The PauseAI Protest: A Photo-Essay
Read the original on Astral Codex Ten →
Summary
First-person reportage from Scott's first protest, a PauseAI march on the San Francisco AI companies. It opens on the unmarked headquarters — X.AI at 3180 18th, Anthropic at 500 Howard, OpenAI at 1455 3rd behind a blobby blue statue — noting that all of this is on Google Maps, that the security therefore stops only malefactors incapable of a web search, and that it 'protects against the stupidest 5% of ways things could go wrong' before pointedly telling the reader this is a metaphor. The protest's ask was deliberately cheap and conditional: that CEOs endorse a pause if every rival, American and Chinese, agreed to the same, so nobody would be the sole holdout — Hassabis has already agreed in principle and there is debate about whether Amodei has. He counts about 120 people against the organisers' 200. Michael Trazzi organised; David Krueger, Holly Elmore, Nate Soares and Guido Reichstadter spoke, with wildly varying competence — some obviously AI nerds who had found a megaphone, while Krueger, holder of the nerdiest credentials present, turned out to be genuinely fiery. Trazzi and Reichstadter had recently hunger-struck outside DeepMind and Anthropic for seven and thirty days respectively (Gandhi's longest was twenty-one), prompting Scott's honest objection: the point of a hunger strike is that it is dangerous, so what is it for when aimed at corporations that will not listen and the only outcomes are starving or quitting? The march ran Anthropic to OpenAI to X with a friendly police escort and chants he reproduces; an attempt to start 'You will not replace us!' was shushed. The core of the piece is an anthropology of the coalition. One faction is rationalists from technical AI safety — adequately groomed, signs more complicated than optimal, loyalties visibly divided, occasionally interrupting the protest to greet AI company employees leaving the building. The other is veteran protesters out of animal rights and climate, wild-eyed and in cause-branded vests, with serious non-AI-generated signs about corporate greed killing children. Scott is candid that he is not fully on board with the second group, some of whom spend a tenth of their energy on AI companies and the rest attacking insufficiently radical allies, risking a circular firing chamber — then notes it would be hypocritical to complain about circular firing chambers while refusing an olive branch, and concludes that a protest with achievable aims and an ecumenical outlook is a reasonable place to start building unusual bridges. It ends on vandalised Google Gemini bike advertisements, which his old liberal self would have deplored and his new proletarian self concedes were kind of funny.
Why this score
Quality 63 · Strong. Solid-to-Strong. Modest in ambition and honest about it, but it earns its place on two things: the two-faction anthropology, which is a genuinely useful description of the coalition problem facing AI-safety activism and is the sort of observation only a participant-sceptic could write, and the opening security-theatre metaphor, which is the sharpest paragraph in the piece. The hunger-strike objection and the admission of his own hypocrisy risk are characteristic and add real value. Held here because it is short, it is reportage rather than argument, a substantial share of the content lives in photographs the text only gestures at, and its interest is largely documentary — a record of a specific afternoon rather than a piece anyone will return to for its ideas.
Claude’s paradigm shift 35 · Slight. Slight-to-Moderate. Event reportage is inherently low-novelty, and the observation that a movement's technical and activist wings mistrust each other is familiar from every other movement. The mildly fresh element is the specific diagnosis of that split inside AI safety at the moment it started holding street protests, and the first-hand documentation of what the pause coalition actually looked like.
Real-world impact 2 · Minor. Minor. Small-community influence: a widely-followed figure in AI safety publicly attending, describing and cautiously endorsing bridge-building with the activist wing, which nudges the norms of that subculture at a moment when it was deciding whether to do street politics at all. No effect beyond the community, and none on laws, institutions or ordinary lives.