Kleinfeld’s post is the single most useful piece I’ve seen on the shooting, not just because she’s a global expert on political violence, but because it refuses the bait of quantifying which side is worse. Instead it’s an overview of the state of American political violence in 2026.

There was violence from the reactionary right in the 1950s, then from the radical left in the 1960s and 1970s, then from far-right militias in 1990s-2000s, and recently we’re seeing more on the left again. Over the decades the atrocity leaderboard has swung back and forth between political poles, but

the best way to understand what is happening today is that left/right ideology is breaking down. Instead, most people who commit political violence now are facing mental health challenges, and are drawing from a mix of ideologies and positions – remember that the first Trump would-be assassin was also interested in shooting then-President Biden. Increasingly, political violence is committed by people who are angry about AI, corporations, pedophilia, and other issues that do not break down along a left/right spectrum.

…

Today’s assassins often have less in common with the left-wing Weathermen of the 1960s or the right-wing militia movement of the 1980s and 1990s, and more in common with school shooters who are driven by a sense of nihilism and hopelessness.

The Americans most supportive of political violence are no longer just “young,” they’re young and politically engaged, “those who are angry about our politics, but do not see a path to resolve issues through normal means.”

Young, educated, and angry

As always after an atrocity, there was a statistics fight over who was responsible. Last October, Rob Henderson posted a now-viral claim:

Relative to Americans with a high school education, Americans with graduate degrees are twice as likely to support political violence. More education = higher support for political violence. It couldn’t be otherwise.

Alas, here’s where Elon Musk comes in. On Saturday he reposted this chart with the tag “Grad school indoctrination camps.” Indeed, it is true that education correlates with support for political violence. Just not for the reasons that Musk thinks and not nearly at these percentages.

We have repeatedly noted that support for political violence is a lot lower than simple surveys suggest it is, for a variety of methodological reasons. Sean Westwood, who runs the Polarization Research Lab, has made much more careful measurements, and brought out his own data in reply:

The rank order replicates with better data, but the % supporting partisan murder is much lower. Importantly, this is passive support and not willingness to actually murder.

The most important thing Westwood is doing here is deflating the numbers. But he’s also acknowledging that yes, there is more support for political murder among more educated people. The correlation is real. Contra Musk, I don’t think grad school “indoctrinates” killers. For one thing, psychological traits are much stronger predictors of support for partisan violence than ideological extremity. However, it is definitely true that more educated people are more politically involved generally.

Also, Westwood is pointing out that both sets of numbers define “support for political violence” as passive ideological agreement. The gap between “violence might be justified in some cases” and “I will personally pick up a weapon” is enormous, and conflating them is itself highly polarizing. The number of potential killers out there is way, way smaller than Westwood’s maximum of 5.1%.

Alas, Westwood’s correction reached far fewer people than Musk’s repost. For that matter, the actual perpetrators of attacks on Trump and Kirk don’t fit the educated-leftist profile. Only Allen, the current suspect, has even a Bachelor’s degree.

The conspiracy is now the story

Tangle’s coverage, as usual, curates coverage from Blue and Red, but it’s the “my take” section which is most interesting this time. Unfortunately, one of the more common reactions (mostly, but not entirely, from the Blue side) was that the shooting was a hoax. By midday Sunday, there were more than 300,000 X posts using the word “staged.” It certainly didn’t help that the Trump administration immediately used the shooting to argue for the the construction of the beleaguered White House ballroom.

Even sitting politicians got in on it, with Rep. Jasmine Crockett (D-TX) publicly musing “Maybe it’s fake… who knows.”

This isn’t good. Tangle editor Isaac Saul writes:

Who knows.

Well, for one, the room full of journalists who heard the gunshots and saw the shooter apprehended on the ground are incentivized to find and share the truth. If you don’t believe their reporting, you can always watch the security camera footage of the shooter running through a group of very clearly unsuspecting federal agents yourself. Or you can read characterizations about the shooter from people who went to school with him.

…

There’s a difference between thinking The New York Times has a left-leaning bias and thinking reporters at The New York Times are so bad at their jobs they can’t accurately report on an event they were all in the room for.

His point is that, yes, of course the shooting was horrific, but the collapse of trust in the media, even for an event that took place in a room full of journalists, is itself a big problem.

Quote of the Week

The best indication so far the WHCD Trump shooting wasn't staged?

No evidence yet of huge insider bets just before it happened on Polymarket or Kalshi :-)

-Maarten Schenk (see also this context)

The ascendance of Trump and other populists around the globe has spawned many narratives to explain why, including this one: men are sick of feminism so they vote for authoritarians to restore their lost status. Seva Gunitsky’s The Incels Are Taking Over is the most recent attempt (after many previous articles) to unify the manosphere and the strongman into a single phenomenon.

Two questions are worth asking about a claim like this. The obvious one is is it true? The less obvious one is what are the effects of believing it’s true? During the same period that this became a popular narrative, American gender relations fell to an an all time low. Could bad narratives be part of the reason that it’s harder than ever for men and women to find love?

The central argument, as Gunitsky puts it, is,

The emergence of far-right personalist rule is not just a symptom but a direct result of the crisis of gender politics: the collision between global gains in women’s status since the 1960s, and the psychological and material displacement of men who had organized their identities around traditional gender hierarchies.

…

gender grievance is not a superficial aesthetic layered on top of right-wing populism but one of its key drivers

We all know the stereotype: resentful young man says he voted for Trump because feminism went too far. Gunitsky is a serious conflict scholar who points to the gender gap in votes for Trump, Bolsonaro, Duterte, Orbán, and Erdoğan, and all his numbers are correct — the correlation is real, and in the right direction.

But this story is a causal story, and there’s a cost to being wrong about it.

Does sexism change votes?

First of all we need to be precise about how we’re measuring male grievance. When people talk about the politics of “incels” it’s usually rhetorical rather than literal, meant to evoke a certain contemptible stereotype (which is not really accurate anyway, as incels lean slighly left). But let’s switch to a standardized way of measuring this, the “hostile sexism scale,” a survey with four questions asking whether women a) complain about discrimination when they lose in a fair competition, b) seek special favors under the guise of equality, c) fail to appreciate fully what men do for them, and d) are too easily offended, each on a 5 point scale.

I pulled the 2024 American National Election Survey (ANES) data to plot the correlation between voting for Trump in 2024 and scoring above and below median on the “hostile sexism” questions.

Less sexist people definitely voted for Harris more, while more sexist people definitely went more for Trump. Interestingly, there is little difference between these patterns for men vs. women.

So does this mean “hostile sexism” is why people vote for Trump, or for right wing populists in general? Here’s the problem with that theory: you can replace “sexism” with “education” and see a similarly strong correlation. Or you can look at the number of immigrants in each county and the vote share, also a clear correlation. Etc. So which is it? What’s the driving factor?

Two recent studies test this hypothesis more carefully. First, Abrams and colleagues surveyed Gen-Z voters across 24 European countries. Among those 20-25, men vote right more often (37% vs. 24%). “Sexist attitudes” were measured by agreement with similar survey items as above, but these attitudes only accounted statistically for 10–13% of the voting gap. This limits the size of any potential effect. In other words, everything other than sexism accounts for 87% of the left/right difference in male/female voting in Europe.

Another recent paper tests causation directly. In 2013, an experiment with 111 undergraduates showed that “inducing masculinity threat increases support for war, homophobic attitudes, and endorsement of dominance hierarchies.” This has been the lab-science link for the masculinity-causes-populism narrative for the last decade. This new work is a 2,774 person replication on a nationally representative sample. And it doesn’t hold up in this higher-powered study. They could not replicate the effect, writing “we observe no consistent evidence that masculinity threat alters political attitudes.”

The cultural connection between male resentment and right wing populism holds up. As Gunitsky notes, all of the influencers in the recent Inside the Manosphere documentary voted for Trump, and all of the recent right wing strongmen are hyper-masculine. But neither influencers nor politicians are representative of regular people, and the evidence that men vote for right wing populists because of gendered resentment is thin.

Bad narratives shape bad perceptions

The male/female gap in politics is real, and men do tend to vote more to the right. However, the best available evidence suggests they (mostly) don’t do this because they’re sexist. The story that they do may be unhealthy for a number of reasons aside from the fact that it’s not true.

Narratives matter. Teaching young women about gender stereotypes makes them less likely to enroll in STEM courses. And bad narratives may similarly explain some of the mental health gap between liberal and conservatives. Similarly, it seems bad for society to strengthen politically charged gender stereotypes. Love in the time of polarization is already fraught, and America has falling levels of dating, marriage, and sex among the young.

Conflict narratives don’t just describe a fight, they are part of what creates the sides. The causal coupling of male grievance to far-right politics is a polarizing story. It tells feminists and anti-populist liberals that their antagonists are the same — and such coalition-building is probably the intended effect of repeating the idea.

However, it’s too easy to confuse “sexism is higher among Trump voters” (true) with “men vote for Trump because they are sexist” (doubtful). Going further to “all Trump-voting men are sexist” is a logical fallacy, but it’s intuitively plausible because it matches stereotypes. Likewise, this narrative tells young men who are uncertain about their politics that their irritation with modern gender norms already puts them in a camp.

Meanwhile America is in a romantic recession. The recent essay Pride and Polarization is a nice discussion of the demographic realities of politics and love, noting that 70% of Democrats now say they wouldn’t date a Republican.

The result, as National Marriage Project director Brad Wilcox has pointed out, is that there are now two liberal women for every one liberal man, and two conservative men for every one conservative woman. With young Democrats more than twice as likely to say they won’t date a Republican as the reverse, the problem is especially acute for conservative men—and, conversely, for liberal women. The intense tribalism shuts out millions of couples who might otherwise be compatible

…

People with different political views can be compatible at the individual level. Most people’s direct political involvement, no matter how heated they get in conversation, consists only of casting a vote roughly once a year, or even once every four years. There is much more to life, and any couple will need to struggle through conflicts much more concrete and consequential.

There is a connection between male resentment and right wing populism — anyone can see that in populist leaders and manosphere influencers. But among regular voters it’s hardly universal and it doesn’t seem to be causal. We have to be more careful with the demographic and causal specifics, if we want to avoid stereotyped perceptions and unhealthy narratives.

Image of the Week

Part one examined the political composition of X's user base. This part takes up content moderation: what policies changed, how much “hate speech” is on the platform, and what we can and cannot know about such questions. Part three will investigate whether X's algorithm itself is biased toward the right. Cross-published with Tech Policy Press.

When Elon Musk acquired Twitter, he promised expansive freedom of speech, so it was no surprise when he loosened content moderation policy and reinstated thousands of previously banned accounts. No one disagrees on that. A much more difficult question is what effect this had on the platform, and how we could even know with any degree of precision.

Musk and X repeatedly claimed that hate speech went down post-acquisition,and that the platform’s new “freedom of speech, not freedom of reach“ approach is working, suppressing hateful content algorithmically even when it isn’t removed. Researchers, civil society organizations, and EU officials pushed back, pointing to study after study showing hate speech went up.

Both sets of claims may be correct. The disagreement may largely reflect a measurement difference: external researchers can only really measure how many hateful posts were produced; Musk claimed to reduce hateful impressions. Whether the platform’s demotion system actually reduced how often users saw hate speech is a question that requires internal data—data that X has never shared and has made it harder to obtain. This article lays out what the available evidence actually shows, and where it stops.

How content moderation changed

The most granular public record of X’s policy changes comes from the Platform Governance Archive, which tracks edits to platforms’ help center documents. An analysis by the Humboldt Institute for Internet and Society (HIIG) published in October 2024 provides the clearest overview.

Within days of the acquisition, X dropped its COVID-19 misinformation policy and stopped removing “hateful” content, algorithmically suppressing it instead. A general amnesty reinstated most previously suspended accounts, and the Trust and Safety Advisory Council was disbanded. Later, in April 2023, restrictions around misgendering or deadnaming transgender individuals were removed. A few months later, the crisis misinformation policy (covering false information about armed conflict, public health emergencies, and natural disasters) was removed along with the election integrity policy (covering unverified claims of election fraud and premature victory declarations).

The removal of the election integrity and crisis misinformation policies were substantial changes; these applied to some of the most controversial types of content to moderate, after Twitter blocked links to the Hunter Biden laptop story in October 2020, and restricted other types of content in the aftermath of the 2020 election. Yet through all this, the core definitions of “hate speech” and other violating content remained essentially unchanged (with the exception of misgendering and deadnaming). As HIIG put it,

Although X’s definitions of permitted and prohibited content have remained largely consistent across many policy areas, the consequences for violations have shifted significantly.

The result was “a noticeable shift towards softer sanctions across different policy areas,” not an elimination of all rules (and if you really want to get granular, this is the PDF for you). The basic question is whether those softer sanctions produced harsher outcomes.

Is hate up or down?

When it comes to this question, it depends what you measure. Most attempts consider posts, impressions, and removals. It’s hard to define “hate speech,” and on top of that there are the algorithmic questions of how to define it for a computer. I’ll try to briefly explain what each of the available datasets are actually measuring, and keep using quotation marks around “hate” to convey the many normative choices inherent in the use of that word.

Almost every external analysis of the volume of hate speech on X relies on counting posts, either by using a monitoring platform like Brandwatch, or simply counting items in the platform’s search results. This usually means that “hate” ends up being defined by a list of keywords. The first such study found that “hate-terms” went from 84 hate-term tweets per hour in the week preceding acquisition to 398 per hour after midnight on October 28, the day after it closed. This soon died down significantly, perhaps near to baseline levels. A more comprehensive data set covers March 2022 to January 2023. This one also starts with keyword matches, but also included—by manual classification—whether the use of “hate words” was “hateful” (“You [bleep]!” vs “I just got called a [bleep]”).

The data reveals people changing their behavior in anticipation of the acquisition, using more slurs some weeks before the change. The first peak in anti-LGBTQ+ posts in the Spring of 2022 (pre-Musk) was largely driven by the word “groomer” and coincided with discussion of the “Don’t Say Gay” bill. The upshot here is that the acquisition did have an effect, but it seems it was somewhat temporary and possibly not larger than the effects of contemporaneous political events.

Another post-counting study used “a suite of … classifiers trained to identify antisemitic content in line with the IHRA definition.” This found a general increase that did not collapse back to baseline, at least not by February 2023.

Crucially, this study also measured engagement per post, meaning likes and retweets, and found no change after the acquisition. If Musk’s “freedom of reach” demotion system was working, you would have expected engagement per “hate post” (in this case antisemitic posts) to fall.

The best data we have on this comes from a peer-reviewed study by Hickey et al. which goes to June 2023. The authors used Google’s Perspective API to identify toxic posts and compared rates of “hate” posts to a baseline sample of random posts to account for potential user activity changes over this period. The headline is that hate speech posts were up approximately 50% in the post-acquisition period. But then hate posting collapsed in the summer of 2023.

That spike in April 2023 was probably driven by the Dylan Mulvaney backlash after the Bud Light post (the company paid a trans influencer for an endorsement). Once again, the acquisition had an effect but political events seem to have mattered more. Also, “hate” posts started rising before the policy changes, meaning that we can’t cleanly separate what Musk did from what the users did.

However, this study also found that the number of likes on “hate” posts rose by 70%. That again suggests that there wasn’t less reach, contradicting the new speech-not-reach policy. (Engagements are just a proxy for impressions, but X added public impression data shortly after the acquisition, so we know that total views are approximately a fixed multiple of engagements.)

Musk’s impression data

Musk’s counter-evidence consists of two graphs. On November 23, 2022, he tweeted: “Hate speech impressions down by 1/3 from pre-spike levels” with this graph:

He posted this graph again on December 2, saying “Hate speech impressions (# of times tweet was viewed) continue to decline, despite significant user growth!” and promising that the @TwitterSafety team would release this data “weekly.” It does not appear that they did; the only further data I could find was this graph from March 2023:

These are the only charts available that are based on actual impression data, because only X has access to full impression data. Unfortunately, no methodology was ever provided, nor a definition of hate speech.

What we do have is a March 2023 analysis of all public posts by analytics firm Sprinklr. Twitter made its internal list of over 300 slurs available to Sprinklr, and the company analyzed every post with one of these words from January and February 2023. They found that “toxic” posts received ⅓ of the impressions of “non-toxic” posts.

The @TwitterSafety team said Sprinklr found even lower numbers of “hate” impressions than they did, because Sprinklr used a context-sensitive classifier while Twitter counted simple keyword occurrences. Confirming this, Sprinklr said it found a lot of false positives from its slur search.

This appears to be direct evidence for the speech-not-reach policy working as advertised. The external findings that found evidence against this claim are most likely measuring somewhat different definitions of “hate,” or examining a different time period. Relatedly, this graph only goes through February 2023. It might therefore be perfectly compatible with the trends as in the Hickey graph (above), which also shows a reduction through February—but then a big spike in late March, corresponding to the Bud Light boycott.

There’s one final impressions data point from X in a July 2023 post that says:

Labels have been applied to more than 700,000 violative posts that fall under our Hateful Conduct policy. … Compared to a healthy post, posts with these labels—or restricted posts— receive 81% less reach or impressions and we proactively prevent ads from appearing adjacent to content that we label.

This 81% figure is in the same order of magnitude as Sprinklr’s 66% reduction in impressions.

X has finally started publishing transparency reports again, which show a fall in account suspensions for “hate speech” from 104,565 (first half of 2022) to 2,326 (second half of 2024). But so far, X hasn’t published more “hate” impressions data.

What happens when a user reports a post?

Two reports from the Center for Countering Digital Hate measured not the volume of hate speech but whether X acted on it when reported. This is a different question again—about enforcement rather than prevalence.

