(I began writing this post several weeks ago, but political events are moving much faster than I expected, so I am publishing now out of fear that otherwise the message will arrive too late to have an impact.)

I

In this post I want to explain a concept, and issue a warning based on it. But I expect the warning will be superfluous if my explanation is sufficient. If you want to convey the idea "the rattlesnake has venom in its fangs, so don't let it bite you", you won't need a hard sell for the concluding advice if the listener understands the initial statement about venom.

The word for the concept I want to illustrate is partisanize, which means to align an issue with a political tribe. It is modeled on politicize, but the latter word is not useful here. It would be meaningless to say "Don't Politicize AI Safety": the project is intrinsically political. It involves international diplomacy, consensus-building, the willingness to sacrifice near-term economic growth for long-term human values, and a brutally difficult coordination problem. AI Safety is inescapably political, but not inevitably partisan. It's possible that, like issues such as infrastructure or wilderness conservation, the topic will remain nonpartisan. Unfortunately, there are many issues that started out neutral and were partisanized via a process of tribal reasoning. As I explain below, I fear AI Safety is at grave risk of following the same trajectory.

I was motivated to write this essay by a recent stratospherically-upvoted LessWrong post entitled Why I Left Google Deepmind. The author, Dr Alex Turner, documents his extensive political efforts to convince Google leadership to adopt a more principled stance on issues related to AI Safety. Unfortunately, these efforts were unsuccessful, leading to Turner's resignation from his lucrative and prestigious job at GDM. I concur with the community's enthusiastic admiration for Turner's willingness to make serious personal sacrifice for his beliefs. However, I also believe his post might actually harm the cause of AI Safety, by linking it to Blue tribe politics and thus partisanizing it.

A Red tribe citizen who read only a high-level summary of Turner's story would almost certainly be strongly supportive of his sacrifice. The Red tribe is generally skeptical and hostile to Big Tech companies, which are widely perceived as extremely Woke (cynical observers claim that Big Tech adopted Wokeness as a way to distract the Left from the vast amounts of wealth inequality they created). Red intellectuals are aware of and enraged by Big Tech's willing participation in the Woke censorship regime of the late 2010s. The American public is increasingly suspicious of AI and there is no reason to think this suspicion has a partisan skew. In general, conservatives by their very nature are resistant to change, and AI will obviously bring about enormous social upheaval. A conservative utopia has safe streets, high employment, strong familes, cohesive communities, and widespread religious participation. There is little reason to believe that AI will promote these goals, and many reasons to suspect it will be toxic to them.

Unfortunately, a conservative who read the actual post (as opposed to a summary) would almost certainly respond negatively, and this reaction would likely contaminate his opinion of AI-Safety activism in general. The reason is that Turner opens the post with a discussion of an extremely partisan topic: the killings of ICE protestors Renee Good and Alex Pretti. His initial activism was not about AI at all, but rather about cloud services: he wanted Google to refuse to work with the federal agencies who are responsible for immigration enforcement (ICE and DHS). While anti-Trump rhetoric is not the main point of his essay, in some places he uses accusatory language such as "federal agents should not be able to kill citizens in the street"—as if the agents had just walked up to two hapless bystanders and executed them in cold blood.

Most Red tribe members find this kind of emotional, logically unexamined accusations to be deeply offensive (though they of course produce similarly illogical retaliatory accusations). While I don't want to litigate Culture War topics here, I do want to convey the Red viewpoint on this issue. Red citizens see ICE as fulfilling a democratically decided mandate to enforce immigration law — a mandate that the Biden administration, in a brazen betrayal of democratic principles, simply ignored. All types of law enforcement are intrinsically dangerous, because some people simply will not follow the law unless the threat of violence is present. Finally, conservatives champion the right of all citizens to use lethal force in self-defense. So, in the Red view of the story, the ICE officers were doing their difficult and dangerous job when they were threatened by Good and Pretti. The officers needed to make a split-second judgment call with their lives on the line; they choose to respond with lethal force. This outcome is regrettable, but it hardly justifies the kind of moralistic insults and accusations thrown around by the Left. Certainly it does not justify Turner's original crusade to convince Google to stop providing cloud services to the federal government.

