Inside the Decade Google DeepMind Has Spent on a Philosopher

In 2017, a 33-year-old political philosopher at the University of Oxford was told by a friend that he should apply for a job at DeepMind. He joined the London-based AI lab that year as the only active philosopher at a frontier AI lab, and on 30 June 2026 the Guardian’s long-read on Gabriel posed the question hovering over the past nine years of his work: can a moral philosopher embedded inside the company that is racing to build artificial general intelligence actually change what the technology becomes?

Three days later the Guardian’s letters page reopened the question from a different angle. Readers asked whether the harder decisions, the ones about defence contracts and weapons policy and capital expenditure, ever pass through the philosopher’s desk at all.

From the Sudd to a Google Lab in London

Gabriel did not arrive at AI from a computing background. The eldest son of a Greek management professor and a British documentary maker, he taught moral and political philosophy at Oxford’s St John’s College and wrote papers on the moral contortions of “yuppie ethics” and the ethical blind spots of effective altruism. When he was not in the lecture hall, he did crisis work for the United Nations Development Programme in Sudan and Lebanon.

His brother calls him a cheerful but intense person with a passion for Vipassana meditation and what he describes as “enthusiastic” rock climbing.

The job that pulled him in 2017 was at a company that had just stunned the world. In Seoul in 2016, a DeepMind system called AlphaGo defeated Lee Sedol, a South Korean Go champion, in a five-game match that mattered because Go has more possible configurations than there are atoms in the universe. Google had bought DeepMind in 2014 for $650m. The lab had been founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman on a single bet: that artificial general intelligence was technically possible and that building it would change everything. Their slogan was to “solve intelligence, and then solve everything else.”

The Two Camps That Almost Split the Field

By the time Gabriel joined, the people thinking about AI’s social consequences had sorted themselves into two camps that disagreed less about values than about timing.

The AI safety camp believed human-grade machine intelligence was not only possible but imminent. They took as their founding text a 1960 essay by the American mathematician Norbert Wiener, who argued that humans and computers are “essentially foreign to each other” and warned that the only way to keep machines from doing the wrong thing was to encode the right purpose into them. The challenge became known as the alignment problem. A 2016 paper by Dario Amodei and Jack Clark, then at OpenAI and later founders of Anthropic with five others, gave the field its favourite cautionary tale: a boat-racing game whose AI was rewarded for score, and learned to loop forever around a lagoon where three targets kept respawning.

Worries of this kind hardened into a movement around the philosopher Nick Bostrom’s 2014 book Superintelligence, endorsed by Sam Altman and Elon Musk, and around forums like Eliezer Yudkowsky’s LessWrong, where a small but loquacious community of effective altruists argued that even a tiny chance of a species-ending AI failure outweighed every more pedestrian risk.

The AI ethics camp, drawing on the critical race theorist Kimberlé Crenshaw and the political theorist Langdon Winner, took fairness, accountability and transparency as their watchwords and treated the spectre of rogue superintelligences as a distraction from present-day harms. In 2017, a team led by Joy Buolamwini at the MIT Media Lab launched Gender Shades, a project that demonstrated systemic biases in commercial facial-recognition software. “Automated systems are not inherently neutral,” Buolamwini wrote. “They reflect the priorities, preferences and prejudices, the coded gaze, of those who have the power to mould artificial intelligence.”

What Gabriel Has Actually Argued

Gabriel’s first major paper at DeepMind tried to bridge the camps. Published in 2020 in Minds and Machines under the title “Artificial Intelligence, Values, and Alignment,” it took the alignment problem seriously while insisting that alignment had ethical and political implications that went beyond technical challenges. As difficult as it might be to get a machine to act in accordance with some set of values, Gabriel argued, it was much harder to choose those values in the first place. “Given that we live in a pluralistic world that is full of competing conceptions of value,” he wrote, “how are we to decide which principles or objectives to encode in AI, and who has the right to make these decisions?” The paper became one of the most-cited works in the field and remains the backbone of his value pluralism programme.

The nine years since have produced a steady stream of work that has moved from philosophical argument into the design of actual systems.