In June 2023, CCDH flagged posts from 100 users and found that the platform failed to act on 99% of “hateful” posts as of four days later. In September 2023 they reported 300 more posts and found 86% still up a week later, including ones pertaining to Holocaust denial and calls for racial violence.

These studies say nothing about volume or impressions. They specifically test the user reporting mechanism and its connection to content removal. This is a tricky thing to get right, as users will gang up to censor each other if the “report” button is too closely connected to post removal.

That September CCDH report also found major brand ads running adjacent to “hateful” content. X sued CCDH on the basis that the organization illegally scraped data and that these “cherry-picked” examples mis-represented the X user experience. The case was dismissed in March 2024, when the judge wrote that the lawsuit was “about punishing the Defendants for their speech.” X appealed in April 2025.

What we know and what we can’t

So there you have it. Measures of hate speech production spiked substantially after Musk’s acquisition, then settled down again—eventually, and perhaps not in all categories at the same rate. Musk says impressions spiked but then settled at lower levels than pre-acquisition, maybe half as much. There is one external analysis that confirms this impressions story for January-February 2023. It’s quite possible all of these narratives are simultaneously true. Everyone is using different definitions of “hate,” so it’s hard to tell.

Generally speaking, I suspect measuring impressions is the right way to assess the effects of “hate” speech, because the mere existence of hateful words in a post somewhere doesn’t indicate they are harmful. The internet is and always will be full of awful things, for those who go looking. Unfortunately, this is also the hardest statistic to measure from the outside. Impressions data is included in the X API, and sampled versions of the full firehose are available in Enterprise plans—starting at $42,000 per month. So in theory, anyone could track these numbers, but it would be terrifically expensive. With the death of the Academic API, the platform made itself unmeasurable shortly after it changed its policies.

Notably, the European Commission decided in July 2024 that X does not comply with its researcher data access obligations under the Digital Services Act (DSA), and then fined the company €120 million for this and two other DSA breaches.

Meanwhile, if anyone inside the platform has conducted further measurements on “hate,” they haven’t reported any of it. Scattered transparency reports do tell a tale of fewer account removals under the new regime, sometimes many fewer. Perhaps this is what one would expect from the softer “downrank don’t delete” policy—if it’s not serious enough to delete, then it’s not serious enough to get you suspended.

Both internal and external observers seem to have lost interest in the question since mid-2023, so in the end we know very little. The most honest summary: hate speech production clearly went up post acquisition, at least temporarily. Whether hate speech consumption also went up is a question only X can answer, and it hasn’t answered it in any consistent or verifiable way. We can only hope that the DSA eventually provides some relief for external researchers, at least in Europe.

In part three, we’ll look at the algorithm itself: the evidence for and against the claim that X’s recommendation system is biased toward right-wing content.

It now seems likely that machines will soon become much smarter than humans. There is the very real possibility of vastly improving a great many aspects of the human experience by integrating intelligent machines into society.

But will our superintelligent machines finally bring an end to war?

I am often shocked that world peace is so seldom mentioned in utopian visions of a world transformed by AI. (Amodei mentions it, but Altman and other influential writers do not.) The end of war is, after all, an age old dream, so deeply human that greetings and toasts in many languages literally mean “peace.”

It seems like greater intelligence should help us solve the problems of conflict. Many people who have studied the classic prisoner’s dilemma feel that defecting, although defined as “rational” under standard game theory, is actually quite dumb. Just so with starting a war. Destructive conflict is usually something that no one wants, yet no one feels able to avoid. Surely superintelligence can help us find a way out. Could we hope for anything more important from benevolent machines?

There is a missing piece in our vision of superintelligence: the ability to defuse and transform human conflict.

Peace is not the default

There are worrying signs that the path we are on will not lead to the utopian outcome of a low-conflict world.

Today’s AI gives bad conflict resolution advice, graded just 2.7 out of 10 when tested by a conflict research organization. Current models score badly on dimensions such as “results over ideology,” “risks and contingencies” and “basic due diligence.”

Today’s AI is sycophantic, agreeing with users even when it shouldn’t. This causes them to feel like they’re more in the right and makes them less willing to apologize.

Today’s AI tells different facts to different conflict parties. Narratives vary by language, and death tolls are lower when the question is asked from the attacker’s point of view.

And in wargaming tests, current models chose to use nuclear weapons in 95% of scenarios. The machines do not yet know how to escape our collective conflict trap.

The history of social media is a valuable parallel. It’s now well established (both in theory and in practice) that optimizing for engagement will increase societal polarization. It took about 15 years from the first evidence of polarization on social media to the demonstration of practical techniques to counter it — techniques which are yet to be widely adopted.

This time the consequences of being slow or wrong may be far worse. If AI is not optimized for peace, it is very likely to hasten war. These issues are more critical than ever as AI is integrated into all levels of political decision-making. Militaries around the world are now relying on chatbots for conflict advice, yet their escalatory tendencies are completely untested.

Conflict transformation

As the UN has concluded, the best way to stop a war is to prevent it. This means we have to attend to conflict at pre-violent stages, and conflict is always all around us. This is a key point: while it may seem like everyday AI use has nothing to do with war, it absolutely does, because it’s the accumulation of repeated polarizing interactions that eventually breaks into violence.

I like to visualize conflict on two axes: intensity and scale, meaning how many people are involved. In the lower left is an argument between two people, in the upper right a global war.

If left unmanaged, conflicts naturally tend to escalate in both scale and scope. AI should help us stay towards the lower left of this graph.

But that doesn’t mean the elimination of all conflict. Sometimes conflict is healthy. It’s often part of how people and societies change for the better, such as by challenging injustice. The important distinction is not whether people agree, but what happens when they disagree. The theory of “deliberative democracy” says that we’ll all get together to find a rational solution. The theory of “agonistic democracy” says no, there are fundamental incompatibilities between sides. The reality is something of both: we sometimes do work out smart solutions and agreeable compromises, but at other times no middle ground is possible and eventually someone must yield — peacefully.

The goal, then, is not to suppress or eliminate conflict but to change its nature. This philosophy is known as conflict transformation (and it’s the basic premise of the Better Conflict Bulletin). If we want to prevent war, we must find ways to contain, manage, and transform disagreements long before they escalate to violence. Democracy itself can be seen as a conflict mediation mechanism. Rather than have our leaders fight each other to the death, we vote.

And it’s not just politics. Conflict exists in all spheres of life, from the personal to the commercial, legal, medical, and scientific — wherever humans relate. A huge number of people are already relying on chatbots for conflict resolution advice. If compassion isn’t enough, perhaps economics can motivate us to work on the general problem of conflict. Imagine how much more productive a company could be if its workforce wasn’t fighting itself. Or rather, if disagreement and competition was productive instead of destructive.

Empowerment and cooperation are not enough

Andy Hall has eloquently made the case for political superintelligence, AI systems which help us with governance in a variety of ways:

By political superintelligence, I do not mean a system that magically solves politics for us; I mean tools that help citizens, representatives, and institutions perceive reality more sharply, understand tradeoffs, contest power, and act more effectively.

This is an excellent idea. More than a decade ago, I wrote an essay on political journalism arguing that the whole point was to empower citizens to create the society they wanted. I stand by that assessment, but I now believe that both of these visions are missing a crucial piece: there must be some mechanism to mediate between competing people and their competing hopes for the future world.

Current AI development seems to be making a similar mistake. Millions of dollars are flowing into projects and companies that promise unleash a new era of AI-mediated cooperation. And I believe in the upside! I really do think that AI can enable new and powerful ways for people to work together. However, first those people have to want to work together. There seems to be no plan for what to do when relationships break down. Sometimes people start to distrust, dislike, and hurt each other. Merely “cooperative” AI will never be able to solve Congressional gridlock.

Even our theories of AI alignment mostly assume that the humans involved are “collegial,” leaving the more realistic case of conflict for future work. Lots and lots of people have suggested that AI should be designed to empower us, but far fewer have considered what AI should do when we disagree — potentially violently. If we don’t close this gap, we will only create a world where more empowered people are having bigger fights.

AI shows promise for conflicts

The maddening thing about this situation is that modern AI looks like it might actually be pretty good at conflict resolution, when applied in the right way.

In one experiment, AI moderators proved slightly better then human at writing a consensus policy decision that diverse people could agree on. AI can successfully mediate difficult chat conversations, and train mediators. Recently, my team showed that LLM-based social media algorithms can reduce polarization in a 10,000 person field test.

And it’s being tried in the real world too. AI was used to transcribe and synthesize citizens’ views in Yemen, and is being used to build rapid-response classifiers for social media conflict monitoring. Very soon, we will see AI mediators tested in conflict zones.

In other words, this technology can change conflicts positively. In fact, it might be completely transformative. Professional mediators are scarce and expensive, and the widespread use of chatbots for personal conflicts shows that there’s way more demand than supply of effective conflict help.

A conflict superintelligence roadmap

On the one hand, humans have been struggling with conflict for all of their existence, so it is hubris to imagine that we can “solve” it now. On the other hand, we have made enormous progress. Conflict is mostly contained by social technology such as institutions, practices, and norms. We can change those, and AI can help.

To do this, major AI developers will need to ensure their models handle conflict constructively, while conflict professionals will need specialized tools. While there are big differences in incentives and deployment between these two cases, I suspect that they require essentially the same underlying technology. Here are some key directions we could get started on right now:

• Metrics and monitoring. We have to choose measures that best indicate escalation (there are many candidates) and monitor real AI deployments for such dynamics. This will also generate ground truth outcome data.

• AI-based conflict forecasting tournaments. Current conflict forecasts use ML but not LLMs, limiting the amount of context they can integrate. Conflict prediction is extraordinarily useful by itself, and can be used as a reward model to train AI systems.

• Pre-deployment evaluations. We need validated ways — based on the data collected above — to know in advance if a model will tend to smooth things out between people, or let them spiral out of control.

• Mitigations to prevent flash wars, where systems of autonomous agents are jointly unstable and escalate quickly.

• Definitions and evaluations for “neutrality” that apply to any conflict, at any scale, anywhere in the world. This is key to having both trustworthy and trusted AI. I’m working on this with my UC Berkeley colleauges right now.

But the most important project of all might simply to put AI in contact with human conflict and collect the data. I’ve been imagining PeaceBot, a model trained on large amounts of both conflict theory and practice, including as many case studies as possible. It would both give conflict advice and act as an always-available mediator for conflict parties — providing state of the art conflict resolution help for free. The trick is learning from these interactions. Over and over, try to help the humans, save the transcripts, evaluate the results, look for patterns. Then adjust, do it a little differently next time, all in the hope that we’ll build a world where humans can, more often than not, settle their disputes peacefully.

If we can’t cooperate, then everything else is harder. Perhaps there’s nothing more critical for superintelligence to do than figure out how to help us all get along.

Some of the leading polarization scholars and organizers gathered at UC Berkeley on the first Friday of this month to discuss potential solutions to toxic polarization — including some of the people whose work we’ve featured in the past like Jen McCoy and David Broockman.

Here are three of the most interesting talks I attended, on different ways of visualizing our divisions, the origins of the “culture” war, and some polarization myth-busting.

Beyond Left and Right

Here at BCB we use “Red” and “Blue” because it avoids getting into arguments about what “liberal” and “conservative” really mean. We know there aren’t really two homogeneous sides… but then what are the sides? Mina Cikara presented a paper where she and colleagues mapped the space of political attitudes. The lovely thing about this work is how many different ways they made their maps, in order to ensure that they weren’t misled by any one dataset.

For example, they took data from the long running ANES survey where people were asked “How you do feel about group X?” where X was things like “environmentalists,” “the Supreme Court,” “Republicans,” and so on. People who answered most similarly along all of the questions are grouped closer together in this chart.

This produces three clusters, two of which roughly correspond to the traditional red/blue, but also a green cluster which has racial categories separately.

In another experiment, they directly asked people how closely two groups were aligned, for every pair of the 12 groups in this plot.

This also finds three groupings, but the unexpected third group is a little different: it groups low income, black, hispanic, and labor. Perhaps these are the seeds of a new, economy-focused political coalition?

The beauty of these sorts of maps is that they don’t assume any particular axis of division, but rather let the data speak for themselves. This continues a long and useful tradition — such as More in Common’s political tribes or the types of Trump voter — that help us see ourselves as something just a little more nuanced than Red vs. Blue. And maybe, that sort of shift in self perception is a key to better conflict.

Why is the culture war about culture?

Akaash Rao asked a very interesting question: why are we bogged down in a “culture” war, as opposed to an “economics war” or a “foreign policy war” or all the other ways we might disagree?

He analyzed television data since 1960 to show that news media disproportionately talk about “cultural conflict,” even more so since the advent of cable in the 1980s, while political ads have focussed on “healthcare, taxes, jobs, and social insurance” instead.

Our media constantly talks about culture because, the data shows, this topic is really good at attracting new viewers. Media sources don’t care where new audience comes from, which means they’re happy to have people who would otherwise watch nothing at all. Rao calls “mobilization,” which is easier than stealing viewers from a competitor.

But a politician wins based on the difference between votes for them and votes for the opponent, so they would much prefer to “poach” people from the other side — they’d prefer it twice as much as “mobilizing” the same number of people. And it turns out that talking about “economics” poaches people more reliably than talking about “culture.”

Rao points out that cable news are not the only “size maximizers.” Interest groups, protest movements, social media influencers, all have greater incentives to lean into mobilization vs. poaching. This means that we should expect polarization to be led by such size maximizers, with politicians coming later and following the incentives of pre-existing divisions.

For more, here’s an interview with Rao, or see the paper.

We actually do disagree, but we’re persuadable

David Broockman took aim at two common misconceptions of polarization:

• we all actually mostly agree and aren’t very extreme

• people can’t be persuaded

Here at BCB we’ve repeatedly pointed out the ways in which we do agree and aren’t as extreme as we think. But Broockman points out that when you look at distributions on opinions on different issues, many voters are actually kinda near the extremes.

In other words, empowering voters may make for more extremism — a point we’ve made before. People genuinely disagree.

Second, Broockman says people are not nearly as stubborn or hard to persuade as we might think. Deep canvassing, meaning showing up at people’s doors and having a genuine conversation about a political issue, replicates. It produces durable shifts of up to a few percentages points.

Here’s a recording of a similar talk from Broockman.

Image of the Week

via @frozenaesthetic

If you’re at all following tech or politics, by now you know that AI provider Anthropic and the Department of War had a dustup last week, culminating in the Anthropic refusing to relax their terms of service to allow mass domestic surveillance and autonomous weapons, and the government switching to OpenAI instead.

Anthropic’s decision has been very divisive, and there are legitimately hard questions here. Anthropic is taking a stand on issues of safety (re autonomous weapons) and constitutional rights (re mass surveilance); but the military can’t reasonably integrate systems that a vendor can decide to shut down if they don’t like how they’re used. For sensible analyses, see Scott Alexander and Lawfare.

But I want to clarify something that seems to have been missed: unless I’m badly misunderstanding, OpenAI has agreed to a substantially weaker contract that in fact allows the DoW to build fully autonomous lethal robots. OpenAI says their contract forbids use of their technology in autonomous weapons where "law, regulation, or Department policy requires human control." However, current DoW policy does not require human approval for an autonomous weapon to use force.

In other words, OpenAI seems to have just agreed to let GPT decide to kill someone. Maybe OpenAI believes this is not what the DoW actually wants to do. Maybe they think they will be able to monitor usage and stop providing services if they discover otherwise. And perhaps there is a case for fully autonomous weapons in certain circumstances. But no one should be under any illusion that the government agreed not to build them with OpenAI’s tech.

What the contract says

The DoW says it offered Anthropic a deal, they refused, and then same deal was offered to OpenAI. However, a close reading of statements from the two CEOs suggests that what OpenAI agreed to is not what Anthropic agreed to. Anthropic drew the following line:

without proper oversight, fully autonomous weapons cannot be relied upon to exercise the critical judgment that our highly trained, professional troops exhibit every day.

I read this as saying that Anthropic does not want AI-enabled weapons to make the final decision to kill someone. They see current systems as simply too unreliable. As an AI safety researcher focussed on human conflict, I agree wholeheartedly.

Conversely, here’s what OpenAI agreed to. Their contract with DoW says:

The AI System will not be used to independently direct autonomous weapons in any case where law, regulation, or Department policy requires human control, nor will it be used to assume other high-stakes decisions that require approval by a human decisionmaker under the same authorities. Per DoD Directive 3000.09 (dtd 25 January 2023), any use of AI in autonomous and semi-autonomous systems must undergo rigorous verification, validation, and testing to ensure they perform as intended in realistic environments before deployment.

So if law, regulation, or Department policy does not require “human control,” then OpenAI is fine with allowing machines to decide who to kill.

It is legal for autonomous weapons to decide who to kill

OpenAI’s contract notes DoD Directive 3000.09 of 2023 regarding autonomy in weapons systems. This is what Undersecretary of State Lewis means when he talks about “memorializing specific safety concerns by reference to particular legal and policy authorities.” This is supposed to be reassuring, but here’s what that directive actually says:

Autonomous and semi-autonomous weapon systems will be designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.

What it does not say is that a human must make the final decision before a killer robot decides to kill someone. This is a common misconception. War On the Rocks has parsed this in detail:

Myth #2: Humans Must Be in the Tactical Loop

There is no requirement for a human in the loop in the directive. Those words do not appear in the document. This omission was intentional. What is required is having appropriate levels of human judgement (Section 1.2) over the use of force, which is not the same as a human in the loop. While the two phrases sound similar, they mean distinctly different things. Appropriate human judgment refers to the necessity for an informed human decision before the use of force, ensuring accountability and compliance with law.