II

Noted LessWrong contributor, AI researcher, and civilizational theorist Richard Ngo recently posted on the LW shortform about a job offer that had been rescinded:

I had a long conversation with Jesse and Dan, and later a long conversation with Geoffrey. In the latter I learned that, although the decision was prompted by me sending them my curriculum, Geoffrey in particular was more concerned about my tweets. I've tweeted some fairly controversial things over the last few years... It seems like he's excited about hiring academics who are new to alignment, and worried that those people might not join if they hear about such tweets... Geoffrey claimed that he would have made the same decision if I'd posted equivalent left-wing takes.

Ngo reports that one of the Resolution executives cited this tweet as being particularly objectionable. I'll quote just a section, which occurs to me as plausibly the most inflammatory:

The tech left views (the country) as a charity. To them, wanting others not to receive what you’re been given is hypocritical. This view is also empty, because it doesn’t tell you where the windfall actually comes from—and once you start to talk about the benefits of culture, ethics, institutions, etc, it becomes clear that citizens (and their ancestors) built that windfall rather than just being given it.

For someone on the tech left, whose alleged viewpoint was being critiqued, I can see how this comment has a lot to complain about. Ngo suggests that the citizens of a prosperous country deserve their windfall of wealth, education, safety, healthcare, etc, because the conditions that produced those goods were created by themselves or their ancestors. But it's self-evident, from a brief review of any social media app, that modern societies have millions of people who don't contribute very much. And given the ongoing debate about the ethics of inheritance, it's not clear why modern people should reap such enormous benefits from their ancestors' efforts.

While it's completely valid to argue against Ngo's tweet, the point of mentioning it is to highlight that he cannot possibly be considered so far out-of-bounds politically as to make rescinding a job offer a reasonable response. He did not call for political violence or hurl profane insults at anyone. I see dozens or hundreds of far more offensive comments by Americans of all political persuasions during my daily surf of X.com and Facebook.

III

One of the core insights of the rationality movement is that people are far more inclined to reason based on images and associations than real logic and argumentation. Suppose a person is faced with the question of whether to support policy X. A naïvely optimistic view of human nature is to suppose that the person will find arguments and evidence that bear on the question of whether policy X is helpful, and decide based on weighing this evidence.

The vast majority of humans simply does not follow the rational approach. Instead, they look around to see what other members of their tribe are saying, and adopt that opinion as their own. This process of observation and imitation happens continually and rapidly, since new social questions emerge all the time. It is important from a social-strategy perspective to avoid making a faux pas by expressing approval of an idea the tribe has already decided to condemn, or vice versa.

Since the observation-imitation process happens so rapidly, people have learned to take shortcuts. A full assessment of another individual's tribal loyalties is usually too time-consuming to perform regularly. Fortunately, there are a wide variety of superficial characteristics that can be used to infer this information. Hair color, sexuality, hobbies, media preferences, fitness habits, career choices, and education levels all correlate significantly with tribal affiliation. While each individual correlation might be weak, you can produce a quite powerful prediction by combining several of them together (this is the idea of the classic ML technique called boosting). This works quite well in practice: if you meet a blue-haired librarian in Berkeley, you can be quite confident that she did not vote for Trump.

So normies are constantly observing people at a superficial level, gauging their tribal affiliation, and using the information gained thereby to guide their thinking on sociopolitical issues. On some technical topics, we might trust someone who is tribally misaligned: the blue-haired librarian might trust a Texan redneck to fix her car. But for politically impactful issues, we are going to trust people who look like us and act like us.

Given this fact, suppose a normal conservative is shown the following image, and told that the people in the image support policy X. How will their opinion about policy X shift as a result of viewing the image?

