A case for slowing down AI over bias and future existential risks, despite its alleged inevitability, the AI arms race with China, and safety by experimentation
Sigal Samuel / Vox : Tweets: @sigalsamuel , @upfromthecracks , and @maosbot Tweets: Sigal Samuel / @sigalsamuel : The AI race has gone berserk. We need to SLOW DOWN. “No!” I hear some people say. “We can't, because... 1) tech is inevitable 2) China!!! 3) we need to play with strong AI to figure out how to make strong AI safe” I tackle all those in this new piece https://www.vox.com/... @upfromthecracks : “AI” is not like electricity, it's like multilevel marketing https://www.vox.com/... https://twitter.com/... Michael A Osborne / @maosbot : Please read this most thoughtful piece, populated with excellent thinkers https://twitter.com/...
Context & Ripple Effects
Days before this piece ran, the New York Times argued humanity must either adapt at AI's pace or make a collective, enforceable decision to slow development — Sigal Samuel's Vox essay picks up that second branch and supplies the rebuttals. It lands two weeks into the fallout from the open letter to pause AI training, whose signatories were already walking back their positions amid fakes and expert dissent, leaving the slowdown camp without a working playbook.
Samuel's targets map directly onto positions in the related coverage: the China-arms-race objection echoes Beijing's parallel moves — multibillion-dollar state VC funds for hard-tech startups, domestic-equipment mandates for chipmakers, and draft AI rules of its own — while the safety-by-experimentation objection runs against what an ex-OpenAI safety researcher later described as labs racing to AGI and cutting corners on alignment.
First-order effects
- The essay gives the slowdown camp, freshly demoralized by the pause letter's collapse, a structured rebuttal to the three arguments — inevitability, China, experimentation — that labs and their defenders use to justify current speeds.
- It puts Vox squarely in the safety-leaning media bloc alongside the Times' earlier framing, sharpening the editorial split with outlets and executives who argue the US should not rush regulation the way it says the EU did.
Second-order effects
- Executives like Box CEO Aaron Levie, who want to avoid EU-style haste, now face a published counterargument that names the China race as a rationale rather than accepting it as a constraint, forcing the pro-speed side to defend the arms-race logic on the merits.
- The billionaire investors of the internet era fighting over whether AI's future is concentrated safety or unfettered advancement gain a sharper fault line: Samuel's case makes 'we must build strong AI to learn how to control it' an explicit position to reject rather than a default.
Third-order effects
- If the pattern holds — safety insiders escalating warnings, security researchers divided on catastrophic potential while global regulatory frameworks lag — the slowdown debate migrates from op-ed persuasion toward the kind of collective, enforceable decision the Times called for, with bias harms cited today becoming evidence in that case.
- The China argument cuts both ways structurally: Beijing simultaneously subsidizes chips and hard-tech startups while drafting rules restricting chatbot content, suggesting both governments may end up racing on compute and regulating on deployment — a combination that makes unilateral US slowdown harder but not obviously pointless.
The trend: The AI governance debate is shifting from whether to slow down to how a collective, enforceable slowdown could work, as safety researchers inside labs and security experts outside them keep escalating the risk case.