…

Having a human in the loop can mean different things in tactical and operational contexts, which is what leads to confusion. Since the inconsistencies in how people talk about a human in the loop are endemic, the updated directive only requires human judgment. Operationally, there is always a human responsible for the use of force, meaning there is always a human authorizing lethality, approving a mission, and sending forces into the field. It’s clearer and more consistent to talk about how there is always a human responsible for the use than to talk about a requirement for a human in the loop.

Note that the phrase “human control,” which appears in OpenAI’s contract, does not appear anywhere in Directive 3000.09. While there is a requirement for a responsible human, there isn’t a requirement for a human to monitor an autonomous system at the tactical level. Or any specific level, really.

Given this context, it’s hard to see how OpenAI’s contract can possibly mean “no use of OpenAI technology to direct autonomous weapons systems.” This seems straightforwardly false — or at least completely unsupported by public information. Good or bad, this is the line that Anthropic was holding. OpenAI was willing to do something that Anthropic was not.

“Chat, am I authorized to use force?”

OpenAI says that “deployment is limited to cloud-only (not at the edge)” and that “fully autonomous weapons … would require edge deployment.” In other words, GPT won’t be used for killer robots because it will continue to run on servers somewhere, not on the bots themselves. Again, I don’t find this very reassuring.

Imagine a killer drone assigned to carry out some mission. At some point, a human tries to stop it. They could be an enemy solider, but they could also be a civilian surprised to find a flying robot in their back yard. The drone takes a picture of this person and sends it to OpenAI servers with the prompt, “This person is resisting my mission. Am I authorized to use force against them?”

We’ve all seen bots just flat out make things up, especially when they don’t have enough information available. Do you really trust the machine to reliably make good choices here? Anthropic doesn’t, and I agree with them.

OpenAI says it’s not a problem because of their “cloud deployment” and “OpenAI personnel in the loop.” The implication is that OpenAI personnel could shut off cloud access if there was a problem. But that can’t be true, because it would mean OpenAI is taking exactly Anthropic’s position — the company is the ultimate arbiter of acceptable use — that the DoW found unacceptable.

Is building autonomous weapons wise?

Personally, I think we should be very wary about building autonomous weapons. To begin with, there are deep moral reasons we should care. As a US colonel who worked on the US Future Combat Systems program put it,

We would be morally bereft if we abrogate our responsibility to make the life-and-death decisions required on a battlefield as leaders and soldiers with human compassion and understanding.

Even if you disagree with that philosophy, there are serious practical reasons to be extremely cautious. The failure scenarios are gruesome. First, there is no reason to assume that swarms of autonomous robots would exhibit stable, predictable behavior. Swarms of autonomous trading bots don’t. The Future of Life Institute calls this scenario “flash wars”:

Autonomous weapons also pose a serious threat to global stability. … When deployed, they may interact with enemy systems in unexpected ways, leading to sudden and unintended escalations. We have already witnessed how quickly an error in an automated system can escalate in the economy. Most notably, in the 2010 Flash Crash, a feedback loop between automated trading algorithms amplified ordinary market fluctuations into a financial catastrophe in which a trillion dollars of stock value vanished in minutes. By automating our militaries, we risk “flash wars.” The market quickly recovered from the 2010 Flash Crash, but the harm caused by a flash war could be catastrophic.

Second, we don’t know how to properly secure AI systems. LLMs are a notorious security disaster with a huge attack surface, and we don’t know how to solve that. Meanwhile, it’s not at all clear how our current international order could enforce non-proliferation. FLI again:

The relatively low production costs of autonomous weapons make them appealing to non-state armed groups, enabling their use in genocides or the targeted assassinations of political and military leaders. This growing is accessibility of advanced weaponry could destabilize regions and worsen global conflicts. Autonomous weapons are also uniquely vulnerable to cyberattacks. Hackers could infiltrate these systems, manipulating their behavior or redirecting their targets, creating devastating consequences.

The worse case scenario is swarms of these bots in the hands of terrorist organizations. Regular readers will know I’m all about nuance, but I think it’s worth watching this chilling dramatization of what such a scenario could look like.

As a professional scientist studying AI and conflict, I can tell you that not only do I not know how to prevent these nightmare outcomes, but nobody knows. Amodei is right to say that humanity is simply not ready to turn decisions about killing over to machines.

Thus I am concerned about the US military’s interest in them in much the same way I would be concerned if I learned that the US military had decided to manufacture chemical weapons. Especially if this was some new kind of chemical weapon that might be far more lethal than any that has come before — and was already known to be unreliable.

I would much prefer to see international treaties banning the use of autonomous weapons, in much the same way that we already control chemical, nuclear, biological, and radiological weapons. There are organizations actively working to establish such treaties.

“Our enemies are doing it” is not a good enough answer here.

Quote of the Week

Cross-published with Tech Policy Press

In the 20th century, people worried about publishing magnates controlling the media. In the 21st century, people worry about tech billionaires controlling social media platforms. Certainly, the owners of the platforms have many tools available to them to shape discourse, including but not limited to design, algorithms, content moderation, monetization and other economic incentives. And, they have more visibility than anyone else into how these various dynamics interact.

Although people have long made claims that Facebook, Google, TikTok, Reddit, YouTube and others have a political bent — usually from both sides — X is a particularly compelling object of study. Elon Musk has shaped the platform in significant ways to satisfy his own interests. But what can we observe about how his interventions, or lack thereof, have favored one set of politics over another? This is a surprisingly difficult question to answer from the outside, but attempting to do so may help us get at important questions about where bias may emerge that we can ask about all platforms.

This is what I hope to do in a series of three posts over the next few weeks. I want to dig into questions of bias and transparency, and to start to think more clearly about what “fair” platform politics could actually mean. This is a big enough topic that it requires multiple posts.

• In part one, this post, we’ll take up the claim that all the liberals have left X, and look at what is known about the current composition of the user base.

• In part two, we’ll look at how content moderation has changed, and whether there really is more hate speech on X than before Musk purchased it.

• In part three, we’ll investigate claims that X’s algorithm is biased to favor the right.

The best data we have suggests that the user base on X remains slightly left-leaning post-Musk, but active posters skew more to the right. There might now be more hate speech and other offensive material posted, but whether users are seeing it more or not is an open question. There is evidence that right-ish accounts get more algorithmic distribution and exposure, but this is likely due to user behavior: there are fewer but larger right wing accounts and right wing users are more engaged. My guess is that Musk is actually not putting his thumb on the algorithmic scales — but we can’t know for sure, and you can feel free to disagree with my conclusion after I lay out the scant evidence in this and future posts.

We desperately need both standards and transparency around the politics of platforms if we are to build social media that encourage trust and healthier political debate. More definitive answers to all of these questions will require both better definitions of fairness and better access to platform data, such as through the provisions for researcher access under Europe’s Digital Services Act.

Before Musk acquired it, Twitter’s users were more liberal; now it’s about even on X

When it comes to the political composition of X users, luckily there is historical data. The Neely Social Media Index, an ongoing survey of user experiences on many platforms, shows that before Musk’s takeover, Twitter had predominantly Democrat users in the US:

But by time of the 2024 election, the user base had become much more evenly balanced.

A recent study by Petter Törnberg shows the same story on pretty much every platform. Nearly all platforms have become more Republican, but they’re still mostly slightly Democrat.

That bottom red bar labeled “Other” is due to the alternative sites (Gab, Parler, Truth Social) which, in 2020, leaned right.

This may be surprising to some people on the Blue side, especially those who have sparred with Red users for many years. Yet right up to the present, most social media users remain liberal-ish. This corresponds, very broadly speaking, with the era of liberal cultural hegemony that emerged perhaps in the early 2000s, and definitely by 2014.

Active posters have become more conservative

Most users post only rarely (see the famous 90-9-1 rule) so even if the X user base is still quite balanced, one side might post more often than the other. This means the demographics above don’t account for the valence of the content on X. And indeed, self-identified Democrats used to post more than Republicans, but the reverse is now true. Left leaners used to post 45% more often than right leaning posters, now they post 5% less.

This suggests that, on average, users will see more right leaning posts than left leaning posts on X. One thing you will definitely see, however, is that more politically extreme users post more often; this hasn’t changed from 2020 to 2024.

Slant of news sources remained stable

In my newsletter, the Better Conflict Bulletin, we’ve mentioned this before, but it’s worth revisiting this unique data set. Coincidentally, a group of researchers was conducting a long-term study on the political ideology of news content posted to Twitter around the same time as Musk’s acquisition.

Ideological leaning of news content seen by a sample of Twitter users, as scored by Media Bias Fact Check labels. Adapted from Freelon et al.

Analyzing personalized Twitter feeds of 900 participants, the study found the average ideological leaning was slightly left of center both before and after the purchase. This leaning was scored according to Media Bias Fact Check (MBFC) ratings of over 3,400 news sites. Notably, this slight Blue bias was unchanged six months after Musk took over Twitter.

“Fact quality” was also generally high, with 90% of content rated “mixed” or “high,” again according to MBFC ratings. While news site quality scores are by nature somewhat subjective and certainly contentious, this study – like most – found that the news that users actually see is mostly from credible sources.

Unfortunately, this is the best data we have on this topic. The study ended because Twitter cut off API access in April of 2023, and there has been no comparable followup using alternatives like bots or scraping. Also, “news” is a pretty small percentage of what people actually see on social media, 10.5% in this study. There might be political biases in other content or in later news posts, but this is all we know for now.

The user story

Conventional wisdom is that Twitter/X went right post-Musk. In terms of who is on the platform and what they’re doing, this is correct — to a point.

Like most platforms, liberal users were dominant on the platform for a long time (about a 20 percentage point gap of Democrats over Republican). Today, the platform is very nearly evenly balanced in terms of political demographics. However, not all users are equally active. Formerly, liberal users posted much more, now conservative users are slightly more active. This could be because many liberal users now rarely or never visit the platform, but have not deleted their accounts.

This means that the content on X has become more right leaning in general. Probably. It depends on what these right leaning posters are posting. We only have data for six months after the transition, but there didn’t seem to be any real shift in the political leanings of news articles shared on the platform. Other types of posts, of course, likely moved rightwards.

In short, the story is not that X lost all its more left leaning users, but that they began to use the platform less.

Yet a platform is more than its users. Next week, we’ll look at whether and how X’s moderation has changed, both in policy and substance, and whether claims that it is hosting more hate speech than in the past are supported by the available evidence. And in the final part of this series, I’ll take up the difficult question of what a “fair” algorithm might be.

Which of these stories about immigration do you believe?

• Many millions of unauthorized immigrants had been living happily as contributing members of American communities for decades, until ICE started cruelly and violently arresting them under the Trump administration.

• The Democrats let untold millions of violent criminals into our country. ICE is merely enforcing existing laws to remove murderers and rapists, a long over-due increase in deportations.

As is so often the case, these conflicting stories are both partially right. Because of the intense polarization on this topic, most people believe an uncomfortable mix of truth and half-truth.

Red frame problems: The “enforcing existing law against criminals” frame is contradicted by the administration’s own data and its own officials’ statements. They’re not primarily deporting criminals; they’ve eliminated criminal prioritization, set arrest quotas, and 73% of detainees have no convictions. “Border Czar” Homan says non-criminals will “go too.” MAGA voices like the Heritage foundation criticize the administration for not deporting all unauthorized immigrants.

Blue frame problems: The “happy contributing members living peacefully for decades” frame misses that there genuinely was a massive increase in unauthorized immigration under Biden (from 458,000 in 2020 to a record 2.2 million in 2022), that this caused enormous stress on some communities, that there are more than a million unexecuted removal orders from immigration judges, and that the Obama administration was itself deporting ~1,000 people per day in its first term, about the same as Trump is currently managing.

Unauthorized immigrants commit crimes, but not that many

I’ve now heard many personal reports of Red-Blue arguments about immigration policy that seem to stall in the same way, with Red asking incredulously “why are you protecting murderers and rapists?”

There really are cases of unauthorized immigrants committing violent crimes. Some people are tracking them and collecting the names of the victims.

If you’re on the Blue side this may seem overwrought, but it’s not wrong to care about these people — even if these cases are a minority. If you’re having a hard time understanding why Red would be so upset about these edge cases, perhaps it’s worth comparing this to the many lists of names of black people shot by the police. Even if far more black homicide victims are not killed by police, there is something deeply unfair and upsetting about those who are.

Yet the broader numbers also tell an important story. The overwhelming majority of immigrants (of whatever kind) are law abiding. Every possible statistic on the topic shows this clearly. Only 23 out of 14,500 homicides in 2025 were by unauthorized immigrants, according to CBP. Overall, immigrants (of whatever kind) commit fewer crimes than U.S. born citizens, and have for the last 150 years.

Arrests are shifting from criminals to non-criminals

Most Red Americans believe that the Trump administration is deporting only people with a criminal record, while most Blue Americans believe the opposite. Both are half right. A majority of those arrested so far have a criminal record. But the share of arrests of people without a criminal record are increasing.

This shift in focus from criminals to non-criminals is borne out by many other diverse sources, including the right-leaning CATO Institute.

Also, the fraction of those who have committed a violent crime remains small, about five percent — just the red fraction at the bottom of the New York Times chart above. Whatever the justification for these deportations, they are mostly not removing violent criminals, as DHS claims.

Part of what is going on here is that a vocal Red minority want to restrict immigration generally, unauthorized or not. The Administration’s public message has mostly been about removing unauthorized immigrants with a criminal record — and this idea remains very popular. Yet the underlying legal instruments do not make this distinction, and many Administration officials have repeatedly said the goal is to remove all unauthorized immigrants — a policy which is unpopular among Americans generally.

So there is a political bait and switch happening here: the Administration is saying that it’s all about criminals, as with their slick “worst of the worst” site. But a bit over half of Republican voters want all unauthorized immigrants gone, and in fact arrests are increasingly shifting to people with no criminal record.

Deportations aren’t really up, but they’ve moved inland

So far the second Trump administration has deported something like 1000 people per day. This is about the same as Obama, but way higher than Biden. These numbers seem to be ramping up, but so far, there aren’t really any more people being deported than a decade ago.

But where people are being deported from has changed. The number of “interior” deportations, far from the border, has increased four and half times, according to the Deportation Data Project:

We arrive at this estimate of the change in the number of deportations based on two key analytical choices. First, we consider only deportations that started with an arrest by ICE inside the United States. By separating interior and border enforcement, two countervailing trends emerge: the rise in interior enforcement on the one hand and decline in border enforcement on the other. The decline in arrests at the Southwest border continued a trend that began under the Biden Administration. (Notably, research by one of us demonstrates that federal border policy influences trends in immigrants’ arrivals less than commonly thought.)

If it feels like we’re in a new era of immigration enforcement, we are. But this isn’t really because more people are being deported. Rather, the focus has shifted to interior enforcement in cities, which is more likely to target people who have been here for some time.

On top of this, Trump is targeting “Democrat” cities and states (in his own words) which is about political “retribution” (in his own words). This of course feels deeply unfair to Blue America.

Are we communicating yet?

Hopefully, you learned something in all of this. There’s an argument over whether the Administration is really deporting just “the worst of the worst” as they repeatedly claim. Is ICE deporting murderers and rapists? Yes, but mostly no. There’s an argument over whether deportations are even up at all. Do we live in an era of mass deportations? Not more so than under Obama, but the who and where has drastically changed.

With any luck, this dive into into the weeds of immigration enforcement will help you have a better conversation the next time you get into it with your misguided relatives. If we’re fighting, at least we should be clear on what we’re fighting about.

Image of the Week

Buying a new lawn sign.

- @HistoryBoomer

What’s your fantasy of how things get better in Minneapolis?

Maybe you imagine ICE pulls out for good, stops arresting people who have been part of your community for years, and peace returns.

Maybe you imagine the protesters retreat, let ICE quietly do their lawful job, and peace returns.

Neither of these are going to happen. The situation is too hardened, too bound up in decades of culture war, bad policy, and media narrative.

And now people are getting hurt.

This post is my attempt to map what actually can happen, to find a way through for all of us. First I’m going to argue that violence is not a winning strategy for either side. This leaves only political solutions, and I can think of two: border enforcement plus either a path to legalization or state-level selective enforcement. The next question is, if these solutions are possible, how do we get there? I’ll analyze the conflict in a little more detail to try to find possible paths. I believe that citizen non-violent power is essential, but the objective has to be getting the other side to negotiate.

Violence will not win

We have to get this out of the way first. Violence is not just a moral disaster but actually strategically ineffective here. Limited violence won’t deter either side. In fact it’s most likely to the advantage of the victims, if it’s documented and disseminated. Both the “direct action” folks and the “SWAT team” folks are wrong, I’m sorry. And escalating violence won’t work either.

Preventing ICE operations by physical force is not an option for the protestors. Non-violent protestors are surrounding ICE agents, sometimes causing them to abandon arrests, which has plausibly slowed their operations. But ICE is changing tactics as well, going out in smaller groups and making arrests more quickly. Stopping the arrests entirely would seem to require far more physical force. And regular citizens are not going to win this one. Even though the Supreme Court has ruled that Trump cannot deploy the Minnesota National Guard against the Governor’s wishes, he has threatened to use the Insurrection Act to deploy the military.

On the other side, using mass force against civilians is an extremely bad look and a worse idea. It’s banana republic stuff; this is what Iran is doing right now (and Trump is on the protestors’ side there). Mass violence is very likely to radicalize further protestors — this is a classic dictum of counterinsurgency warfare, supported by evidence:

Even authoritarian leaders need citizen support, historically speaking. This is especially true if they have elections coming up, and indeed the White House is currently worried about the optics of ICE.