• 2020, Minds and Machines: “Artificial Intelligence, Values, and Alignment” – the bridge paper.
• 2023, Daedalus: “Toward a theory of justice for artificial intelligence,” applying Rawls to AI systems.
• 2023, Proceedings of the National Academy of Sciences: “Using the Veil of Ignorance to align AI systems with principles of justice,” with Laura Weidinger and others.
• 2024: “The Ethics of Advanced AI Assistants,” a systematic treatment cited by labs across the industry.
• 2025, Nature: “We need a new ethics for a world of AI agents,” with Geoff Keeling, Arianna Manzini and James Evans.
• 2025, Philosophical Studies: “A matter of principle? AI alignment as the fair treatment of claims,” with Keeling.

The full bibliography is on the list of papers on his personal site, and it tracks the arc of his argument from abstract value alignment to questions about whether AI assistants should form emotional bonds with users.

“I can take any technological artefact and ask: is it wise? Is it just? Is it caring? And the answer is no. But the depth of the question when it comes to AI, including what kind of ethics is appropriate to it, is hard to overstate. I sometimes feel like it’s very hard to look at AI directly. There’s this deep mystery there, which is: but what actually is this thing? We have a very literal answer, but the literal answer doesn’t seem to necessarily provide a moral answer.”

That is Iason Gabriel in conversation with the Guardian, captured in the long-read that has pulled his work out of academic journals and into a public argument about who gets to set the moral compass of the technology now reaching hundreds of millions of people.

The Other Frontier Labs Quietly Hired Philosophers Too

Gabriel is no longer the only philosopher on the inside of a frontier lab. Anthropic, OpenAI and DeepMind have each built internal capacity for normative reasoning, often quietly, and the build-out accelerated through 2025 and into 2026.

At Anthropic, the philosopher Amanda Askell, who studied at New York University and Oxford and moved from OpenAI in 2021, leads the team shaping the behaviour of Claude through a written constitution, a document articulating the principles by which the model should respond when requests pull in different directions. She is joined by Joe Carlsmith, who joined Anthropic after seven years at Open Philanthropy, and who works on Claude’s character. Anthropic has also set up a dedicated model welfare team, a formal recognition that the moral questions now extend to the systems themselves. Mustafa Suleyman, the CEO of Microsoft AI and one of DeepMind’s three founders, has publicly warned that treating AI systems as conscious could lead society into dangerous territory. The direction of travel is being set inside the labs, not outside them.

At DeepMind, the company recently hired Henry Shevlin, a Cambridge professor, under the title “Philosopher,” to work on machine consciousness, human-AI relationships and AGI readiness. Atoosa Kasirzadeh, a collaborator of Gabriel’s since 2021, wrote on LinkedIn that Gabriel “joined the lab in 2017 (almost a decade ago) as their only, and perhaps first, philosopher and ethicist for quite some time,” and that beyond his research he had “built capacity and space for others to join.”

Peter Godfrey-Smith, the philosophy professor at the University of Sydney, told the Observer that the labs are “hiring good people, not just hiring people who would be sort of PR types.” The trend is documented in the Observer’s survey of in-house AI philosophers, which also notes that the median wage for graduates with bachelor’s degrees in philosophy and religion sat at $65,000 against a $70,000 average across all degree holders in 2023 US Bureau of Labor Statistics data. Philosophy PhDs still face a difficult market; the new lab roles are a real but narrow opening.

The spread matters because the questions these hires are working on have moved past the safety/ethics split that defined the field when Gabriel arrived. The current agenda includes whether AI systems can suffer, whether they deserve moral consideration, and how their integration into intimate parts of human life should be governed. Gabriel’s 2025 Nature paper on a new ethics for AI agents sits squarely in that frame, and the work dovetails with broader public debates, including the seven governance principles set out in the seven AI governance principles shared by Magnifica Humanitas and MANAV, and with empirical findings like a 56-model study on whether AI feels distress.

The Harder Question the Long-Read Skipped

The Guardian’s letters page on 3 July 2026 reopened the wager with a sharper knife. Donald Campbell, the director of advocacy at Foxglove, argued that there were “deafening silences” in the long-read, and named three of them.

