"Thinkers Into Hinkers? Do AI Companies Want Philosophy, or Just Its Compliance?" An essay H. Mustafa Akyol (Keywords: AI Ethics; Governance;Institutions;Epistemic Independence;Public Philosophy)
If you enjoy reading this, please consider becoming a Patreon member or making a donation. The Philosopher is unfunded and your support is greatly appreciated.
The recent trend of AI companies hiring philosophers has largely been celebrated as a vindication for the discipline. The Economist recently reported that American philosophy graduates now find work at higher rates than their computer science peers, and that leading AI laboratories are recruiting philosophers into senior roles -– not merely to sharpen a model's chains of reasoning, but to draft the documents, such as Google's published 'AI Principles', that are meant to govern how a system behaves. Where criticism of this trend exists, it has tended to focus on familiar concerns: ethics washing, the risk of philosophy becoming a public-relations exercise, and the migration of talent away from universities that Luciano Floridi has called a 'haemorrhaging'. These concerns are real. But they may be missing a further, less visible risk, one that has nothing to do with where a philosopher's paycheck comes from.
Floridi's own diagnosis is more specific than the headlines suggest. In a recent essay, he argues that AI laboratories are not really buying philosophical knowledge but a temperament – the disposition Keats called negative capability, the ability to remain in uncertainty without an 'irritable reaching after fact and reason'. Philosophy departments cultivate this temperament unusually well, which is part of what makes their best graduates so easy to poach. His proposed remedy is structural: let philosophers hold two posts at once, one foot in the university and one in the laboratory, on condition that the university chair remains theirs to keep and their right to publish and dissent remains intact.
It is a generous and historically literate solution – he is right that Descartes was a soldier, that Leibniz was a diplomat and mining engineer, and that the philosopher confined to a single university post is a nineteenth-century invention rather than the discipline's natural state. But Floridi's remedy answers a question of institutional custody: who keeps the philosopher, and on what terms. It does not quite reach the question I want to raise, which is what happens to the practice of philosophy itself once it is put to work inside an organisation, regardless of who signs the contract. As AI ethics increasingly becomes a matter of compliance, auditing, and risk management – the very apparatus that produces documents like the 'AI Principles' The Economist describes – philosophy itself risks being transformed from a practice of questioning into a function of implementation. Even a philosopher with an unimpeachably independent university chair, free to publish and dissent, can find that the daily work reshapes what 'thinking' means long before anyone notices the shift.
A thinker asks whether a framework is justified. A hinker helps implement it more efficiently.
This is not a claim that compliance is worthless. Laws, safety standards, and ethical baselines matter, and few people would argue that AI systems should be exempt from them. The trouble starts when compliance is mistaken for ethics itself. Across contemporary AI governance, 'responsible AI' has increasingly come to mean audits, checklists, certification schemes, and risk registers. These mechanisms have real value; they create accountability and catch certain harms before they metastasise. But they also perform a quiet substitution. Questions become metrics. Values become key performance indicators. Moral reflection becomes process management. The resulting apparatus is very good at answering 'how do we comply with the ethical requirements we've been given?' and almost never asks 'why were these the requirements, and who decided that?'
It is worth being specific about the scale of this, since it is no longer a matter of a handful of isolated hires. Anthropic's philosophy contingent includes Amanda Askell, who leads the team responsible for Claude's 'constitution' – a document Anthropic has described as drawing on sources 'as diverse as Immanuel Kant, Apple's terms of service and the Universal Declaration of Human Rights', and which has been nicknamed, inside and outside the company, its 'soul doc' – alongside Joe Carlsmith, Ben Levinstein, and Jackson Kernion. Google DeepMind's roster includes Iason Gabriel, who leads the lab's AI morality work, together with Atoosa Kasirzadeh, Henry Shevlin, and others. OpenAI's Sam Altman has said the company consulted 'hundreds of moral philosophers' while drafting the principles behind ChatGPT. Several of these hires left tenured academic posts to take the roles, which may itself suggest how seriously the work is taken on both sides. None of this should be read as a claim about what any particular person is doing, still less as a judgement on them; taking a position at an AI lab is, as the historical examples above suggest, an entirely reasonable choice, and one this essay has no interest in second-guessing. The relevant point is only that a large share of what these roles are for – writing constitutions, drafting principles, building the documents that will govern how a system behaves – sits, by its own description, closer to the framework-building end of the spectrum this essay has been describing than to open-ended inquiry. Whether that reflects what these institutions most want from the philosophers they hire, or simply what they find easiest to reward, is the question the rest of this essay tries to answer.