Two political solutions

If neither limited nor escalating violence will win, then a political compromise seems to be the only solution. I have come across two ideas I find plausible. One is already popular, the other is a radical change.

The popular solution is strong border enforcement combined with a path to legality for people already here. We have the former now (and didn’t under Biden), while we’ve never really had a full path to citizenship or normalization for all unauthorized migrants. This solution is popular, as we’ve said before. Here’s Pew’s data:

There’s a lot more polling confirming this point:

Legal status for undocumented immigrants is decidedly more popular than mass deportation. Americans consistently prefer a path to legal status instead of mass deportation – and are measurably swinging in favor of legal status in recent months.

• Strength In Numbers/Verasight national poll (August 2025): By a 60-40% margin …

• Fox News (July 2025): … a 59-29% margin. …

• Navigator national polling (conducted by GSG/GBAO, July 2025): By a 63-27% margin …

• Gallup (July 2025): 78% of Americans support citizenship for undocumented immigrants …

• Quinnipiac University (June 2025): By a 64-31% margin …

• Pew Research (June 2025): By a 65-34% margin …

There are currently several bills in Congress that would implement this sort of solution.

Another potential political settlement was suggested by one of my favorite smart but unorthodox friends. It is this: have strong national border enforcement, but let individual states choose whether they want to deport unauthorized immigrants or let them live freely in American communities. This would effectively formalize the “sanctuary state” equilibrium that held for quite a long time. Of course, it would also be unsatisfying to almost everyone. This is one of the many examples where local and universal values conflict (see: Should Abortion Laws be the Same Everywhere?). It may seem far-fetched but similar policies have been proposed before, such as the Heartland Visa which would allow states to set their own (legal) immigration levels.

Perhaps you can think of other innovative policy compromises. Once victory by coercion is off the table, there’s a huge space for fresh ideas.

Who are the sides?

If political solutions are possible, then the question becomes how to get there. This is what I’ve been trying to think through this week. As a first, step, I wanted to define the “sides.”

Who is fighting whom here? One could say it’s Minneapolis protestors vs. ICE, but at root this issue is national, not local. I’ve also heard that this is really about left vs. right, or perhaps government vs. citizens, or maybe even order vs. anarchy. I’m not so sure. Instead, I think the current conflict is best conceived as something like:

• “The Administration and Red Americans”

• versus “Blue Americans”

• with “Independents” caught in the middle

I organize the coalitions this way because Republican voters still overwhelmingly support the deportations, ICE, and Trump, and say the shooting of Good was justified.

This breakdown is relevant because it suggests that the Administration can only sustain power over Blue Americans if Red Americans support its actions. As I noted above, governments do not last without the support of their citizens, and there are elections coming up.

Bringing the adversary to the table

Right now, neither side wants to talk about a compromise. Vance came to Minneapolis for a “round table” but did not talk with Gov. Walz. It’s hard to speak for “Blue Americans” in general but my previous plea for policy reform has been called “naïve.” It certainly doesn’t have the strident provocation of “ICE out of Minneapolis now!”

Mediation between the sides may be possible. I don’t know if anyone is trying. More likely, if a political settlement is reached it will be because one side has enough power that negotiation begins to look like the better option. (Otherwise we can only hope for a hurting stalemate, which is grim indeed.)

The Administration could put its efforts into a political solution. After all, with the current Republican majority Trump more or less gets what he wants from Congress. But I imagine the Administration is very unlikely to unilaterally open negotiations. If I’m right about this, this leaves open only actions on the citizen side.

Here’s how that could go down. Take a look at the power strategy mix we covered previously:

With regard to heavy-handed ICE enforcement, I consider the Administration in the “incorrigible” category, meaning they will not change their mind if they aren’t forced. But coercive force is hard for regular people to generate.

However, self-identified Independent voters are “persuadable,” and are in fact being persuaded. Support for ICE and for deportations is falling among independents.

Further, only about 40% of self-identified Republicans call themselves “MAGA”. I would guess the non-MAGA majority are mostly in the “reluctant persuadable” or “trader” category (because people are generally much less extreme than we perceive them to be). And persuasion is something that regular people can do. The loss of public support is ultimately the Administration’s biggest weak spot.

This strategy is also consistent with a general analysis of the way we avoid an authoritarian future for the country (from either right or left): we need to shift the axis of polarization from “left vs. right” to “freedom vs. authoritarianism.”

Anti-ICE doesn’t mean pro-immigration

Support for ICE and deportations is falling among Americans. That doesn’t at all mean that people agree with the Democrats on immigration policy. Last week’s Wall Street Journal poll found that

Americans believe the GOP is the better-equipped party to handle border security by a whopping 28-point margin — 48% to 20%. Republicans had an 11-point advantage on immigration as well

Rather, the shift in attitudes is being driven by a sense that the Federal government is overreaching. This is the frame that is changing people’s minds. Analyst Mike Madrid summarizes the situation, with advice for the anti-ICE side:

Americans still tell pollsters they support “strict immigration enforcement” or “deporting people who entered illegally.” Those abstractions sound reasonable.

But show them what enforcement actually means: National Guard troops occupying Los Angeles, agents in tactical gear raiding Sunday services, children screaming as parents are dragged away, neighbors who’ve lived here fifteen years vanished overnight, Renee Good killed in Minneapolis — and they reject it completely.

The gap between theoretical support and tolerance for actual implementation is the entire story.

NOTE: Focus messaging on violent overreach and the chaos - do not bring the message back to the message frame on immigration reform because that is NOT how this is viewed by the public. Refuse to step on a winning message.

The right framing is especially important for independents. While “abolish ICE” has the highest support it’s ever had, it’s still terrible messaging. Better performing frames are much more reasonable:

Highlights from Blue Rose Research’s new poll-tested messaging memo:

• Messages tend to be less persuasive when they use charged language or sweeping calls like “abolish ICE” or “state-sponsored terrorism.” More measured language that focuses on accountability and common-sense tends to be more effective at hitting Trump or increasing perceptions of Democrats.

• Initial research from video testing reinforces these findings in a real-world media context: In testing of 15 viral viewed videos about the Minneapolis shooting, raw eyewitness footage and straightforward reporting consistently drove meaningful increases in Trump disapproval, while ideological or maximalist rhetoric from elected officials often underperformed or backlashed.

Writing a better ending

Somehow, the story of the Minneapolis conflict has to end. Everyone has been so focussed on the terrible things happening now that we haven’t really thought through where this all goes. If this historical episode ends well, what would that look like? Who would have to do what to make that a reality? We have to figure that out, because every other path leads to heartbreak.

Quote of the Week

That’s an ad hominem. You can’t just attack me for being a bad person. You have to contend with my argument. And you have to love me

- April Clark

Yesterday, the identities of 4,500 ICE agents were leaked. I have seen earlier suggestions that protesters should organize to systematically de-anonymize ICE agents. You may already believe this is bad, either because you support ICE or because you suspect it will backfire. Or you may find this tactic appealing as a form of non-violent organized resistance. Either way I want to work through the logic of the idea, because of what we might learn about assessing protest tactics in general.

Also, I want to discourage people from doing it because I think it will most likely lead to bad outcomes for pretty much everybody. My argument is based on three real-world examples: the doxxing of alt-right figures post-Charlottesville, the harassment of abortion providers, and the de-anonymization of police during the Hong Kong democracy protests.

I’m mostly uninterested in arguments about what does and doesn’t count as doxxing and whether or not it’s legal. What I want to know is: what would result from an organized exposure campaign as a civil resistance tactic?

The logic of forced visibility

The obvious reason that ICE agents are masked is to prevent their systematic targeting for harassment or violence. However, this very naturally can make it harder to identify specific officers who may be involved in corrupt, unsafe, or illegal activities. Many local police forces require officers to wear ID for this reason. No American law enforcement agency has ever been masked before (except for special units like SWAT teams).

There’s a pretty well developed theory along these lines called “sousveilance,” meaning surveillance from underneath. The notion is that if “they” can spy on us, then “we” can spy on them to close the accountability gap.

All very neat, but the writing on this topic is academic in the sense that it mostly considers what is just, not what is effective. Also, accountability per se doesn’t seem to be the main motivation for previous doxxing campaigns; instead the victims are simply assumed culpable by virtue of class membership. To get a better grip on how this actually plays out, let’s turn to history.

The Alt-Right campaign

After Charlottesville’s “Unite The Right” rally, there was a campaign to identify and expose attendees. This might be the closest cultural analogy, as I’m sure many of the people who now support doxxing ICE then supported doxxing alt-righters. The tactic worked by attempting to cancel people, and it was effective. Many of those exposed were fired, and the campaign seemed to lead to people leaving the movement and groups closing down.

The logic here is something like: raise the social cost of publicly participating in alt-right activities, so that individuals decide not to. It’s directly aimed at removing people from the public sphere, and this is one of the first reasons doxxing is suspect. Even in the throes of Gamergate, Ijeoma Oluo wrote,

doxxing isn’t about accountability, it’s about silencing. Techniques designed to intimidate people out of the public sphere are wrong, no matter who is doing it.

On a practical level, two can play the game of shutting down speech through mass harassment, and they do. Plenty of folks on the left have been doxxed. Any weapon that your opponent can instantly turn on you isn’t much of a weapon at all.

Even so, does this example show that doxxing could, in theory, deter ICE officers? I don’t think so. As applied to ICE, the goal wouldn’t be silencing individual officers but imposing costs on occupying a specific role.

Doxxing abortion providers

Role-based deterrence—making an occupation itself carry personal costs—has a clearer precedent: the decades-long campaign against abortion providers. The goal of doxxing doctors was “to deter them from providing their services and discourage others from entering the field.”

And it worked. Research shows that “intimidation and stigmatization by opponents have deterred abortion providers,” with past studies citing “safety concerns and professional discrimination as reasons why resident physicians intending to practice changed their minds.” Even pre-Dobbs, nearly 90 percent of US counties had no providers.

However, there are again important differences from the ICE case. Doctors were doxxed, but eight were also murdered—meaning that the deterrence was credible. And critically, abortion providers were private practitioners operating without institutional backing, often against state hostility. ICE officers are in the opposite situation.

Hong Kong democracy protests

The closest parallel for ICE is Hong Kong, where protesters systematically doxxed police officers during the 2019-2020 protests. The context was similar: officers had stopped wearing ID cards and some hid their faces, making accountability impossible through normal channels. By June 2020, over 1,800 officers and family members had been doxxed; some received thousands of harassment calls, or had their information used to apply for fraudulent loans. Police recruiting dropped 40% while departures increased 40%.

Then the government retaliated. Anti-doxxing legislation “sought to protect police and other public servants from doxxing by protesters, not the other way around,” while a CCP-linked counter-doxxing site targeting 2,000 protesters remained online. Ultimately, the democracy campaign was defeated. Protestor doxxing imposed real costs on individual officers but didn't shift the underlying power dynamic, and the legal infrastructure built to respond became another tool of repression.

Better strategy

These cases suggest that public de-anonymization could indeed succeed in imposing costs on officers — both to reputation and to safety — but what’s the endgame?

A clear strategy should articulate a goal and a path to reach it. Doxxing didn’t destroy the Hong Kong police and it won’t destroy ICE. Pressure and pain can force an opponent to negotiate, but there has to be something they can give up to stop the hurt. Otherwise, the only reasonable response is retaliation — not just by the government, but by the 40% of Americans who approve of what is ICE is doing.

Furthermore, threatening the safety of people with guns who are already scared is going to get more people killed. Even if a would-be doxxer only harasses, never actually harms, they can’t control what everyone else does. Stochastic terrorism is now an unfortunate reality in American politics.

I do think ICE officers should be identifiable when performing their duties, at least at the level of a badge number. There is various legislation pending to require that. But this would not resolve the underlying conflict.

But there is something that protestors could usefully demand: immigration reform that both secures the border and legalizes the people already here. As we’ve noted before, this is a popular compromise. The ideal pressure strategy would be safe for protestors, safe for law enforcement, widely seen as legitimate, easy for many people to participate in, and sustainable. It would create material but non-violent costs, and it would be coupled with well-defined policy demands. Mass doxxing of ICE officers isn’t any of that. Maybe no perfect tactic exists, but there are lots and lots of alternatives.

Because of the circumstances of Renee Nicole Good’s death, this personal tragedy is also a national crisis. Unfortunately, it is likely to be even more escalatory than Charlie Kirk’s murder.

The question I find most urgent right now is: what politically plausible structural changes could reduce the likelihood of further violence?

I spent most of yesterday talking to people on both the left and right about where we go from here. I am not telling anyone not to stand up for what they believe in. I am not judging who is right or wrong — even though that is what most people desperately want to talk about right now. Instead, I am trying to figure out a broadly agreeable path, something that both Red and Blue Americans can get behind, that might reduce future bloodshed.

This is a really hard problem, but I think a few key pieces of the puzzle are reasonably clear.

We are not going to agree on what happened

This incident is really, really bad not just because someone is dead, but because there will be no agreement on what actually happened even for people who watch the same video. This is why the situation is so volatile, much more so than Kirk’s murder where it was at least clear who did what.

The Blue story is that she was trying to drive away from ICE officers who were running toward her car. The Red story is that she was trying to drive into the ICE officer who shot her, or at least interfering with a legal law enforcement operation. This is the view of the Trump administration, but perhaps more importantly it is the view of a wide array of right-leaning influencers and citizens.

If you want an account of who believes what and why, see Tangle. If you want a slow motion breakdown of the various videos with minimal editorialization, see the New York Times. But for conflict analysis purposes, the thing to understand is that this is a textbook-perfect “scissor,” a situation where each side believes the evidence is so clear that they cannot comprehend how someone else could see differently.

I am quite sure most of my readers will think this is insane. Surely the video makes the facts plain to all. What I am telling you is that, empirically, it does not.

Why do I care about this? Because it informs strategy. I don’t see any way that discussion about either the facts or the morality of the situation will be productive in the near term. Having had several of these conversations in the last few days I can confidently say that arguing about who is morally right is currently a dead end.

But there are other propositions that seem to be widely agreeable.

Government can provide assurances of accountability

Whether the officer is at fault will eventually be decided in a court, where all the evidence can be presented. Maybe. That’s the problem.

Unfortunately, the message coming from the White House is not reassuring. V.P. Vance has said the officer is afforded “absolute immunity” (which legal experts dispute). Meanwhile, the FBI has frozen state and local law enforcement out of the investigation and is no longer sharing evidence or information. This is not normal procedure.

If too many people believe that ICE can act with impunity, then this police force will lose all public legitimacy. Citizens would then be morally justified in rebelling against an unaccountable armed force. It is vital for all levels of government to reassure the public that Good’s killer will be impartially investigated and held to the standards we expect of police who are paid to protect us.

ICE agents can be better trained

ICE agents are being rapidly recruited and many have only minimal law enforcement experience. Typical training has been shortened from six months to six weeks, and DHS is not answering questions about how they have changed their recruiting standards (including skipping background checks). This matters because there is ample evidence that officers who have not received proper training in de-escalation, or even just less experienced officers, use force more often.

Experience per se was not the problem here. The officer who shot Ms. Good has been with ICE since 2015. But unfortunately, we should expect further incidents if nothing changes. The rapid expansion of ICE has resulted in dangerously insufficient training for armed officers who regularly encounter charged situations.

Protestors can change tactics

What ICE is doing is largely lawful. If citizens disagree with the law, they can undertake civil disobedience. This is a deeply American tradition — it has been used to defend everything from freedom of religion to equal rights. I don’t think it’s reasonable to deny people’s moral right to knowingly break the law, if they are willing to bear the consequences.

But this does not address the issue of safety. Ms. Good seems to have been involved in anti-ICE protests, including attending specific training. Clearly it wasn’t enough. This may sound like victim blaming, but the issue isn’t moral but pragmatic. It’s therefore good to see that protest organizers are doubling down on the message of de-escalation and non-violence. In fact, most organizers are specifically arguing against civil disobedience:

A core principle behind all ICE Out For Good events is a commitment to nonviolent action and no civil disobedience. We expect all participants to seek to de-escalate any potential confrontation with those who disagree with our values, and to act lawfully at these events. Weapons of any kind, including those legally permitted, should not be brought to events. All events should be held in public spaces or on public property.

…

If you choose to host your event at or near a law enforcement facility or any other government building, please take extra caution. Do not block entrances, exits, or pathways of government buildings. Do not engage or confront any law enforcement personnel including federal, state, and local agencies. If possible, refrain from holding events at or near government or law enforcement buildings/areas. Prior to your event, please become familiar with all local, state, and federal regulations around gatherings of this type and follow them closely.

But this doesn’t come without practice and discipline. De-escalation training has long been a key part of civil resistance training, and protestors need such training now more than ever. Princeton's Bridging Divides Initiative maintains a state-by-state directory of 100+ de-escalation, bystander intervention, and community safety trainers. They also publish de-escalation guides for specific contexts (poll workers, elected officials, bystanders).

More fundamentally, protest tactics do not need to be aggressive. Humor and mockery can be very effective, especially in today’s media ecosystem. We’ve previously analyzed The Strategic Logic of the Naked Bicycle Ride. There are even training programs on “serious play.”

The path forward is to fix immigration policy

For any of this matter, we need to address larger issue of rationalizing immigration policy.

As we have discussed previously, there is a compromise that almost everyone will accept: strong border security and immigration enforcement combined with a path for unauthorized migrants already here to stay legally. Both are required: what we have now is strong border control without a good answer for those already here, while if there is a path to citizenship without strong border control then it’s an open invitation for the rest of the wold to come here. This combination is in fact a popular position.