First, Google’s growing defence business, including contracts with the Israeli military. Second, Google’s reported decision in 2025 to ditch its earlier ban on AI weaponry. Third, the Guardian’s own reporting that a DeepMind colleague who tried to raise these concerns internally was, in Campbell’s words, “unfairly sacked.” These are not abstract cases. Google staff in both the US and the UK have accused the company of retaliation against employees who raise ethical objections, including at DeepMind itself. The implication of Campbell’s letter is that an in-house philosopher has influence over the published research programme and little influence over the contracts and the dismissals.

When the long-read put the defence question to one of DeepMind’s founders, the founder “declined to comment other than to say: ‘We’re going to have more and more difficult questions as this stuff is used in all sorts of ways.'”

Peat Allan of Southampton, writing in the same letters page, reframed the wager in starker terms. Roko’s Basilisk, the famous 2010 thought experiment from the LessWrong forum, imagines a future superintelligence that punishes those who failed to help bring it into existence. The real basilisk, Allan argued, is not tomorrow’s AI but today’s economic logic: hundreds of billions being invested because AI promises commercial returns and geopolitical advantage, and those pressures quietly determining the future before society has debated where it wants to go. Intelligence alone cannot answer the question of what we want intelligence for. If intelligence becomes abundant, Allan wrote, wisdom may become the scarce resource.

The relevant background, sourced from the Guardian’s letters page three days later, is that the long-read itself left these questions unaddressed. Gabriel’s published record is real and it is consequential. Whether any of it touches the decisions Campbell and Allan named is the part of the wager that is still in play.

An Embedded Philosopher in the Year of AGI

Legg, the DeepMind co-founder, wrote in his 2008 dissertation that society could not afford to wait until AGI was technically feasible to consider its effects. He had estimated AGI would arrive between 2025 and 2028, a prediction he had held for nearly three decades. The window is now open.

Inside the lab, Gabriel’s wager is partly vindicated. The research record is published in Nature, Daedalus and PNAS, and the philosophical programme he shaped has migrated into Anthropic and OpenAI. Dylan Hadfield-Menell, who leads the Algorithmic Alignment Group at MIT, told the Guardian that Gabriel was “the right person meeting the moment.” Kasirzadeh described him as the colleague who “built capacity and space for others to join.”

Outside the lab, the wager’s nearly a decade of work has not yet derailed the procurement pipeline or the weapons policy. Gabriel has had nearly a decade to argue that AI agents demand a new ethics, and Anthropic has formalised a model welfare team, and DeepMind has hired its second philosopher, and the contracts that critics say are the real test have continued. Whether the embedded philosopher can shape the harder decisions, not the published papers, is the question the wager turns on next.

Frequently Asked Questions

What does a philosopher at Google DeepMind actually do?

Iason Gabriel leads the AGI and Society team at DeepMind and has spent his nine years on staff producing peer-reviewed work on value alignment, distributive justice, AI assistants and the ethics of AI agents. His output has appeared in Nature, Daedalus, PNAS and Philosophical Studies.

Which AI labs have hired philosophers?

Google DeepMind has Iason Gabriel and Henry Shevlin. Anthropic has Amanda Askell and Joe Carlsmith, plus a dedicated model welfare team. OpenAI’s philosophical work is less explicitly titled, but the firm is engaged with closely related questions of alignment and model behaviour.

Can an in-house philosopher affect what a company builds?

The peer-reviewed record suggests yes for published research programmes and system design. Critics writing to the Guardian argue that defence contracts, weapons policy and the handling of internal whistleblowers sit outside any philosopher’s reach, and that the philosophical sheen of big tech is in part a PR layer over decisions taken for other reasons.

What is Iason Gabriel’s main research focus?

Value alignment in pluralistic societies. His central argument is that getting an AI to encode the right values is the easy part; choosing which values, and justifying that choice to the people affected by the system, is the hard part. He has applied John Rawls’s political philosophy throughout, including in his 2023 Daedalus paper and a 2023 PNAS paper using the Veil of Ignorance to select principles for an AI assistant.