It is worth pausing on what kind of claim this essay is making, and what kind it is not. This is not a description of what any of the philosophers named above are doing behind closed doors – none of us outside their teams can see that, and I have no wish to guess. It is not an accusation, and it is not a claim that the pattern described below is already the settled state of affairs. It is a risk assessment: an argument that a particular institutional pressure exists, that it would predictably push in one direction if left unchecked, and that a moment of rapid hiring is exactly the situation in which such a pressure is worth naming early, rather than only once it has already done its work.
***
Compliance concerns conformity to an existing framework. Philosophy concerns the examination of that framework. The two activities are related, but they are not the same, and institutions are structurally more likely to reward the first than the second.
A company would generally find it easier to fund research into how to operationalise a fairness metric than to fund a sustained argument that the system should not exist in its current form at all. It would tend to welcome help aligning a model with the organisation's stated goals more readily than a colleague who keeps asking whether those goals deserve to be pursued. This is the trap, and it rarely announces itself as one. Nobody sits a philosopher down and asks them to stop thinking. The transition, where it happens, tends to be gradual and almost administrative: a person hired to interrogate assumptions finds an increasing share of their week spent translating assumptions into governance procedure instead.
The trouble starts when compliance is mistaken for ethics itself... Questions become metrics. Values become key performance indicators. Moral reflection becomes process management.
There is a word for this in Turkish that I think names the phenomenon more precisely than any English term I know: hınk deyici, someone who reflexively agrees with authority and echoes the prevailing view rather than testing it. The nearest English cognate is 'yes-man', but that phrase carries a whiff of cowardice that misses what is actually happening here. Nobody involved need be spineless, or even particularly aware that anything has shifted. Borrowing the shape of the Turkish term, I want to propose a companion English one: a hinker. A thinker asks whether a framework is justified. A hinker helps implement it more efficiently. A thinker expands the space of reflection; a hinker accelerates execution within whatever space has already been agreed. It is worth being precise about which two forms of approval this covers, since they are easy to conflate: a hinker is not only someone who approves of the company, uncritically endorsing whatever it wants to build; a hinker can just as easily be someone who approves of compliance itself, uncritically accepting that 'alignment' and 'auditing' exhaust what ethical diligence requires. The second failure is quieter and, I suspect, more common among philosophers who would never think of themselves as corporate apologists. Most philosophers now working inside AI companies are, I want to be clear, neither cynical nor complicit – many are exactly as rigorous and independent as their academic training made them. The pressure that turns a thinker into a hinker is structural, not personal: it lies in what the institution is built to reward, not in any individual's failure of nerve.
Hannah Arendt gives this something closer to a proper name than anything borrowed so far. Reporting on Eichmann's trial, she was struck less by any demonic hatred in the man than by an almost total absence of reflection: he spoke in the fixed phrases of his office, and seemed unable to step far enough outside his role even to ask whether the role itself could be defended. She called this thoughtlessness, and she meant something quite specific by it – not stupidity, and not ignorance, but a professional habit so complete that it closes off the space in which a person might stop and ask what she is actually doing. One could object, fairly, that this is a strange comparison to reach for: thoughtlessness under totalitarian command is one thing, and a well-paid consultant drafting a fairness metric is a very different and much smaller thing, and the disanalogy is real. But the mechanism Arendt is describing does not need totalitarianism to run. It needs only an institution efficient enough that following its internal logic starts to feel like the whole of the job. Late in her life Arendt described the alternative as thinking without a banister – reasoning that refuses to lean on an inherited rule for support merely because the rule is already there, already signed off, already load-bearing. A hinker, in the sense this essay is using the word, is someone working with exactly such a banister permanently installed. The framework does the leaning on their behalf, and the daily habit of checking a system against it can quietly substitute for the harder work of asking whether the framework deserves the weight it has been given. Nobody need notice the substitution happening. That is rather the point.
Compliance concerns conformity to an existing framework. Philosophy concerns the examination of that framework.