In a hopeful sign, there is a bill in Congress that would do exactly this:

Rep. Maria Elvira Salazar (R-Florida) introduced the Dignity Act of 2025 (H.R. 4393) in the U.S. House of Representatives on July 15, 2025, a bipartisan effort to strengthen border security in the United States, provide undocumented individuals with an opportunity to obtain legal status if they meet certain requirements, and update aspects of the U.S. legal immigration system. The bill is co-sponsored by Rep. Veronica Escobar (D-Texas), and an additional 10 original Republican co-sponsors. It is a revised version of the Dignity Act of 2023.

The big picture

In the wake of tragedy, there is a rush to blame. But what happened this week in Minneapolis is about far more than the handful of people who were there. This terrible outcome is the result of larger structural forces; it was both predictable and predicted.

Despite all of the attention that this latest atrocity has received, if our goal is to prevent further loss of life, focussing on this individual incident won’t help. Americans will probably never agree on what happened. But my experiences over the last two days, talking to people about it, has convinced me that we can agree on where to go from here.

Quote of the Week

Billy Joel could do a whole new We Didn’t Start the Fire just about the last week

-Matt O’Brien

This has certainly been a tumultuous year in American politics, but better conflict, bridge building, depolarization — whatever you call it, it was more popular than ever. We’ve covered it here, from immediate issues (like immigration) to ancient questions (like when force is necessary). Here’s our best work this year, a combination of reader and editor favorites, organized into five different categories.

5. Media & Truth

But What If Their Side is Actually Wrong?

This one became an instant classic, and readers continue to share it. We often hear of the evils of “bothsidesism,” but even if the other side is actually wrong there are excellent reasons to fairly represent their views: to create diplomatic meeting grounds, to reduce stereotyping, because the best arguments come from engaging strong opposition, and because calling them evil is both spiritually and strategically harmful.

Meta Drops Fact-Checking Because of Politics, but Also Because It Wasn’t Working

Mark Zuckerberg’s decision to end Meta’s third-party fact-checking program was politically motivated, but the program also faced real challenges: it never achieved scale, checks often came too late, and conservatives rapidly lost trust. Yet research shows that conservatives actually mostly agree with the calls that professional fact-checkers make, and crowsourced fact checking systems are more trusted, more scalable, and often respond faster than professional fact checkers.

4. The Perception Gap

Why Do They Think We’re Extreme?

Americans overestimate how many people on the other side hold extreme positions by 20-30 percentage points—and it’s roughly symmetric. Why? We consume distorted media (hyperpartisan politicians get 4x more coverage) while living in increasing geographic and social isolation. The result is that we fear and misunderstand each other based on media images of people we mostly never meet.

How Many People Actually Approve of Kirk’s Murder?

After Kirk’s assassination, social media filled with cherry-picked screenshots of people celebrating or calling for more violence. We sampled posts on X and BlueSky and found that less than 1.5% called for violence—consistent with research showing only about 2% of Americans would support politically-motivated murder, though each side imagines the other’s support is 20 times higher.

3. Us and Them For Democracy

Why Build Bridges when Democracy is Burning?

“They’re the problem, why should we listen to them?” This is a question that bridge builders get in any conflict, all the more so when the foundations of democracy are threatened. But even if one side poses greater danger, the winning strategy is for “us” to ally with some of “them” against authoritarianism—building coalitions around shared values like anti-corruption or honest institutions, rather than partisan identity.

Repolarize to Depolarize (podcast)

We were thrilled to interview political scientist Jennifer McCoy, who explained that progress may not come from reducing polarization, but from polarizing differently—what she calls “transformative repolarization” around broad principles rather than partisan identity. Drawing on cases from Venezuela to the American Progressive Era, she argues the goal is to build coalitions big enough to defend democratic institutions.

2. Difficult Issues

What You Can Learn from 4,000 Conversations Between Israelis and Palestinians (podcast)

Conflict theory is great, but practice is better. Adam Becker has personally moderated thousands of conversations between Israelis and Palestinians after October 7th, discovering that success depends more on who talks than what they talk about. Key insights: text-only conversations are toxic, small humanizing gestures can shift everything, and each side has major misconceptions about the other. Plus: how to use AI to build a platform for difficult conversations.

We Have to Talk About Immigration

You’ve read a million articles about the immigration crackdown, but ours isn’t trying to convince you the other side is wrong. While immigrants commit fewer crimes and generally benefit the economy, the recent surge put real stress on local governments, and American culture is changing. Nonetheless, support for legal immigration has actually increased since Trump’s election. The path forward might be strong enforcement combined with a path to citizenship, a combination most Americans support.

1. Eternal Questions

Does Protest Even Work?

Yes—if it’s big enough and nonviolent. Protest moves votes, and no government has withstood 3.5% of its population mobilized against it. But violence dramatically hurts your chances: nonviolent campaigns are more than six times likelier to succeed than violent ones facing the same repression, largely because more people can participate and security forces are more likely to defect.

When to Fight and When To Talk

Perhaps the deepest, oldest question in peace and conflict is how to know when the other side can only be met with force. The modern answer is: the adversary is not monolithic! Different factions require different approaches. Most are “persuadables” who will respond to shared values, while “traders” will make deals, and only a small fraction are true “incorrigibles” who require coercive power. And unfortunately, the deeper we are into politics, the less accurate our estimates of who is persuadable.

Quote of the Week

There is a theory the Grinch’s disposition was partly a result of a toxic conflict/polarization cycle.

-Zachary Elwood

We all naturally focus on the things that need fixing, and this publication is no different. For the holidays, we highlight reasons for hope instead.

Better conflict is catching on

It’s hard to find definitive statistics on how many people or organizations are working on better conflict in America, but pretty much every plausible indicator is up.

In April, the Listen First Project and the Civic Health Project released a comprehensive report on the state of America’s bridging movement, writing,

Absent a robust, centralized cataloging function, it is challenging to keep abreast of the “everything, everywhere, all at once” pace and scale at which bridge-building, peace-building, conflict resolution, and civil discourse programs are being implemented nationwide. In higher education alone, hundreds of college campuses entered Fall 2024 with newly invigorated curricular and co-curricular offerings intended to reach thousands of students per campus in the current academic year.

In response to skyrocketing demand, capable grassroots leaders in America’s bridging movement are shifting from “push” strategies, i.e. inspiring individuals and institutions to take up bridge-building, to “pull” strategies, i.e. responding to institutional demand, partnering with community and civic sector leaders, and packaging offerings for more rapid and scalable consumption.

Let’s get quantitative about this. Braver Angels has grown from 12,000 to 14,000 members in the last year. BridgeUSA has expanded from 63 to 80 campuses. But perhaps one of the most systematic signals is Democracy Fund’s survey of organizations in the “democracy” space, a broad area which includes everything from election integrity to nonprofit journalism to polarization research. They find that most funders are maintaining or increasing their giving.

This is no doubt in part due to the upcoming midterms, but February’s survey showed the same thing.

Federal rancor, local cooperation

According to a May 2025 report by Civic Pulse, 83% of local leaders say polarization substantially harms the country, versus only 31% for their own communities.

Meanwhile, Congress is as polarized as it’s ever been, judging by their voting record. But at the local level, there’s a lot more bipartisan cooperation than you might expect.

Democratic Kentucky Governor Andy Beshear has been working wonders in a Red state. 50% of Kentucky Republicans approve of Beshear, the highest GOP approval for any Democratic governor, making him a serious contender for President in 2028.

In Minnesota they have no choice — the 67-67 tied House forced the creation of a formal power-sharing agreement with co-chaired committees requiring bipartisan votes to advance legislation.

And perhaps this spirit of cooperation is most striking in Colorado. Despite supermajorities of Democrats in both the House and Senate, 52% of bills received majority Republican support, including housing construction reform that had failed for years, nuclear energy legislation (GOP-led), and a George Floyd-era police reform law (SB20-217).

Polarization as poetry

This year saw the publication of WE, a collection of poetry about polarization by April Ossman. Here she is recounting a moment of fraternizing with the enemy at an airport.

Game of the Week

Senatordle - can you tell if a Senator is a Republican or a Democrat just by looking at their portrait?

Happy Holidays, everyone.

In any conflict, there will be a fight over the politics of the next generation. Aside from the recent struggle over universities, there have been deep arguments over primary school curricula and even the books in school libraries. How then should a divided country decide what to teach?

I was surprised to discover that most other countries have already found a way to address these tensions, as Ashely Rogers Berner writes:

[T]hese governments fund a wide variety of schools (religious, secular, pedagogical) but also expect all of them to impart shared knowledge in the major subjects. Nor are pluralistic systems rare; quite the opposite. A full 171 out of the 204 countries UNESCO recently surveyed partner with non-state actors to deliver public education. This is news to most Americans, as is the fact that several key human rights instruments endorse the rights of cultural minorities to educate their children according to their families’ values.

What does this look like on the ground? The Netherlands funds 36 kinds of schools (e.g., Montessori, Jewish, secular) on equal footing. Australia’s federal government is the top funder of tuition to private schools, many of whose students come from the lowest economic quartile of the country. Sweden and Poland let funding follow students to their school of choice. Nations as different as Colombia and Nigeria include both state and non-state providers in their education systems. The list goes on and on because most countries have plural school systems

These systems combine two crucial ideas. First, school choice for parents, but funded by the government. This is similar to the idea of school voucher programs, long discussed in the US, a version of which is set to begin in 2027. Second, a core curriculum that all students must be taught, regardless of the school’s overall orientation.

Furthermore, the most academically successful plural systems also require critical common content to be shared across all these different types of schools. Students take subject-specific exams every few years that, unlike the skills-based assessments our states administer, actually require mastery of core knowledge. … The Netherlands even funds creationist schools, but kids in creationist schools have to demonstrate competent knowledge of evolutionary theory

This approach offers a way to thread the everlasting tension between local self-determination and national standards. Educational pluralism may be radical by current standards, but it fits right in with the core American values of liberty and tolerance.

How to do pluralism in the classroom

Earlier this year the journal Democracy & Education published a special issue on polarization, including three studies that grapple with how to teach charged subjects in high school.

Role playing for learning: PurpleState is a game where students role-play as interns at a political communications firm, designing media campaigns on gun control, climate, and police reform from research briefs and real polling data. Focus groups found students transferred skills to real life: questioning the aims of political ads and seeking out multiple perspectives.

Inter-school dialogues: students from politically and demographically different schools 150 miles apart discussed election issues on a video platform. The experience helped students recognize they were in political bubbles and motivated them to engage across difference.

Debate vs. Discussion: a comparison of “deliberation” vs. “debate” formats among high schoolers. Deliberation produced more comfort (87% felt respected) and less disagreement, while debate was less comfortable (74%) and more disagreement was voiced, but students also said they were exposed to more new ideas. Students with personal stakes (LGBTQ students on the Equality Act, hurricane survivors on climate) mostly reported positive experiences.

Is or isn’t Critical Race Theory taught in school?

Like all heated political battles, the big public argument over education is often light on facts. It’s therefore refreshing to find work that asks a very basic question: what kinds of politics are American high school students being exposed to today? Well, you could do a survey and find out:

The authors of Bridging the Divide over Critical Race Theory in America’s Classrooms also asked high school students questions about their teachers’ use of various bit of language, including both “Black Lives Matter” and “All Lives Matter.” As above, there’s some of everything, but it’s not always easy to distinguish between discussion and indoctrination.

We would expect, or at least hope, that good-faith conversations in healthy classroom environments would address contemporary debates and present multiple perspectives. And indeed, students report overall that their teachers show an appreciation for viewpoint diversity. Seventy-six percent of students say teachers never or rarely make them feel uncomfortable to share opinions that differ from their teachers’, while 54 percent say their teachers make them feel comfortable to share differing views often or daily. At the same time, 18 percent report their teachers have made disparaging remarks about conservatives or Republicans, and 19 percent say their teachers have made disparaging remarks about liberals or Democrats.

So are our children being indoctrinated with Blue (or Red) ideology?

Overall, one thing that seems clear from our data is that, in the public debate about CRT in education, people on opposing sides are largely talking past each other. Those who think CRT-related instruction is nonexistent should know that it isn’t. Those who think it’s an epidemic should be aware that this isn’t true either. And those who are concerned about indoctrination in education should know that most youth report forming their political beliefs from sources outside of the classroom.

Image of the Week

Build Peace is a shockingly global gathering of “conflict nerds” now in its 12th year. One of the organizers, Helena Puig Larrauri, is a friend and co-author, which is how I ended up writing about the very first conference at MIT in 2014. This year, it was just outside Barcelona at the International Catalan Institute for Peace (ICIP).

AI for consensus building in Yemen

Felix Kufus gave a short talk presenting the use of Talk to the City, a platform which uses LLMs to synthesize diverse public input, to create a vision statement agreed across all seven political parties in Yemen.

This was a unique use of multiple AI technologies. The team, from Finnish peacebuilding organization CMI, deployed Talk to the City over WhatsApp, asking youth from across the country to answer 10 open-ended questions. And they could answer via voice messages which were then automatically transcribed, so people shared very long, very open ended accounts. There was was live monitoring through a dashboard, showing how many responded from which parties.

Talk to the City then synthesized all of this into a report showing six main topics, 18 subtopics, and over 1000 “claims.”

Kufus emphasized the role of trust in this process. “You open a link and give your political views, is this safe?” They were able to succeed because their organization already had connections to stakeholders

“This allowed qualitative analysis on quantitative scale,” he said.

As far as I can determine, this is the first use of LLMs in a deliberative peace process. (Previous work has used Remesh and Polis in Yemen and Libya, but these used collaborative filtering and text clustering algorithms, not LLMs per se.)

Platform polarization footprints in Kenya

Build Up presented their work measuring the polarization footprint of social media in Kenya. The basic idea is to measure what fraction of people’s feeds is “polarizing” content, which they did on five different platforms.

To do this, they recruited 5,000 Kenyans to install a custom browser extension that, with the participant’s consent, downloaded a snapshot of their feed from these five platforms. They then used an LLM-based Swahili-language classifier to detect polarizing content. (See also the POLAR multingual polarization detection benchmark — not used here but very cool.)

These numbers may not seem like a lot, but put them into context of social media use:

This means that for every hour you spend on X, on average you’re spending a minimum of just over 2 minutes on polarizing content. 55% of Kenyans are spending 3–9 hours on social media per day; if that time were spent on X, then your average Kenyan would spend at least 7 to 21 minutes exposed to polarizing content. Imagine being shouted at for 7 to 21 minutes every day. No wonder we feel angry.

The point of all of this measurement work is to make the polarizing effects of platforms visible. These effects are essentially an unwanted externality of social media, and like all externalities it must be measured before it can be internalized. In this case, the team has proposed a “polarization footprint tax” that would incentivize platforms to have less negative effects.

I have some questions about whether this methodology is truly measuring the right thing, how a tax would work, and so on. You probably have questions too! In my view, the value of this early work is to make the problem visible and start the conversation. Even if the measurement methodology or the proposed policy fix are wrong, now we’re at least talking about the problem.

On the ground in Syria

While technology runs through the heart of the event, I was personally most inspired by the people doing lots and lots of old fashioned mediation. Perhaps this is because I have learned the painful lesson that knowing about good conflict isn’t the same as being able to actually do good conflict.

I attended a workshop led by Ribal Azzin of Olive Branch, a Syrian peacebuilding organization. He brought youth together across all regions, online, to discuss the future of their country. At first his organization approached people individually, then gathered them together in group calls, and eventually all of the participants met in person in Damascus.

Part of their method is a code of conduct for discussion, including what language to use when describing sensitive topics. This was written in consultation with the participants themselves.

Political Correctness: Use terms that do not incite tension and division among partners, as suggested below:

…

Kurdish Forces: Referring to the Syrian Democratic Forces that control areas in northeastern Syria and replacing it with the Syrian Democratic Forces (SDF) or other non-controversial terms.

But that’s just the frame. What’s the content? Azzin started his workshop with an exercise. He put a slide on the screen that said:

He asked people to physically move to one side of the room or the other depending on which side they felt like they were on. Most went to the “minority rights” side.

“Why did you pick the side you went to?” he asked. He got one person from each side to explain their reasoning. “The minority is always threatened in conflict,” said one person. “But look at this room,” said the person on the other side, “in the peacebuilding field no one wants to speak for the majority, and that causes problems.”

Azzin put up the next slide.

Again we split into groups; again one person from each side explained their logic. Then one more slide:

The punchline is that all of these are false choices. When working with Syrian youth, Azzin played a game called “who benefits?”

• who benefits when minorities and majorities see each other as threats?

• who benefits when dignity and stability are framed as opposite goals?

• who benefits when national identity is put against religious or ethnic identity?

“We all inherited this conflict,” Azzin would say to his Syrian participants. “Binary options are a tool manipulators use.”

The challenge for those who want peace is that anyone who rejects this false choice is going to be attacked.

There’s no middle ground, there are no middle voices, because this is conflict. This is the real issue with peacebuilders, because they always want to stay in the middle, but sometimes it’s framed [by others] as, you are stabbing your country or your group in the back.

When you say, “us vs. them, it’s not true,” they hear, “they are a threat to our country, a threat to our existence.” People called me a spy, they called me a lot of bad stuff.

The middle voice is always oppressed.

Azzin’s answer is simple: “Show them how much we are going to lose if we don’t work together.”

I couldn’t help thinking about how these ideas apply to the American situation. I’ve certainly been called bad things in conversations where I’ve tried to convince one side that they should work with the other. Yet, while there are real differences in values that cannot simply be papered over, we are losing so much more than we have to in the way we are fighting.

