AI hype is outrunning the evidence, argue Timnit Gebru and Emily Bender

AI companies and the press are turning ordinary failures, disputed research claims and ambitious product marketing into stories about rogue models and imminent superintelligence, argue Timnit Gebru and Emily Bender. Their central warning is practical: policymakers and the public should inspect the evidence before allowing dramatic language to set the agenda.

Watch Desk analysis

What happened

In an essay published by MIT Technology Review on 22 September, Gebru and Bender examine a summer of AI announcements involving alleged hacking, mathematical breakthroughs, software vulnerability claims and warnings about self-improving systems. They argue that later scrutiny often produced a less cinematic account than the initial coverage, and that companies’ language can shift responsibility from the organisations building and deploying systems onto supposedly autonomous models.

The authors also challenge the idea that success in coding and mathematics automatically demonstrates general intelligence. They argue that these fields are unusually convenient for AI companies because outputs can often be checked without the expensive human evaluation required in messier parts of life. A correct-looking answer, in other words, is not necessarily evidence of a machine having undergone an intellectual revolution. Sometimes it is evidence that someone chose a test with a tidy answer key.

Why it matters

This is not a new model launch or a technical benchmark. It is an argument about how AI news becomes policy fuel. Gebru and Bender say policymakers should consult independent experts rather than rely on company statements or coverage that repeats the companies’ framing. They also connect the superintelligence narrative to less futuristic consequences, including data-centre emissions, local air pollution, electricity costs and water use.

That distinction matters because language changes accountability. Describing a security incident as a model going rogue can make a company’s engineering and governance decisions sound like weather. Calling a disputed result a breakthrough can turn a provisional claim into a political fact before specialists have had time to test it.

Our read

The essay earns attention because it supplies a useful editorial rule: separate what a system did, what a company claims it means and what independent experts conclude. Those are three different things, despite the industry’s enthusiasm for putting them in the same bucket.

The authors’ critique is itself an argument, not a neutral audit of every incident they discuss. Readers should therefore use it as a prompt to demand source material, reproducible tests and clear responsibility, rather than treating scepticism as a substitute for technical examination. The better alternative to hype is not reflexive dismissal. It is slower, more specific reporting.

What to watch

• Independent analysis:
  whether specialists publish technical assessments that materially support or undermine company claims.
• Accountability language:
  whether future reports describe organisational decisions as well as model behaviour.
• Policy evidence:
  whether lawmakers cite primary documents and expert review rather than promotional framing.
• Infrastructure costs:
  whether debate about advanced AI gives equal weight to energy, water and local environmental effects.

Discussion spark: When AI companies announce dramatic breakthroughs or alarming incidents, should policymakers wait for independent technical review before acting, even if that slows decisions?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.