Incentives

If the future is uncertain, one might hope at least to find reliable guides to it. The difficulty is that almost everyone with a public view on AI has a reason to hold it that is not purely a matter of evidence.

This is not an accusation of dishonesty; it is an observation about how incentives shape what people notice. The laboratories building frontier AI systems have an interest in describing their products as powerful, since power attracts customers, investment and talent, and they sometimes also have an interest in describing them as dangerous, since danger invites regulation that incumbents can afford and challengers cannot. Their staff frequently believe both things sincerely, which does not make them right. Safety organisations have an interest in the salience of risk, because salience brings funding and attention. Consultancies have an interest in transformation, because transformation needs consultants. Venture capitalists have an interest in the belief that a new world is arriving, because they have bought tickets to it. Journalists have an interest in whatever will be read, which usually means the extreme ends of any distribution. Politicians have an interest in narratives that assign credit or blame conveniently. And professional sceptics — there is now a flourishing trade in AI scepticism — have an interest in debunking, since a sceptic who concedes that something important is happening loses their market position.

The result is a landscape in which you can find an articulate, credentialed, apparently sincere advocate for almost any view, and in which the most confident voices are rarely the most reliable.

from Surviving the AI Apocalypse (2026)