The phrase and its history
The motto 'move fast and break things' expressed a genuine Silicon Valley philosophy: that the costs of moving too slowly exceeded the costs of errors made in haste. For a social network, the things that got broken were relatively abstract: privacy norms, platform trust, civic discourse. The same philosophy is now the dominant ethos in AI development. And the things that could get broken are qualitatively different.
What is at stake
AI systems are being deployed in medical diagnosis, legal research, financial advice, hiring decisions, bail recommendations and social benefit determinations. When these systems fail — and they do fail, in ways that are often opaque — the costs are borne by patients, defendants, job-seekers and benefits claimants. The costs are not borne by the companies that built the systems.
Why self-regulation is insufficient
The tech industry's record on self-regulation is not strong. Social media companies were warned about the harms of algorithmic amplification of harmful content for years before any meaningful action. The lesson of virtually every major technology externality is that self-regulation defers accountability until the harms become undeniable.
The counter-argument
Proponents of the current pace argue that AI is too nascent to regulate precisely — that premature regulation will entrench incumbents and slow beneficial developments. There is some truth in this. But 'we should wait for more information before regulating' is also exactly what the tobacco industry argued.