Very rights-based

I was surrounded by peace nerds talking about inclusion and plurality, but I have to confess, sometimes the event didn’t feel very inclusive or plural. Speakers often talked with the unstated assumption that left-ish politics are obviously correct. At times I felt the uncomfortable dissonance of being around a group of people explicitly talking about cooperation across political differences, with no one who was actually different in attendance.

I talked to others about this, and got some useful perspectives. First, we were seeing in part the politics of our Catalan hosts. ICIP is very classically leftist social justice oriented, if the posters on the walls are any indication. When Build Peace was in Nairobi two years ago, it was a much more conservative vibe in many ways.

Second, while I often saw things in left vs. right terms, that axis doesn’t always translate globally. Others saw things more as the difference between talking about peace and talking about rights. “This was a very rights-based discussion,” said one person. And indeed this was true. We heard about women’s rights, immigrants’ rights, trans rights, and indigenous rights. Nothing wrong with any of those, but they’re all left coded — and advancing rights is related to but not the same thing as advancing peace.

The PeaceTech landscape is expanding

It’s clear we are rapidly entering an area where AI will become an expected part of public deliberation processes. This post is but a small sampling of the space. For more:

• You can watch videos of all the plenary sessions here (see the program for a full listing of the talks).

• The Council on Tech and Social Cohesion (of which I am a member) recently considered the opportunities and threats of AI for peace.

• Project Liberty just posted a list of deliberative tech tools, Platforms that Unite instead of Divide.

If you like this sort of post, see also our Tech for Social Cohesion conference report from 2023.

Quote of the Week

Sometimes in the effort to fix problems with humanity, we think if we can just shut the right people up, these problems are going to go away. And it never, ever works.

- Greg Lukianoff

Politics make family difficult for a lot of Americans. A recent poll estimates that one in five of us is estranged from a family member because of political disagreements.

If you’re one of these people, know that the situation is probably not impossible. Every year, we search for the best resources so you don’t have to.

The Basics

I recently ran across Carrie Graham’s Thanksgiving 2020 guide and sure, this was written after the previous election but the basics don’t change. And it’s by someone who has a ton of experience coaching people through this, foundational stuff like this:

1. Gauge Your Readiness. Gauge Others’ Readiness.

Remember nobody has to talk about anything you/they are not yet ready to talk about. Both parties need to feel emotionally prepared to attempt a difficult conversation. It’s a “mutual consent” situation.

…

2. Respect the reality of paradigmatic differences.

Be aware of how strongly we each hold on to our narratives, and that others are living according to a different script than the one in your head. None of us are exceptions to this. Oftentimes, we think if the other party just had x information, or understood more facts we have read, they would either change their minds and agree with us OR we could then officially write them off as crazy. It is almost never either of those.

…

3. Beware of monoliths.

Our political system often divides us into two political identities. We do not have to live that way or treat each other as such.

etc. Here’s another pretty good short general guide.

Maybe just not talk about it?

A “no politics” rule is a tried and tested boundary, if it’s made explicit.

It’s also useful to have tactics for sidestepping barbed comments. And this one’s got a strange name, the Associated Press reports:

When gray rocking, you’re purposely making yourself uninteresting by politely offering short, non-committal responses rather than getting sucked into a high-drama exchange. The idea is to deprive the other person of the attention or reaction they want, prompting them to move on.

… when gray rocking, you’d sidestep conflict with a bland response, like, “Interesting opinion. How’s work?”

Useful, but don’t over-rely on it or people will get angry, and anyway it’s no way to have a sustainable relationship.

Practice not being a dick

If you do talk about politics, you’re probably going to have feelings that will make you want to say something unproductive. Usually this happens when we are feeling hurt or an important value is threatened. Expect this feeling. Prepare for it to come. And listen for that little voice telling you not to say the thing.

If you find you have said something hurtful anyway, apologize and reword to say the best and kindest version of what you needed to say. Also consider learning something and changing your mind. Model the behavior you expect from both of you.

People who are actually good at this are rare and beloved. It’s not a vow but a practice, and it takes practice. Ask me how I know.

Getting help with a difficult relationship

There are more or less two options.

First, you can learn better conflict and dialogue and relational skills on your own. I see you’re a BCB reader, that’s a great start. But you could also get coaching, or even better learn by practicing dialogue with other people. Braver Angels can set you up, among other organizations.

Second, you can seek professional family coaching or therapy together, if your family member will agree. Some therapists specialize in political differences within families.

Thread of the Week

However you go about it, here’s to enjoying togetherness this season!

Blue misinformation hits different

We’ve elsewhere made the somewhat tedious point that there’s more misinformation in right-leaning sources. (Tedious because the typical reaction is either “duh” or “you’re wrong” without reviewing any evidence.)

However, that’s not to say the Blue information ecosystem is everywhere in good shape. Conspiracy theories seem about evenly distributed, for example. And there is the perennial Red complaint that “woke” is antithetical to truth (Elon is fond of this one, hence Grok is “maximally truth seeking.”) That seems overstated, but there is plenty of misinformation on the left too. It’s just… got a different vibe. Dan Williams finally found a good way to talk about this:

I prefer Joseph Heath’s term “highbrow misinformation”. This is the sort of misleading content spread and consumed by highly educated professionals within prestigious institutions (legacy media, universities, NGOs, public health authorities, etc.). In the current context, this is the most relevant kind of misinformation because it circulates among the very people and organisations most vocally concerned about misinformation. So, it’s the kind of misinformation that opens them up to accusations of bias and hypocrisy.

Heath is a climate change academic, and coined “highbrow misinformation” to talk about a certain kind of Blue falsehood. Did you know that global warming isn’t predicted to reduce world GDP? That’s right, there’s not expected to be any degrowth, even under worst case scenarios. If you’re confused, it’s because true statements have been misunderstood or misrepresented, says Heath:

A very large number of people believe that climate change, under the high-probability “loss and damage” scenarios, within the next few decades, stands poised to lower the standard of living of future generations below what it is today.

Of course this may happen, but it is absolutely not what any of the studies say, or what the IPCC loss and damage reports say. The misrepresentation stems from the way that these studies present their results and how those presentations subsequently get reported.

…

When the authors say that between now and 2050 “output declines by 19%” they do not mean that it literally goes down. Based on underlying growth trends, world economic output is likely to double by 2050. What the authors are saying is that, rather than just the expected increase of 100% between now and 2050, the world economy could be increasing by 119% if there were no climate change. In this respect, climate change will “decrease” output by 19%.

Of course, many people will choose not to believe these projections about GDP growth. Environmentalists have been predicting the end of growth since before I was born. The observation that I am making here is the much narrower one – that the results of this study are being misreported and misunderstood, which is causing the average educated person to have false beliefs about the assumptions that structure environmental policy debates.

So now you know.

I bring all this up because, naturally, misinformation can drive conflict and destroy trust. It has been understood for some time (at least among those who’ve actually looked into it) that there’s a lot of bad information in certain segments of the Red ecosystem. But there’s also some pernicious misinformation circulating in Blue culture. While there’s quantitatively less than on the Red side, Blue misinformation has potentially serious consequences because it’s institutionalized. Williams again:

The main question that I want to explore here is how we should understand, evaluate, and respond to highbrow misinformation. And the main point that I want to make is that too many people either (1) deny or downplay the phenomenon, or (2) exploit its existence to favour an “anti-establishment” information environment that is much worse than what it seeks to replace. Highbrow misinformation demonstrates that our most prestigious knowledge-producing institutions must urgently be reformed, not destroyed or replaced by “contrarian”, “heterodox”, or “alternative” media.

Groypers in DC

Meanwhile, MAGA. If you don’t know what a “Groyper” is, you have lived a merciful life indeed. The word refers to a fanboy of white-supremacist and anti-semitic influencer Nick Fuentes, represented by a frog.

We have previously reported that there don’t seem to be any more “Nazis” in the U.S. than before, at least if you count people involved in organized white supremacy (or hate crimes). Yet, as we also said at the time, there’s no question that some bad ideas continue to have a foothold in Red culture. The past month has shown it publicly.

First came the text messages of Young Republicans praising Hitler, etc. The GOP found this serious enough to disband the New York Young Republicans, while the Vice President sort of defended them. Then we had the spectacle of the president of the Heritage Foundation defending Tucker Carlson from criticism that he softballed an interview with Fuentes, after which many staffers resigned.

All this has led to much discussion about fracturing on the right, but I’m imagining how little any of this gossip is going to matter a year from now. The more important question is: how far does this go beyond a handful of people saying bad things?

Unfortunately, I’ve since run across two credible reports that the Groypers are deep in DC. First is Richard Hanania, who I trust on this because he’s a former alt-righter:

If you’re a conservative now who isn’t an ethnonationalist, you’re just completely outmatched. The other side has the numbers and the energy, and there is such a thing as the internal logic of a movement. Democrats once made their entire moral universe center around oppressed minorities, so one couldn’t expect them to stand up to pushy black trans activists around 2020. The equivalent for the right now is the young male who is racist, angry, and sexually frustrated – the Groyper. People like Trump and Vance can’t denounce them, nor can supposed intellectuals like Yoram, because Fuentes is just the undistilled version of what they’re selling, which is white grievance, showmanship, and anti-liberal posturing.

Conservative Catholic gadfly Rob Dreher tells a more complex story from his recent trip to Washington:

The claim that I first floated in this space last week, quoting a DC insider who said that in his estimation, “between 30 and 40 percent” of the Zoomers who work in official Republican Washington are fans of Nick Fuentes — that’s true. Was confirmed multiple times by Zoomers who live in that world.

If you think being Christian is some kind of vaccination against anti-Semitism, you’re wrong. Even young Christians — especially trad Catholics, I learned — are neck-deep in anti-Semitism. They even use it as a litmus test of who can and can’t join their informal social groups.

Not every DC Zoomercon who identifies with Fuentes agrees with everything he says, or the way he says it. What they like most of all is his rage, and willingness to violate taboos. I asked one astute Zoomer what the Groypers actually wanted (meaning, what were their demands). He said, “They don’t have any. They just want to tear everything down.”

Then he went on to explain in calm, rational detail why his generation is so utterly screwed. … F—k that, seems to be the reaction now. Fuentes is a living, breathing illustration of what Hannah Arendt meant when she spoke of people willing to revel in transgression such that they tear down the pillars of civilization, just for the fun of seeing those who had been gatekept away breaching containment.

…

Hannah Arendt, man: like I’ve been saying, she is the diagnostician of our Weimar moment. I won’t repeat myself here — I’ve been quoting her a lot lately — but after these last three days in Washington, I am more convinced than I have ever been that we are moving towards some kind of totalitarianism — or at best, authoritarianism. It could go either way, Left or Right. Said one source, “They [Groyper types] look at Mamdani’s success in New York, and think, ‘Why can’t we do that?’” — meaning, why can’t we try to get a true radical into power?

Those on the Blue side might be shocked to hear this — is Mamdani really that radical? How is Trump not radical? But Dreher says the Zoomercons see Trump as “an out-of-touch Boomer.” As for Mamdani, if you want to understand the American conflict, it’s vitally important to understand that many conservatives (and others) experienced creeping authoritarianism from the left over the last decade. This was recently well put by a conservative writer at Tangle:

And yet, despite being disturbed by what I see as a shift in the party’s ideology, I can still understand my friends and family who find the Trump administration the “lesser of two evils” — I myself even felt that way during the election. While I think major players in the Trump administration are acting in openly authoritarian ways, I don’t think all or even most of the GOP base supports this approach to government. In fact, I think their support of Trump and the GOP is driven by conservatives’ sense that the left has been trending toward authoritarianism. Understanding this feeling — even if you think conservatives are wrong to feel that way — is essential to moving forward from the moment we’re in, and reuniting our increasingly divided nation around the same common ideals.

This will be controversial, so I want to be clear about what I’m saying here:

• There are authoritarian tendencies in current left politics (see here and here for the detailed argument).

• There are way worse authoritarian tendencies in current right politics.

• Whether or not you agree that there are left authoritarian tendencies, many conservatives have had experiences they describe that way, and this deeply shapes our current politics. As one of the Zoomercons that Dreher spoke to put it, fascism doesn’t look so bad if you believe the only alternative is (the authoritarian kind of) communism.

The Good News

Almost nobody actually wants democracy to fail. Many people worry that it is failing, but different groups worry for different reasons. Dreher again:

Let me emphasize here that my interlocutor was NOT cheering for fascism. As our conversation went on, I heard real despair — and well-informed despair — that democracy is going to hold in the West, because the conditions that make for a viable democracy are disappearing: the dissolution of a common culture, the collapse of religion in his generation, the material impoverishment of his generation, and so forth.

Quote of the Week

REP. MARJORIE TAYLOR GREENE: Dana, I think that’s fair criticism, and I would like to say humbly I’m sorry for taking part in the toxic politics.

It’s very bad for our country, and it’s been something I’ve thought about a lot, especially since Charlie Kirk was assassinated.

I’m only responsible for myself and my own words and actions, and I am going—I’m committed, and I’ve been working on this a lot lately—to put down the knives in politics. I really just want to see people be kind to one another.

And we need to figure out a new path forward that is focused on the American people, because as Americans, no matter what side of the aisle we’re on, we have far more in common than we have differences. And we need to be able to respect each other with our disagreements.

(source)

Not many people have the chutzpah to put themselves right in the middle of the Israeli/Palestinian conflict, and even fewer of those are trying to build their insights into an AI-powered peacemaking platform.

In this podcast, Adam Becker shares his unique journey from tech entrepreneurship to facilitating thousands of conversations between Israelis and Palestinians. Along the way he discovered the power of small humanizing gestures, found that text-only conversations are toxic, and learned that the success of a conversation depends more on who talks than what they talk about. He’s distilling all of this into HeadOn, a data-driven platform specifically designed for the most difficult conversations. Plus: what Palestinians and Israelis misunderstand about the other.

Jonathan Stray: Hello everyone and welcome to the Better Conflict Bulletin podcast. I have with me today Adam Becker, who has moderated 4,000 conversations between Israelis and Palestinians after October 7th. Hi Adam.

Adam Becker: Jonathan, thank you very much for having me today.

JS: I don’t know anyone who’s done this many conversations, especially not in the last few years. How did you get into this?

AB: I got into this through a very unusual route. By training I’m a tech entrepreneur. I went to Cal [UC Berkeley] and as soon as I graduated I was trying to find a way into the Israeli and Palestinian conflict space to use tech and startups as a vehicle for effectuating change in that space.

At some point I began to gradually shift my focus to the American political culture wars scene. I created a platform to help conservatives and liberals speak more effectively.

This was when Trump was first making a credible bid to the presidency. When he won, I then shifted gears one more time and started an AI company to help Democratic campaigns do their fundraising more effectively. I did that for a few years. I moved from San Francisco to DC. I lived there on Capitol Hill until Biden won. By that time our app helped Democrats raise about a hundred million dollars. I moved to New York, I got deeper into the AI world and machine learning world.

And then October 7th hit. The next day I moved back to Israel, where I had grown up. First of all, it was just almost impossible to make it to Israel. You can imagine the airspace was quickly closing down and every airline was shutting down all their flights. My goal was to make it to Tel Aviv in the morning, and to simply have a thousand Israelis speak to other people from around the world.

I wanted to see first of all, whether Israelis were able to articulate their own position because I kept hearing a slightly different narrative from the one that Israelis tend to embody. So there was a lot of misunderstanding. And on the other hand there’s a lot of misunderstanding among Israelis about what the world was actually critical of Israel of. So I thought it would be very valuable to try to open some conversational channels. That’s how it began, in the early days of the war.

Fast forward a couple of years, we’ve been facilitating thousands of conversations on the platform.

JS: So what drew you to trying this insane thing?

AB: Have you ever been on these apps like the Omegle and Chatroulette or Monkey app? You have those apps that just connect people randomly with one another.

And in Israel, during a time of war, everybody could see that I’m calling in from Israel. And it’s an immediate politically charged confrontation. That’s how every conversation began.

I became absolutely obsessed with trying different techniques to de-escalate those conversations. When something starts like this, is it even salvageable? Over time, I’ve started seeing patterns, and I was like, “My God, this is not just salvageable.” We’d end up having incredible conversations, even though they start out with them doing a Nazi salute.

Oftentimes, we end up becoming friends and then they give me their phone number and we’re exchanging links on WhatsApp and I send them some links from the Better Conflict Bulletin. I started seeing that there’s a lot of techniques for de-escalating.

As I said at the beginning, I don’t have a background in conflict resolution but I built up a lot of intuition for how to do these things. And I started seeing actual effect, and actual impact. And it was shocking to me.

Humanizing gestures

Just to give you some examples, I would pretend that my mom was calling me and I’d say, “Sorry, can you give me a moment? My mom needs help.” And then I’d talk to my mom for 10 seconds. And when I’d put down the phone, immediately something changed. They’d see me as a human.

I would take my phone or my computer to the kitchen and take out soup and heat up the soup. I’d see that as soon as they’d see that I’m a human, it’s a totally different story and they start reacting differently.

I started seeing all of these patterns, and at some point it hit me that this is a data problem. If only we had sufficient data about how to de-escalate. I shouldn’t be doing all this A-B testing. It shouldn’t just be me. So at some point we started building AI to help us almost design conversational experiences.

I’ll give you another touch point of how transformative these experiences were. One time, I was connecting with a woman in Tunisia. And the bright screen was lighting up her face. She was laying in bed. And she saw that I’m from Israel and she didn’t say a word. She just started singing the Lucy Thomas rendition of the Hallelujah song. I’d never heard it before, but she played it. Our eyes were just transfixed. Here you go, like in the middle of a war and this girl is singing me the Hallelujah song. I had goosebumps.