At this point, I want to invite rationalist, Blue-aligned readers who are concerned with AI Safety to take a step back and view the material I've mentioned above through the lens of tribal reasoning. Think about how a typical conservative, who has no understanding of AI Safety or the culture that surrounds it, might respond to these messages. Here's my take:

1. (TurnTrout post) : the author is deeply Woke and also concerned with AI safety. Woke people hate me, so fuck AI Safety.
2. (Ngo job offer rescission) : AI Safety orgs are so deeply Woke that they would take back a job offer because of an extremely unspicy conservative tweet. Since I'm much more conservative that Ngo, AI Safety people must hate me too, so fuck AI Safety.
3. (Aella photo) : AI Safety people are weird Berkeley sex orgy cultists. I don't want to be around those people, so fuck AI Safety.

To be very clear, I don't endorse this reasoning process. I respect TurnTrout's willingness to make sacrifices for what he believes; I don't have a problem with weird Berkeley sex practices; I'm sure Ngo will find an excellent position that makes wonderful use of his talents. I'm simply telling you how right-leaning normies will respond, because of human tribal reasoning-by-association.

IV

I'm part of the unfortunate generation whose entire adult life has been dominated by the Culture War (the Duke Lacrosse rape hoax of 2007 is a good upper bound for the starting point). The conflict has cost me several friends, and severely weakened my relationship with my family. I've read enough rationalist material to understand that my tribal enemies are not actually bad people as a whole, but I also know that many individuals despise me for being a straight white dude with generally libertarian beliefs. These conflicting intellectual pressures induced me to spend many years contemplating the philosophical aspects of the war. The deepest conviction produced by this long thinking is that the closest historical analogy to the American Culture War is the European Wars of Religion that laid waste to much of that continent in the 16th and early 17th centuries.

Here's a quick overview of the similarities (see my essay the Hatred of the Believers for a more extensive discussion). Society in both eras is held together by a unifying ideology (Christianity / liberal democracy), which both sides affirm, and which is used as a justification for the conquest of foreign nations. The ideology is preached, interpreted and defended by a moralizing class of highly credentialled professionals (priests & monks / journalists & university professors). The cause of the conflict is corruption, real and perceived, of the moralizing class (simony / DEI). The attacking side propounds a new doctrine which undermines the power and prestige of the moralizing class (Sola Scriptura / Do Your Own Research). Because people believe that the ideology protects them from unspeakable horror, disagreements about precise technical components of the ideology (transubstantiation / transgender rights) cause bitter resentment and hatred, which ironically causes people to abandon the direct stated commandment of the ideology (brotherly love / civic tolerance of dissent). The conflict is accelerated by a new information technology that the establishment cannot completely control (printing press / social media). The attacking side deploys a full spectrum of criticism, ranging from highbrow intellectual argumentation to literal shitposting. The defending side attempts to sideline its critics by shutting them out of polite society (excommunication / cancellation).

If you accept that "religious conflict" is the best reference class to use for forecasting consequences of the Culture War, there are several important implications. First, there is no rational sense in which one tribe is right and the other is wrong. No serious modern intellectual would try to argue that the Catholics were right and the Protestants were wrong, or vice versa. Pundits condemn each other as traitors or heretics against the ideology, but the precise content of the ideology is exactly the topic of debate. The sale of an indulgence is either a blasphemy against God or a perfectly acceptable means for the Church to raise funds for its good work in His service. Partisans who recognize this philosophical difficulty also advance arguments grounded in objective (non-ideological) facts. For example, a medieval Protestant might claim that his creed is superior, because Protestants had a far higher literacy rate. Similarly, modern Blue often love to observe that their side is better-educated and wealthier, and thus Blue policies lead to a healthier society. But it is simply a category error to draw far-reaching ethical conclusions on the basis of these modest correlations.

The second implication of the historical analogy is that the Culture War is going to continue for many years. The European Wars of Religion lasted for roughly 120 years (the 95 theses appeared in 1517 and the Peace of Westphalia was signed in 1648). In the wake of the 2024 election, many observers argued that Wokeness had been defeated. This is a dangerous illusion. While it is true that Trump's return was a setback for the Woke Left, it was simply a single battle in a multi-decade struggle. The core questions of the conflict are not close to being settled. The recent success of so-called democratic socialists such as Zohran Mamdani and Darializa Avila Chevalier and the fact that (as of August 2026) Alexandria Ocasio-Cortez is the prediction market favorite for the Democratic presidential nominee in 2028, are evidence that the Left has no plans to retreat from its more extreme positions in the face of Republican electoral wins (of course the Maga faction is similarly uninterested in ceding ground). Neither side has any interest in diplomacy; both sides view the only acceptable outcome as the unconditional surrender of their enemies.

