Reflections on the largest AI safety protest in US history
On a sunny Saturday afternoon two weeks ago, I was sitting in Dolores park, watching a man get turned into a cake. It was, I gather, his birthday and for reasons (Maybe something to do with Scandanavia?) his friends had decided to celebrate by taping him to a tree and dousing him with all manner of liquids and powders. At the end, confetti flew everywhere. It was hard not to notice, and hard not to watch. Something about the vibe was inspiring… I felt like maybe we should be doing something like that. I was there celebrating with another fifty or so people from the Stop the AI Race protest march we had just completed, along with another hundred or so others. 1 We were marching, chanting, etc. to tell the AI company CEOs to say the obvious thing they should be shouting from the rooftops: “A
On a sunny Saturday afternoon two weeks ago, I was sitting in Dolores park, watching a man get turned into a cake. It was, I gather, his birthday and for reasons (Maybe something to do with Scandanavia?) his friends had decided to celebrate by taping him to a tree and dousing him with all manner of liquids and powders. At the end, confetti flew everywhere. It was hard not to notice, and hard not to watch.
Something about the vibe was inspiring… I felt like maybe we should be doing something like that. I was there celebrating with another fifty or so people from the Stop the AI Race protest march we had just completed, along with another hundred or so others.1 We were marching, chanting, etc. to tell the AI company CEOs to say the obvious thing they should be shouting from the rooftops: “AI is moving too fast! We want to stop! If governments can solve the coordination problem we are SO THERE!”
It was a good time. Everyone involved seemed to think it went well and that it felt good to be a part of. It got media attention, there were some great photos, and videos, and speeches. Big props to Michael Trazzi and the other organizers.
Berkeley statistics professor Will Fithian’s speech was the stand-out. He’d just come from his son’s birthday party, and was visibly moved talking about the prospect of his children not having a future, and imagining telling his son years later about the grown-ups who came out to protest so that he (the son) would get a chance to grow up himself. It was heart-wrenching.
Confronting the reality that AI could kill us all, and yet people just keep cheerily building it, brings up a lot of emotions. They can be overwhelming. A lot of people end up shutting their feelings out and treat AI risk as an abstract intellectual exercise, or with gallows humor. It’s a problem because the emotional reality is so important to staying grounded and to communicating with people who haven’t considered the issue before. It’s such a terrifying, horrifying, sickening, appalling state of affairs. It’s really hard to grapple with. And then you don’t just want people to give up, either...
I spent a good chunk of time preparing my own speech, which I actually wrote in advance (I can only recall two other times I’ve done that).2 My speech was about refusing to accept the unacceptable, and the lie of AI inevitability. I was a bit thrown off because a homeless man was shouting disruptively at the outset, but I think it still turned out pretty well, you can take a look and let me know what you think.It seemed like the protesters were mostly people who think AI is quite likely to go rogue and kill everyone; the “If Anyone Builds It, Everyone Dies” type of crowd. It’s actually impressive that we managed to turn these people out, since they’ve mostly not turned up for previous protests, e.g. run by PauseAI. I’m not sure what made the difference here, probably timing and branding both played a role.What gets people to come to a protest? For all of the work people put in flyering and promoting the action, it seems like virtually everyone was there because they had a personal connection to someone else who was attending. Getting people to turn out feels a bit like getting people to come see your band play at the local bar. You’re just not going to get many random people showing up because they think your posters look cool. At the outset, it’s basically going to be whatever friends and friends of friends you can drag along.
Could next time be different? I don’t see why not. One way you can grow is moving from “friends of friends” to “friends of friends of friends”. But I want so much more. I know so many people around America are worried that AI is moving too fast. As inspiring as it was, I’m left asking how we can get those millions of Americans into the streets.
The protest was on a Saturday, so there weren’t that many people around, but the ones who were seemed supportive; we got honks and cheers, etc. But somehow watching the spectacle of the cake-man made me feel like there was so much more potential for getting people’s attention… Dolores park was full of hundreds and hundreds of people hanging out, way more than attended the march. I felt a sense of potential… How many of these people could we get to join us next time? It feels like the question is more like “How do we get the audience to start dancing?” than “Why don’t they like our music?”3
Thanks for reading The Real AI! Subscribe for free to receive new posts and support my work.
1
This was the largest “AI safety” protest, specifically. I imagine there have been larger protests organized around resistance to AI (or related things like datacenters) from other motivations.