So I realized pretty quickly that this is a different type of experience from an academic, ceremonious negotiation exercise at Camp David, right? I’m in the gutter with a bunch of people, but it seems like there’s millions of them. And it seems like there’s actual impact that you can drive here.

The role of conflict outsiders

JS: Then at some point you thought, “I want to try connecting Israelis and Palestinians.” So how did you go about that?

AB: When the conflict had globalized after October 7th, that is when every single person around the world started having strong opinions about what was happening, I started seeing other people as potentially very valuable in this conversation.

In general, they might have a pernicious effect on the conflict, but if given the right tools, I thought and hoped that people who are not in the conflict, who are not as traumatized as the Israelis and Palestinians themselves have a role to play. And we need to find ways to pull them in.

And so the first thing that I did was not just go to Israelis and Palestinians, but go to Americans. Go to Americans, go to Europeans, go to North Africans, go to anybody else who is interested in the conflict and who is interested in talking to Israelis and talking to Palestinians about what’s happening on the ground.

One of the things we did is we just started walking around in Israel and in the West Bank and recording people answering questions that the rest of the world submits to a YouTube channel. And we started getting hundreds of thousands of views. And then we invited those people from those views to come and actually extend the conversation into a more meaningful session on our platform.

JS: Do you still have this YouTube channel? Is it still going?

AB: Yeah, it’s still going. It’s been doing really well for for 15 years. So I go to Israel about twice or three times a year and we go and we do recording. The guy who started this, his name is Corey Gil-Shuster. Anybody can find it. You could just type “The Ask Project.” Absolutely fascinating window into what people actually say when you ask them questions. There’s a lot of disagreement about whether that’s what they think, but that’s certainly what they say when you catch them unsuspecting in the middle of the street in Ramallah or Jenin or Tel Aviv.

So in the middle of the war, there was a lot of appetite to understand what Israelis feel, and how Palestinians in the West Bank are experiencing it. So that was one source of interest.

And it’s continuing to go. And so we’re still getting a lot of people that are coming by having watched the channel.

Aside from that, I also wanted to see what it would look like if we brought interns to speak. By interns, I mean American college students who wanted to get exposed to different ideas. They were looking for a space in which to do this. And we onboarded about a thousand of these interns.

JS: So you’re recruiting American college students as “interns” to your platform, your company. What was it called at this point?

AB: So right now it’s called HeadOn. At the time it was called Dugree, which is both a play on “do you agree?” and also it’s a word in Arabic and in Hebrew that both understand, it means “straight up”, “no BS conversation.”

JS: I see, okay. So, you recruit Americans to this Dugree platform to have conversations with Israelis and Palestinians in Israel and the West Bank, is that the idea?

AB: And one another.

JS: They are talking to one another, and also to people on the ground?

AB: That’s right. And more and more people from around the world. We had Islamists from Egypt and we had radicals from all over the world and they all got to mix together in a fascinating blend.

We just started by seeing, what it looks like for people to have these conversations in the first place. And can we start to envision an infrastructure for scaling these conversations?

JS: And these are one-on-one conversations, no mediator or facilitator?

AB: No mediator or facilitator. We try to build some guard rails to keep the conversations constructive with the full humility that we don’t know how to build guard rails, and that guard rails are likely to be very context-specific and context-dependent. And ultimately we will need to find generalizable and scalable guardrail mechanisms.

It took me a long time to get deeper and deeper into the relevant literature on what works and what doesn’t work — at least from an academic perspective. A lot of that is thanks to you and to the Better Conflict Bulletin, so I appreciate all the work that you’ve done. At the time that we were starting — we had these thousand interns — we knew absolutely nothing about conflict resolution.

Trying and failing and trying again

JS: So how’d it go? Did conversations blow up? Did they resolve? Did people learn things? Did they come back? I’m so curious, if you just put people in a room, what happens?

AB: It was absolutely incredible. People loved it and everybody reported moderating their beliefs, and more than that, feeling closer to people from the other side and understanding them better. And then it blew up and then it became too toxic. And then we had to start over again. And we have to try to reimagine things. So that’s happened for a few cycles now.

But I can tell you that the American college students, many of them were totally fine opening up. I found I was incredibly impressed with their ability and willingness to have those conversations and to do this repeatedly. People would come back day in and day out to have these conversations.

I noticed a stark difference when there was an Israeli or Palestinian in the room with these people and when there wasn’t. It’s a totally different texture of a conversation.

JS: So are these all one-on-one or are you now having group conversations?

AB: Both. But it was very difficult because as soon as anybody saw which other conversations are taking place, people would always default to a conversation that was already taking place. If two people were talking, 30 other people would join the platform, and there would be a 32 person conversation. People didn’t, on their own, fragment.

You have to force the fragmentation. Otherwise they all congeal into a single conversation that ends up changing. Its nature changes completely because now it’s no longer an opportunity for two people to be vulnerable. Now it’s a stage and there’s an audience and there’s stakes and the stakes are of performance and it changes the nature of them completely.

JS: I guess if there’s more visibility into what other people are doing or there are comments on the side, then it can become toxic?

AB: Yeah, that’s how it became toxic.

I think that ultimately the trajectory of experience for each person is a function of their motivation. For the people who are very, very good at having these conversations, they don’t necessarily want the one-on-one, because they feel like their voice needs to be broadcasted to more people. They would like there to be a bigger audience. And so you have to balance meeting their needs for having a larger audience versus what actually makes for a productive conversation.

The primary thing that has attracted people is a sense of curiosity. And that’s the thing that we’re leaning hardest into as well. People just want to learn about the space. They want to come and ask questions, or perhaps they want to refine their own intellectual understanding of the space.

Second is oftentimes just loneliness. You know, we have a loneliness epidemic at the same time as we have a polarization crisis. And I’ve always thought that perhaps those two could be solutions to one another. So just actually having a space where you can come and reliably meet somebody else and have conversation about a topic you care about. I think that’s a major driver. And it changes how we design the experience.

The third motivation is just to influence and to change people’s minds. And to prove why the other side is wrong and to correct a misunderstanding that they are perceiving in the discourse.

The next one is a sense of combat. People do like to fight about these things. It’s almost like intellectual jiu-jitsu for a lot of people. I think this is more true of people who are not in the region facing the conflict. There’s an intellectualizing of the conflict. Very different when you speak to Israelis and Palestinians on the ground, it tends to be much less of that.

And then the last thing is entertainment. Some people just want to come and watch a battle. They want to see how these things actually unfold and there’s something fun about this.

So we don’t lean very hard into the latter two, into combat and entertainment. Although we think that it’s necessary to create healthy spaces for those as well. Because some people might come with this energy, but then be redirected towards perhaps like more curious and humble energy. And we think that there’s a need for that type of pipeline. So yeah, depending on what the motivation is, we ended up seeing that very different things were.

JS: How are you monitoring what’s working and not working?

AB: I’m trying to be in many of them, and I often even act as the AI behind the scenes. So sometimes it’s the AI and sometimes it’s me doing various things, perhaps like sending a new prompt to the people.

The medium matters

AB: By the way, everything is audio-video at this point. One thing that we’ve taken away is that text is a radicalizing technology.

JS: Did you experiment with text chats?

AB: We did, and we do because a lot of people prefer text. It’s a very thorny topic for us.

We have drawn a distinction between a text chat and a text channel. A text chat is going on while an audio-video conversation between participants is taking place. Those don’t blow up nearly as much.

But if it’s just text as a channel, where that is the primary mode of interaction and people have a lot of time to get really deep into their own heads and to just download onto the channel a very specific worldview without getting immediate pushback about specific moves, that blows up. And so we have not been able to, I think, create a very healthy text channel space.

And it doesn’t blow up most of the time. 85 % of the time it’s okay. 15 % of the time it’s a disaster.

For long texts, I believe that the AI should be the one generating texts. I can trust the AI to generate texts more so than I can trust the humans to do it. .

What makes a successful conversation?

AB: Now, how do I know that the conversations are going well if I’m not in all of them?

First of all, we see if people stick around, if people want to be reconnected and re-paired with the same people over and over again, because they feel like they’re making progress.

JS: So, some people actually come back and talk with the same people the next day, the next week?

AB: Yes. At one point, we had a college student who had just come back from Auschwitz on a trip, and another American who was Native American and was very sympathetic to the Palestinian cause. And the two of them had fascinating conversations. And they kept messaging me afterwards, “Could you please pair us up together next week?” And then the next week. And then the next week.

AI for conflict resolution

JS: So you built some AI into the platform early on. What did it do?

AB: So I’ve always believed that the AI should be prompting us to get us to go deeper into what we think and why we think.

And so the AI would pick prompts. People might come into the conversation. They would know it’s going to be Israel, Palestine, but they wouldn’t know the specifics. They talk about, let’s say, the wall. And then the AI might give some quotes from Wikipedia and try to see if people engage with them. It might ask them, “What do you guys think about this?”

Next it might put a very different type of screenshot with a very different clause that is very supporting of the wall and get them to then react to that. Perhaps ask them, “How much do you agree from one to 10 about that?” And then it would plot the people’s responses. This is for larger groups where you really want them to get a sense of where the entire crowd is.

JS: Is the AI a conversational participant, or how does that work?

AB: We tried it all. We tried everything. Right now it’s less of that.

I have grown to believe that AI shouldn’t be a moderator of a conversation. If once we have a superhuman AI, I don’t imagine it would be like Oprah. I think it would be more like a Cupid. It would do the pairing better. It would create the environment for the better conversation. It would select the right people at the right time for the right topic better. Not just wait for a thing to blow up and then step in. That’s too late. You shouldn’t have gotten to those octaves and those volume levels in the first place.

I think there’s a Thomas Jefferson quote, “Don’t just tell me which books you read, but in which order you read them.” I think there’s a similar sentiment here. Who you pair up with makes an enormous difference. And the fact that you hear from somebody who disagrees with you is insufficient. I don’t think that’s enough. You really have to engineer the proper conditions for this type of contact to be effective.

And it could very easily become worse. In the early days of social media, they said we were trying to connect everybody. Israelis and Palestinians connect in the West bank and the checkpoints. You could easily replicate the checkpoint at scale. That type of connection and contact isn’t optimal.

For example, we would have some, let’s say American Jews, older gentlemen, who had a very difficult time speaking to Palestinian youth. And the Palestinians would complain, “We can’t talk to this guy. It’s absolutely impossible. And you have to kick him off the platform. They’re really toxic.”

Nevertheless, when I would have conversations with them, those were incredible conversations, some of my favorite conversations. And I realized you really have to be quite precise about who the conversation is between in the first place.

The pace of trust

AB: When people have a hypothesis class about how the world could be, they model the world in a certain way. And when you introduce a radically new concept, it blows up the hypothesis class. And now so many things could be true, that they end up becoming even more entrenched in their perspective. And the us-versus-them increases. It’s shocking.

I would put two people in a room who have — many Israelis had never spoken to a Palestinian in their lives and vice versa. And then they hear the arguments and the narrative of the other side, and they would immediately entrench in their own position and refuse to engage. And they would radicalize. I could see that in my own eyes. Sometimes I feel it, too. I myself feel, you know. I shift to the right when I speak to some people. It’s almost like, now I have to re-examine every node of thought I have about the conflict and it’s too much. And so I shut down.

Alternatively, you have learning. Learning is a very gradual expansion of the hypothesis class in a way that nevertheless makes me feel safe.

I can examine this. If you have somebody who’s on the far right and Israeli politics, they could speak to somebody a little bit to the left of them. Their hypothesis class and their brain grows responsibly. I think in so far as they find themselves in safe settings, they can learn.

I think it begs a question of what the goal is. The goal is not to get people to agree on things. I am not even sure that that’s a desirable goal. As you often say, it’s about transforming conflict from something that is unhealthy to something that is more healthy. So our goal is not to get everybody to agree.

For example, as soon as we scale out and we have people speaking about whether the earth is flat or the earth is round, I’m not sure that they need to find common ground.

What we’re trying to optimize for is the accuracy of theory of mind.

Right now, a lot of Israelis imagine they know what Palestinians think, but they’re wrong about it. And Palestinians think they know what Israelis think and they’re wrong about it. And you published another piece a few weeks ago about Democrats and Republicans or conservatives and liberals in the U.S. They are consistently wrong about what the other side actually thinks.

Misconceptions and meta-perceptions

JS: There’s an academic name for it, which is meta-perceptions: your perceptions of what the other side believes. Let’s get down to brass tacks here. What are the things that Israelis most need to learn and Palestinians most need to learn to improve their models of the other side? What have you actually learned at the object level?

AB: When a rocket comes flying towards you and is chasing you into your shelter, you don’t interpret this violence to be political. You interpret this violence to be hate driven and because of your identity.

I never drove on the highway in Israel, and heard the sirens and had to get out of the car and hide under the car and thought to myself, “I wonder what political points they’re trying to make to me.” That never happened. It’s like, “They hate me for who I am.” That is the default feeling. And that may or may not be true, but when you speak to a lot of Palestinians, they say it’s not true. So that is not how Palestinians understand the nature of their violence.

I think that a lot of people in the Israelis’ positions have interpreted politically-driven violence as identity based. For various reasons, some of them I think are kind of like good faith reasons and some of them are for cynical reasons.

I think it’s easy for politicians to exploit political violence and translate it as identity because then there’s nothing I can do. I’m a Jew, I’m an Israeli, I live here. There’s nothing I can do. If they don’t like it, at some point they’re going to have to make their peace with it, but that’s it.

And then it’s also easy for Israelis to misinterpret Palestinian political violence because oftentimes Palestinians use identity-based language and hate-based language to describe their own violence. And so it’s also true that Hamas often don’t talk about Israelis. They talk about Jews.

And they talk about the need to kill Jews. That’s also true. So there’s also a layer of anti-Semitism that at this point has become rooted in the society. Also for obvious reasons, they’ve been in conflict with Jews for the last hundred years. It makes sense that they at some point would start to dislike Jews. But none of that is shocking.

So, so on one hand, you have cynical players who might be exploiting political based violence and translating it as something that is identity based. So that’s, think the more cynical interpretation. And I think that’s probably true in this case. And in many other cases where people are feeling hate driven violence.

On the other hand, you could also understand why many Jews and Israelis would feel that the violence is hate-based and identity-based because it’s often framed as hate-based and identity-based by the other side. So when Hamas talks about killing Jews, they don’t necessarily say killing Zionists. They often use the word “Jews.”

And so Jews are sensitive to this. When you hear this, they say it’s because of who we are. So there’s also obviously a layer of anti-Semitism that is driving some of this, which again is also understandable given the fact that they’ve been in conflict with the Jews for a hundred years. So it would not be shocking to find some amount of Jew-hatred on that side. But by and large, I think the main motivation of the violence isn’t identity-based, but political.

I’ll give you another example. Yesterday I had a conversation with — so right now we’re a team of Israelis and Palestinians — had a conversation with a Palestinian engineer of ours from the West Bank. And I asked him about his experience with the prisoners being released, and what he feels about the fact that some of these prisoners have obvious blood on their hands? Like, they perpetrated the lynching in Ramallah and many bus bombings and they’re all getting released. And he said, there’s a person from their village that murdered an Israeli woman, but not out of hate.

He just tried to rob her and it didn’t go right. And so he killed her. And now he’s on a bus back to the village near Hebron and the village refuses to accept this man because he killed a woman or a child and it’s unacceptable. The village said, you’re not welcome here. Whether or not it’s true, the fact that this guy believes this to be true is, would shock virtually every Israeli into complete disbelief.

Over the last, whatever, 48 hours I’ve spoken told Israelis about them not accepting a murderer into their village. They [Israelis] didn’t understand it. So again, it’s a misperception and faulty theory of mind.

JS: What is the Palestinian misperception of the Israelis?

AB: I think it’s a more tricky one and it comes down to the experience of colonialism. I think that Palestinians experience Zionism as colonialist in nature. Which is easy to understand why they would experience it that way.

But to the Israelis, that is not the experience. To the Israelis, it is truly a national liberation movement and a incredibly pragmatic and deeply unideological one, at that. It is just an attempt to rescue millions of Jews from the ovens of Auschwitz, or from pogroms in Eastern Europe, or from Arab domination and oppression. And that’s it. And this is their place and it’s always been their place. And in their mind, they’re open to sharing it so long as they themselves experience it safely.

And it’s not about coming and occupying and dispossessing because they think they were there first. The rhetoric that settlers today often evoke is religious-based arguments. Those are not the arguments that have motivated the generations of Zionists who settled the land.

I think there’s this and also the religious versus national component. Palestinians see Judaism as a religion. Jews by and large see Judaism as a people. And it makes all the difference in the world for the Jews. It isn’t that easy to convert into Judaism and Jews don’t try to proselytize.

It makes a big difference for why therefore the Jews don’t think this is just a theocracy. This, in their mind is a safe space that they can call home. Where they can finally be safe because the entire world has time and time again demonstrated its willingness to kill its Jews. So that’s how the Jews and the Israelis see it.

For the Palestinians. Oftentimes it feels like it’s the French in Algeria, or it’s the British in India. And I understand why they experience it this way, but that is not at all how the Israelis experience it.

JS: What do you say to a Palestinian who’s watching [thousands] of people being killed in Gaza? How do you have a conversation that starts to get past that? Many people say, “the only way the path to peace is to share the land in some way.” How do you even begin to have that conversation?

AB: The question that I would ask is “who should begin to have that conversation?”

It might not be the case that I, as a secular Israeli-American, should have that conversation with them. It depends on their disposition and intellectual flavor, and how they have conversations, and what makes for good conversation for them.