The third implication is that, while the Culture War does raise serious ethical concerns, these are utterly overshadowed by petty religious hatred. History is replete with examples of religious strife: Protestants vs Catholics in Europe, Sunni vs Shia in Islam, Hindu vs Muslim in India. From an outside view, it is clear that none of these were grand ethical struggles of good against evil. However, it is extremely difficult for someone involved in such a conflict to realize this fact. In the war of Protestants against Catholics, one of the bitterest points of contention was the principle of Sola Scriptura. Very few people today even know what this principle means, and an even smaller minority would be willing to die for it. If it were not for the particular historical accident of AGI, historians of the future would likely view the Culture War with little interest. From their future perspective, it would be just another religious and tribal conflict, where ambitious and greedy leaders manipulated the masses into hating each other for their personal political gain. Robin Hanson correctly argues that, because of cultural drift, future people will have little interest in our modern debates about transgender rights or the Second Amendment: they'll be too busy debating their own futuristic moral conundrums.

Unfortunately, the Culture War will, in fact, have vast historical impact, not due to its innate intellectual significance, but because almost all the work of AGI development will be done within its spatiotemporal extent. This purely random historical contingency poses a dire threat to the future of humanity.

To see the danger of this accident, suppose we lived in a alternate timeline where science had made faster progress earlier in European history, so that AGI development overlapped with the Wars of Religion. Furthermore, suppose that the geographical epicenter of AGI research occurred in an fervently Catholic area, such as Rome or Paris. In this timeline, it is inevitable that the Catholic church would place enormous pressure on researchers to ensure that AGI promoted good Catholic teachings. For example, the Church might require that AGI tools must refuse to answer questions about the Bible for fear that Protestants could use the technology to learn Christian truth without depending on the priesthood. More concretely, the Wars of Religion would likely derail any attempt at political agreement to ensure that AGI development proceeds at a cautious pace. It is hard to coordinate with your tribal enemies after they've starved your cities to death or viciously insulted your spiritual leader, even if you happen to agree with them on some technical topics. The conflict might also precipitate an AI arms race, with the Catholics insisting on rapid development of the technology to ensure that it did not fall into the hands of the foul Protestants.

V

I want to persuade you of the following advice:

If you care deeply about a issue, don't let it become partisanized

An issue has become partisanized when it begins to be used by one tribe to attack the other, especially if those attacks are mindless or shallow. Some examples of partisan wedges that have arisen in the last couple of decades are climate change, #MeToo, Black Lives Matter, illegal immigration, transgender rights, and vaccine mandates. Naïve advocates might imagine that partisanization is good for their cause, because it brings the topic into public consciousness. But I suggest the opposite is true: the net effect of partisanization is negative even from the perspective of people who care about the issue. Climate change is a good example of this. This issue always had a partisan skew, due to the Red tribe's enthusiasm for masculine industrial projects like resource extraction. But this skew was not strong enough to prevent Republican leaders such as John McCain, Lindsey Graham and Olympia Snowe from supporting legislation to reduce greenhouse gas emissions through a "cap-and-trade" approach in the 2000s. Unfortunately, as the Culture War picked up steam in the 2010s, and climate issues became overwhelmingly Left-coded, hope for a bipartisan action has vanished.

A crucial fact about the wedge issues mentioned above is that most of them have little intrinsic partisan valence. Consider vaccine mandates. Trump rightly boasted about the development of the COVID-19 vaccine as one of the great accomplishments of his administration. Knowledgeable commentators generally agree that Operation Warp Speed was successful in accelerating the development of the vaccine, likely saving hundreds of thousands of lives. Furthermore, Big Pharma is a perfect villain for the Left-wing model of the world in which corporate plutocrats enrich themselves by bribing government officials to help them loot the public purse. So there is no reason a priori to suspect that vaccine mandates would ultimately become a Left-coded issue. The partisanization of this issue cost many lives, by pushing the Red tribe towards vaccine skepticism. In the eyes of the Blue tribe, these deaths were widely seen as poetic justice, but those few who care more about public health than tribal politics might view the outcome as a tragedy.

