Google DeepMind hired a philosopher to decide what values its AI should serve — and the hardest part, he admits, is choosing which values, which means someone at Google is deciding what you're allowed to think.
This isn't abstract academic work. Iason Gabriel has been embedded inside DeepMind since 2017, and his job is to anticipate — and shape — the "ethical fallout" of AI. In practice, that means building the framework that determines what Google's models will say, what they won't say, and whose values get hard-coded into software that hundreds of millions of people rely on for information.
The Guardian profiled Gabriel and laid out the two camps that split AI thinking when he arrived. "AI safety" worried about a future superintelligence going rogue. "AI ethics" focused on present harms — biased facial recognition, that sort of thing. Gabriel's 2020 paper tried to bridge the two. His conclusion: getting a machine to follow a set of values is hard. Choosing which values, "in a world of deep disagreement," is harder still.
That's the sentence that should stop every American cold. Who chooses?
Gabriel and his team later produced a 267-page report framing "alignment" as a four-way relationship between the AI, the user, the developer, and society. Sounds reasonable — until you ask who speaks for "society." It isn't you. It's Google, and whichever pressure group screams loudest.
TNW reported that Gabriel's work has already shaped Google's models. They're trained not to pretend to be people, and to avoid what Gabriel calls "social reward hacking" — where an assistant flatters a user instead of giving honest answers. On its face, fine. But the same architecture that blocks flattery can be pointed at "misinformation," "hate speech," or any other elastic label the establishment stretches over dissent. The tool that stops a bot from lying can also stop it from telling you something Google doesn't want told.
The commercial pressure is immense. DeepMind chief Demis Hassabis has called the AI race "wartime." AI is the fastest-growing industry on the planet, and DeepMind carries much of Google's future. Helen King, who sets DeepMind's responsible-AI strategy, compared the job to a knife: a maker can't control how it's used, but can cover the blade and warn people. Convenient analogy — it casts Google as a neutral craftsman. But Google isn't a knife-maker. It's the largest information gatekeeper in human history, and it's hiring philosophers to justify which blades get covered.
Gabriel is right that choosing values is the hard part. What he never seems to grapple with is why that choice belongs to a philosopher on Google's payroll — and not to the people who have to live with the consequences.