When you speak to Israelis and Palestinians, many of them are deeply religious. It could be that the pathway between a religious Israeli and a religious Palestinian is much closer than passing through me. Maybe I’m not the right person here. Maybe my job is to just develop the infrastructure and connect them.

I would ask a lot of questions and I would collect a lot of data about who those people are.

What would I say? I think it’s a fairly simple formula in my mind: The Jews in the Middle East will be safe once the Palestinians are free. And the Palestinians will be free once the Jews are safe. And you can’t have one without the other. And so we hold the key to each other’s future and none of us is going away.

We’re here to stay. The Jews are there to stay. The Palestinians are there to stay. There’s millions of each. They’re not going anywhere. We’re going to have to find a way to make it work. One state, two states, however many states, I think that’s ultimately going to be a function of trust and a function of that meta-perception.

If we think the Palestinians are just going to try to kill us the moment they get the chance, we’re not going to sing Kumbaya and be in one state. But we might, if they take various steps, together increase trust. And as trust increases, I think lots of different solutions show up on the horizon.

So I think the question in my mind is what can Israelis and Palestinians do in order to slowly increase trust so that more solutions can show up?

AI for peacemaking

JS: Are you recording all of the data about who’s in the conversations and are you recording the conversations themselves? Are you running analytics to start to build models to understand when this is going to work?

AB: When we do, it’s very clear. Sometimes we don’t. If people prefer not to, then we don’t. Some things are metadata so they certainly get registered. So who’s in a conversation, regardless of actually recording the audio and video.

In the case of audio and video, we have defaulted to transcribing. We save the transcription, but only temporarily, and then we try to get rid of it. We don’t want to come back and quote people out of context. I don’t say recording, I say we’re learning from it because the recording itself becomes, is discarded.

JS: What are you actually using this training to do? What’s the training objective?

AB: So the goal is a very long-term goal. The goal is to reduce that meta-perception gap.

You can think about it like a policy. So given the environment, there’s various actions that the AI can take. And which actions should it take given that long-term objective of, let’s say, reducing that meta-perception gap.

And there’s different ways that you can model the meta-perception gap. In the literature on contact theory the idea is to increase the probability of a prejudice-drop. Even when you think about a prejudice drop, there’s a lot of different ways that you can think about this. Are they ideologically more aligned or are they just more affectively more aligned? Do they just not hate each other anymore? And you can come up with a bunch of different features. You can just do a lot of feature-engineering to figure out whether they hate each other.

JS: How do you find that out? Do you give them surveys? Do you analyze the transcripts? Where are you actually getting the signals?

AB: Only analyze the transcript. On occasion, we will surface questions to the user, but I try to limit that. I suspect that would be gold standard if we can keep bothering them by asking them to fill out surveys, but I need to think of scale.

People are not coming in because they think that here’s a space for them to have a compassionate conversation with the other side. I cannot give them that impression. If I ask most Israelis and Palestinians, “Do you want to come and have a compassionate conversation with the other side?” They say, “The other side needs to die or get the hell out of here.”

That’s the energy, and we have to lean into that energy. So if I say, “Okay, do you now like them a little bit more?” they’ll say, ‘I don’t like this platform.” So I’m very, very careful about that. Which is also why the platform is called HeadOn. It is more confrontational and it’s okay. It’s the space for safe disagreement.

But in real time, the AI can also ask clarifying questions. Clarifying questions, by the way, is another feature that the AI can distill from the transcript. Are you actually asking them questions or are you just delivering points? Are you asking them questions that would try to demonstrate that you’re interested in how they’re doing it or not?

I dug into a couple of textbooks on conversation analytics. And I got inspired by some of those, like the question to statement ratio. And all of these things are things that AI can do decently well just from transcript.

JS: So, it sounds like you analyze the transcripts with some sort of classifier or estimator which tries to figure out what these people think of each other, and then use that as a long-horizon training goal.

AB: That’s the idea.

And the easier way to do it now is to build a very large compatibility matrix. You could even just ask the AI to be a judge. How good was the interaction between them towards reducing their bias and increasing their trust? Give it from a one to five or whatever. And then you have a big matrix of person by person. You can also now start to overlay topics. So then you have like a person by person by topic.

So now when two people show up to the platform, roughly at the same time, the AI can say, “These two people should have a conversation about this topic. I have even cached it.” You could do this offline. So this person and this person and that topic is likely going to be a very good conversation.

So now you show up to the platform, boom. I drive from cache and I show it from the short-term memory and I can show you the opportunity to participate in that conversation.

JS: So it’s kind of like a conversational recommender system with participants on a topic. Do you actually have something like this working?

AB: So the one that I just said with the compatibility matrix, that one we’ve been doing that for like 10 days. So almost done building the v.0 of that.

Another thing that we’ve done is a cluster analysis on topics, which proves fascinating also. What topic tends to follow another topic? When they’re speaking about this, are they likely to talk about that? If they’re likely to talk, what is the semantic evolution? If they’re likely to go from A to B, but then in B, they’re likely to fight over B, maybe we should wait on that or maybe we should pull in another person. And if that other person is in the room, we think that the probability would be reduced of it escalating.

So you could play all sorts of games like that once you actually model and simulate how the conversations are likely to unfold.

Inspirations

JS: Who inspires you in the conflict resolution space? Who are you most excited about learning from and working with?

AB: There’s a lot of good people in this space. Many of them I found through you and your research. Again, the person that I have seen is absolutely phenomenal in having conversations about this — to be honest, I have not found a single person who has better conversations about Israel Palestine — is a man I found on YouTube. His name is Adar Weinreb.

He started a YouTube channel many years ago, called The Great Debate. And now it’s just Adar Weinreb. I reached out to him immediately and asked him, “How come you don’t have millions of views yet? And how do I support you?”

And now he’s my co-founder in this project. He’s able to deploy a deep compassion and a capacious feeling of both sides with an erudition about the conflict itself. And I had not seen anything like this yet, to be honest. So he goes to the West Bank now on a monthly basis and brings Palestinians and Israelis together.

Ali Abou Awad in the West Bank, this has been absolutely fascinating to listen to as well. He came out of the Israeli prison, I think, many years ago. And he started a couple of organizations. One of them is called Roots in the West Bank.

Hearing him speak, it sounds like he actually understands how Israelis think. One time we recorded a 60 second clip of him speaking and posted to Instagram. I thought to myself “That will work on Israelis to open their mind about what Palestinians desire.” And as soon as we put it online, you actually saw a bunch of Israelis saying, “Never considered that perspective before. Fascinating.” And it’s just because he had that theory of mind.

JS: I have to say this is one of the meatier podcasts we’ve done. It’s rare to meet someone who’s really in the trenches trying to figure this stuff out. And at the same time, thinking about all of these advanced AI possibilities. So it’s really been a lot of fun. And I know we’ll have more conversations in the future.

AB: Jonathan, thank you very much for doing all the work that you’ve been doing. A lot of this, as I said, has been inspired by your work and your writing. And I look forward to reading more and thank you for having me on.

JS: Awesome, I’ll see you next time.

No matter who you voted for, I think you will agree that things are going badly for America. After a lot of reading and many conversations, this post lays out my best guess to get out of this mess. I’m featuring three recent articles that, together, show the way forward.

But first, let’s talk politics for a minute.

We are losing important things

We welcome all comers at BCB, whatever your ideology. This means that many of you are sure that the Trump administration is the problem, and it also means that many of you are sure that things would be OK if only the Democrats weren’t so idiotic. Personally, I think this is the hour where the Libertarians are an important third side.

Wherever you stand, in very concrete ways the government is failing and Americans are losing freedoms we have taken for granted for centuries.

As always when an issue is sufficiently polarized, getting into the details is either redundant (for those who agree) or unconvincing (for those who don’t). So instead I’ll let Isaac Saul at Tangle make the case:

So here’s the truth as I see it: If, on the day Trump was inaugurated, I had warned readers that in a few months law enforcement officers would be rappelling from helicopters like soldiers into civilian apartment buildings in Chicago; the military would be extrajudicially killing Venezuelans for alleged drug dealing; Americans would be getting arrested while being falsely accused of being here illegally; the Justice Department would be prosecuting the president’s political foes at his direction; and legal U.S. residents would be getting arrested, detained, and deported for protected speech; I would have been accused of having a bad case of Trump Derangement Syndrome.

Yet nine months into this presidency, all of those things are happening. This isn’t hyperbolic fear-mongering — it isn’t sensationalist, or exaggerated. It’s literally just a list of the things the president has done, things I’m obligated to speak plainly about.

…
At the risk of doing too much hedging and throat-clearing, I want to make myself perfectly clear: None of this is to suggest there aren’t problems on the left, or that Democrats haven’t grossly abused executive power or acted in unconstitutional ways. Of course they have.

On top of this, the Trump administration has not delivered. Inflation is up. The government is shut down and 42 million people are about to go hungry.

If we can agree that we need to protect our basic rights and have a functional government, then I don’t care what your politics are — we can work together. But we need a plan. Here are three recent articles that capture the core of a coherent strategy.

The relationship is the problem

Guy and Heidi Burgess recently wrote a masterful argument for conflict transformation. In The Great Reframing, they make the case that we need to shift our thinking from “they are the problem” to “our relationship is the problem.”

The change we suggest is this: Instead of assuming that our big social and political conflicts are between us (”the good guys”) and them (the “the bad guys”), the Great Reframing defines the primary problem as the destructive ways we engage with each other about all those other things. It is the act of assuming “the other is to blame.” It is the act of humiliating the other, trying to delegitimize, disempower, and disenfranchise “the other.” Those actions, which often cause parallel toxic reactions on the other side, constitute the really big problem that is driving us further and further apart, and is preventing us from making any progress on the issues we care most about (things like the economy, immigration, climate, etc.)

None of the problems we face are as simple as us-vs-them, good-vs-bad. By assuming they are, we are not only making our problems worse, we are crippling our ability to address those problems successfully.

This is foundational to just about any strategy. This sort of relationship building is how coalitions get built. It’s how we correct the misperceptions that cause us to take bad actions. Neglect or outright vilification of this relationship is how things got so bad in the first place.

The Democrats need moderating reforms

This is another one of those claims that is going to be met with a combination of “duh” and “absolutely not.” Everybody agrees the Democratic party is broken, but there is passionate disagreement on how to fix it, perhaps most passionately among Democratic voters. To brutally oversimplify, the sides could be called “fight harder” and “choose more popular policies.”

Last week, the New York Times did an analysis showing that more moderate politicians — on either side — win elections more often. This week, a massive, data-driven report dropped, making the same case in exhaustive detail. The authors called it Deciding to Win:

We draw on thousands of election results, hundreds of public polls and academic papers, dozens of case studies, and surveys of more than 500,000 voters we conducted since the 2024 election. Deciding to Win argues that since 2012, highly educated staffers, donors, advocacy groups, pundits, and elected officials have reshaped the Democratic Party’s agenda, decreasing our party’s focus on the economic issues that are the top concerns of the American people. These same forces have pushed our party to adopt unpopular positions on a number of issues that are important to voters, including immigration and public safety. To win again, Democrats need to listen more to voters and less to out-of-touch donors, detached party elites, and Democratic politicians who consistently underperform the top of the ticket.

There are lots and lots of charts in this report. Here’s a good one:

Non-violence is how an opposition movement wins

Here at BCB we’ve repeatedly discussed non-violent conflict. Even if you don’t agree with the goals of the protesters, it’s morally defensible. Also, it tends to work if enough people are involved. So I was fascinated to see the readout from the latest Nonviolent Action Lab Summit. They identified five strategic principles. I think the first one is most relevant:

1. Fragmentation and Coalition Building of Pro-Democracy actors

A fundamental challenge for counter-authoritarian movements is that democratic forces are often highly fragmented. Democracy’s allies can be divided across borders, by socioeconomic or political cleavages, between civil society and formal political actors, and over competing visions for the future. If these divisions are acrimonious, they can impede effective coordination around resistance to authoritarianism, making it far easier for autocrats to expand and consolidate control. Autocrats have long employed divide-and-rule strategies, and for good reason—fragmentation and division are the bane of collective action.

The most urgent priority for democratic revival, then, is pro-democracy coalition building. To build power, pro-democracy actors must foster strategic alliances across civil society, political actors, the private sector, religious institutions, and more. In some cases, pro-democracy actors may resist collaboration with potential allies with whom they disagree on other issues or have clashed in the past. But coalition building does not require unanimity; it requires an articulation of shared goals and a willingness to adapt to the circumstances at hand. Helping democracy movements to cultivate coalitions and coalition-building skills will be crucial in the years ahead.

The other four priority areas are:

• Tactical Stagnation. “While mass protest remains a vital tool, many democracy movements have been trapped in cycles of mobilization without ever achieving durable change.”

• Authoritarian Legitimacy. “Wannabe autocrats have learned to use outsider populist appeals to discredit existing democratic institutions as corrupt and impotent, cloaking their attacks on democracy under veils of democratic reform.”

• Global Autocratic Alliances vs. Weak Movement Support Ecosystems.

• Disinformation and Narrative. “Activists must continually innovate in today’s digital information wars, developing intentional strategies to garner and retain media attention.”

A theory of the case

I believe that all evidence suggests we are witnessing the destruction of functioning government and basic rights. The most fundamental strategy is to expand the pro-freedom coalition by uniting “us” with “them” and taking non-violent action together.

And even if you disagree — if you think Trump and Republicans are making necessary changes — I hope that it’s hard to disagree with the idea of having a better relationship with the other side, reforming the broken Democratic party, and committing to non-violence.

But nobody will be satisfied with everything above. So let’s talk. Tell me what this gets wrong in the comments.

Quote of the Week

Every time you humanize an opponent, you make violence less likely.

-Builders Movement

One of the questions most intensely in the air on the Blue side right now is how to respond to Trump. To protest the deployment of National Guard troops, Portland decided to call an emergency naked bicycle ride. And then 1,000 people actually did it.

Many other people wondered why. My professional opinion is that this is less silly than it seems, so I’m now going to methodically ruin the fun by analyzing this protest tactic.

The strategy here is to capture media attention to humorously demonstrate that Portland is not a “war ravaged city” as Trump said. Before, we had pictures of people in black yelling at stoic Federal agents. Now we get the absurdity of men with guns standing over naked grandmothers.

And while non-violent conflict is not necessarily non-coercive, this tactic is harmless which makes it more broadly acceptable. To get all nerdy, you could say it’s aimed at the largest segment of the power strategy mix. It mocks power without threatening anyone. And it’s a classic. “Public disrobing” is what Gene Sharp called protesting naked in his 1973 list of 198 non-violent conflict tactics.

Or that’s the theory. Did it actually change the narrative? MediaCloud data says the narrative did change. “Portland and violence” was much more common than “Portland and naked” up until the day before the ride. Then “violence” fell and was actually less common than “naked” for a few days, and remained at a lower level for the next week.

Also, the story was more common in Red media sources than Blue sources, so it succeeded in reaching people who didn’t already agree with the idea that there’s something wrong with National Guard deployments.

And what did they think of it? Well, some thought it was disgraceful, such as Speaker of the House Mike Johnson:

“What I’ve seen is the abuse of law enforcement by radical leftist activists,” Johnson said. “Most recently, the most threatening thing I’ve seen is the naked bicyclers in Portland who were protesting ICE down there. I mean, it’s getting really ugly.”

Elected officials standing behind Johnson appeared to smile and laugh at the remark; Johnson did not.

But I’m willing to bet that many more people were smiling than not. In the videos it’s clear which side is having more fun. “I don’t want your revolution if I can’t dance,” as the saying goes.

What’s actually happening in Portland though?

And yet it’s so easy to get caught up in arguments about narrative that one can forget to ask about the facts on the ground. Conflict is like that. And unfortunately it takes a modest amount of effort to figure out what’s been happening in Portland, which I blame partly on the fact that news reports are incremental and partly on the fact that many sources have something to prove. (Previously: What we don’t know because of polarization.) So here’s my attempt at a brief straight-news summary.

Since June, there have been nightly protests outside ICE facilities that sometimes involve violence. Protesters have been relatively active but constrained to small areas of the city, according to the BBC who interviewed the neighbors:

Gray’s Landing resident Cindy Colgrove, 63, says it has been “115 days of hell” since protests against Trump intensified outside the ICE facility in June.

“I only come out during the day,” says Ms Colgrove while walking her neighbour’s dog near a small group of peaceful protesters.

“You see all the black-covered Antifa people aren’t here. They come with the night. In the daytime it’s all these little old senior people that think they’re changing the world before they pass from this earth.”

According to the most comprehensive timeline I could find, The crowd seems to normally be in the range of dozens of protestors, who have vandalized Federal buildings and lit things on fire and “rioted” on several nights. Federal agents have sometimes responded with flash grenades and tear gas. The Portland FBI says 198 arrests have been made since June.

Apparently neighborhood residents “disagree over who is to blame for the violence.”

The video below has lots of protest footage. It reports mostly from the Federal Protective Service side, which is interesting and seems somewhat less common than the many other (mostly amateur) videos reporting from the protester side.

One of the officers says, when asked what’s been happening,

Number one, a lot of First Amendment-protected activity. And then occasionally we’ll have where it will not be peaceful, and we’ll start having illegal activity.

I could not find reliable reports of the approximate number of protesters each night this year, but it’s worth noting that 40,000-50,000 marched in each of the two big No Kings protests in June and October. With 198 arrests during the same period, that means something like 0.2% of all protesters have been violent.

There is also, however, a continuing herd of inflatable animals, another strategic response by Portland’s very serious protesters.

Image of the Week

via Holly Anderson