Immigration is another issue that has no innate partisan valence, and only recently became a wedge. In 1996, the official platform of the Democratic Party stated explicitly that "We cannot tolerate illegal immigration and we must stop it". As recently as 2015, Bernie Sanders mocked open borders as a "Koch Brother proposal" in a Vox interview with Ezra Klein. In the traditional Democratic outlook, immigration is undesirable because it undermines the bargaining power of labor relative to capital. Cultural and ethnic affinity creates social cohesion, which increases support for progressive social policies such as universal health care. Immigrants from traditional societies are less likely to support progressive thinking on homosexuality and gender equality. So the fact that immigration became a partisan issue was not at all inevitable. But now that it has become partisanized, we have the worst immigration policy imaginable: ping-ponging from a complete lack of enforcement under Biden to harsh crackdowns under Trump.

How can sincere advocates avoid allowing their issues to become partisanized? I believe the secret is to practice partisan hygiene, which has two aspects. First, more obviously, advocates should ensure that their public image contains no overt expressions of tribal affiliation. As an example, Second Amendment defenders should avoid confirming the stereotype of gun owners as rural white Christians who listen to country music and talk radio, even if that stereotype actually describes them. Second, advocates should be disciplined about refusing to acknowledge the partisan skew of an issue even when it is present. It may be that many supporters of policy X also agree with policy Y, but advocates for X should ignore that correlation — even if they also support Y! For an example of how to fail at partisan hygiene, consider the following statement from Al Gore's web page:

Entrenched, systemic racism in our country has led to disproportionate impacts of pollution on communities of color... The need for climate action is bound together with the struggle for racial equality and liberation.

This statement lends support to the Red suspicion that climate change activism is really a Trojan horse for far more wide-ranging progressive and socialist reforms.

One very good rationalist reason for refusing to see partisan valence of an issue you care about is that your own tribal loyalties may cause to you overestimate it. If you believe strongly that Policy X is good and true, and are a loyal member of Tribe T, you are likely to seriously overestimate the correlation between support for X and membership in T. Also, your lack of understanding of Tribe S may cause you to overlook significant enthusiasm for X in that population, or fail to realize that they have little incentive to support Policy Y. I believe the Blue tribe fell victim to this illusion during the intense debates about climate change in the 2010s. Blue saw Red as being (recklessly) pro-industry, and thus opposed to any policy that might hurt the energy companies, especially the large and profitable Big Oil sector. But, in fact, the most direct effect of stricter carbon policy would simply be to reduce the use of coal, because coal burning has the worst ratio of energy output to carbon emission. This reduction might actually help the oil companies, which are really oil-and-gas companies, since those two products are cleaner fuels than coal.

As I mentioned above, the AI safety issue is like immigration and vaccine mandates: it has no innate partisan valence. Conservatives, by default, will be very happy to support policies to slow or pause AI development... unless those policies become Left-coded. People who are passionate about AI safety should obsess over partisan hygiene and scrupulously avoid even the appearance of tribal affiliation.

As a closing thought, it's already a partisan-hygiene issue that Berkeley is the epicenter of AI Safety work. I would strongly encourage new research groups to locate in more "purple" regions such as Colorado or New Hampshire. Just as it's unwise to underestimate how simple images and associations can undermine strong intellectual arguments, don't overlook how cheesy gimmicks can help a cause (the Trump-as-a-McDonalds-worker photo is a great example of this). I suggest someone buys Eliezer a Ford F-150 and a cowboy hat, and organizes a road trip where he drives around to remote rural towns to spread the word of AI safety. This kind of stunt could become an iconic moment in American cultural history, comparable to Ken Kesey's cross-country bus trip with the Merry Pranksters.