2
One was the opening statement I prepared for my two appearances before the Canadian parliament earlier this year. The other was as a senior in high school back in 2007, when I gave a speech to the school about what I viewed as the moral obligation to end factory farming and fight global poverty, after which I ran a largely unsuccessful campaign to raise money for mosquito nets… I donated ~$5,000 dollars I’d made working minimum wage at The New Scenic Cafe, I think we got <$1,000 from other students. It’s a bit embarrassing that I didn’t do a better job inspiring others to give, but my leadership and social skills have improved a lot since then…
3
I don’t mean to suggest that the Stop the AI Race message, framing, etc. is what’s going to resonate most with most people. But I think it’s already appealing enough that you could get many more people to join in.
LessWrong AI
https://www.lesswrong.com/posts/dkMGrCN7AaJfqAAND/reflections-on-the-largest-ai-safety-protest-in-us-historySign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
companymilliongovernment
Chinese AI rivals clash over Anthropic’s OpenClaw exit amid global token crunch
Chinese tech companies are engaged in a public war of words as they compete to capitalise on US start-up Anthropic’s decision to pull its industry-leading Claude models from open-source AI agent tool OpenClaw. The development comes as AI agents have triggered a huge increase in demand for AI tokens – the core metric of AI usage – raising questions about the long-term ability of industry players to meet this demand amid a growing global crunch in computational power. On Sunday, Anthropic...
Agentic Link Construction for Environment and Intent Aware 6G Communication
arXiv:2511.05094v2 Announce Type: replace Abstract: The emergence of sixth-generation networks heralds an intelligent communication ecosystem driven by the rapid proliferation of intelligent services and increasingly complex communication scenarios. However, current physical-layer designs-typically following modular and isolated optimization paradigms-fail to achieve global end-to-end optimality due to neglected inter-module dependencies. Although large language models (LLMs) have recently been applied to communication tasks such as beam prediction and resource allocation, existing studies remain limited to single-task or single-modality scenarios and lack the ability to jointly reason over communication states and user intents for personalized strategy adaptation. To address these limitat
Help Converts Newcomers, Not Veterans: Generalized Reciprocity and Platform Engagement on Stack Overflow
arXiv:2604.03209v1 Announce Type: cross Abstract: Generalized reciprocity -- the tendency to help others after receiving help oneself -- is widely theorized as a mechanism sustaining cooperation on online knowledge-sharing platforms. Yet robust empirical evidence from field settings remains surprisingly scarce. Prior studies relying on survey self-reports struggle to distinguish reciprocity from other prosocial motives, while observational designs confound reciprocity with baseline user activity, producing upward-biased estimates. We address these empirical challenges by developing a matched difference-in-differences survival analysis that leverages the temporal structure of help-seeking and help-giving on Stack Overflow. Using Cox proportional hazards models on over 21 million questions,
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Laws & Regulation
Fast Best-in-Class Regret for Contextual Bandits
arXiv:2510.15483v2 Announce Type: replace Abstract: We study the problem of stochastic contextual bandits in the agnostic setting, where the goal is to compete with the best policy in a given class without assuming realizability or imposing model restrictions on losses or rewards. In this work, we establish the first fast rate for regret relative to the best-in-class policy. Our proposed algorithm updates the policy at every round by minimizing a pessimistic objective, defined as a clipped inverse-propensity estimate of the policy value plus a variance penalty. By leveraging entropy assumptions on the policy class and a H\"olderian error-bound condition (a generalization of the margin condition), we achieve fast best-in-class regret rates, including polylogarithmic rates in the parametric
Power one sequential tests exist for weakly compact $\mathscr P$ against $\mathscr P^c$
arXiv:2604.03218v1 Announce Type: cross Abstract: Suppose we observe data from a distribution $P$ and we wish to test the composite null hypothesis that $P\in\mathscr P$ against a composite alternative $P\in \mathscr Q\subseteq \mathscr P^c$. Herbert Robbins and coauthors pointed out around 1970 that, while no batch test can have a level $\alpha\in(0,1)$ and power equal to one, sequential tests can be constructed with this fantastic property. Since then, and especially in the last decade, a plethora of sequential tests have been developed for a wide variety of settings. However, the literature has not yet provided a clean and general answer as to when such power-one sequential tests exist. This paper provides a remarkably general sufficient condition (that we also prove is not necessary).
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!