It would be tempting to conclude from this that the safest place for a philosopher is as far from any institution as possible – that the banister only appears once someone accepts an office to lean on it in. But that conclusion does not survive much contact with the discipline's own history, and it is worth setting the temptation aside before it does any more work. Descartes served as a soldier before he became the author of the Meditations, and by his own account it was the itinerant years, not the stove-heated room, that first unsettled his certainties. Leibniz spent decades as a working diplomat and mining engineer, drafting political memoranda between treatises on monads. Hume was a librarian and, for a time, a civil servant under-secretary, and complained more than once that his administrative posts paid rather better than his philosophy did. The philosopher entangled in worldly institutions is arguably the historical norm rather than the exception; the mid-century image of philosophy as a cloistered, purely academic pursuit is itself the historical anomaly, and a fairly recent one. So the question was never whether philosophers should work inside powerful organisations, or whether doing so automatically hands them a banister to lean on. The question Arendt's diagnosis actually poses is narrower and harder to escape by staying clear of industry: whether a person inside such an organisation keeps enough distance from its internal logic to ask, periodically and seriously, whether that logic still deserves the trust it has been given.
With the term on the table, the question becomes empirical rather than definitional: are the philosophers now working at AI companies actually drifting toward hinker-work, or is the risk, so far, mostly theoretical?
***
An obvious objection is available, and it deserves more than a brief concession. Why assume that a company's interest in implementation crowds out its interest in genuine critique, rather than the two running together? There is a real economic case that they do run together. As the industry commentator Gerald Pallor has put it, a certification or audit process functions as a kind of bouncer at the door of enterprise sales – no certificate, no deal – and insurers are increasingly willing to offer better terms to firms that can document real oversight and bias testing, which means that documented rigour carries a cash value quite apart from whatever ethical good it does. On this view, a philosopher embedded in an AI company is not competing with the firm's commercial interest; she is serving it, and a company that wanted only the appearance of ethics could presumably get that more cheaply than by hiring philosophers with doctorates from Oxford and Berkeley.
This is a fair challenge, and the honest answer is that it relocates the risk rather than removing it. The claim here is not that firms lack a commercial interest in genuine philosophical rigour. It is that firms have a much easier time rewarding, measuring, and defending the parts of that work which yield a legible artefact – a certificate, a published constitution, a benchmark score – than the parts which yield only a harder question. This is a specific version of a general problem in organisations that assign people several kinds of work at once: when some of what an employee does is easy for others to observe and verify and some is not, incentives drift toward the observable part even when the organisation formally values both, because performance review, promotion, and resourcing all have to be justified in terms someone else can check. A philosopher who spends a quarter drafting a constitution has something to show for it. A philosopher who spends a quarter arguing that a product should not exist in its current form has, for the organisation's own bookkeeping, produced nothing any dashboard can register – even if the argument turns out, years later, to have been the most valuable thing anyone in the building said. Naming this risk does not require assuming AI companies are cynical about ethics. It requires only the weaker and much more defensible assumption that measurable work tends to crowd out unmeasurable work inside large organisations generally, whatever the people involved sincerely believe they value.
A society can build a perfectly auditable, thoroughly certified AI system and still have failed to ask whether it is heading in a direction worth going.
There is a small piece of evidence that speaks to where this boundary tends to fall in practice, even if it does not settle the matter. Joe Carlsmith writes, alongside his work on Claude's constitution, a long-running series of personal essays on philosophy of mind, AI moral status, and related questions – open-ended, unresolved work of exactly the kind this essay has been arguing philosophy should keep doing. Nearly all of them carry some version of the same disclaimer: 'I work at Anthropic, but I am here speaking only for myself and not for my employer.' The disclaimer is unremarkable as professional practice and implies nothing improper. But it marks, with more precision than most abstract argument could, exactly where this essay's distinction tends to fall: the constitution is the company's, drafted collectively and owned as an institutional document; the open questioning happens under a heading that keeps it from counting as the institution's own view. Whether that questioning ever finds its way back into the document it is formally kept apart from is not something an outside observer can verify, and outside observers are not always convinced that it does. Asked about the influence of in-house philosophers on product decisions, one AI industry engineer interviewed by the International Business Times put the doubt plainly: 'I'm not really confident whether they are actually taking philosophers into account at all, to be honest ... Primarily, all these companies, their main aim is revenue generation.' That is one person's impression, offered about the industry in general rather than any single lab, and it may undersell the people involved. But it names, from outside, roughly the same gap this essay has been trying to describe from within: the distance between a framework a philosopher helped write and the deeper interrogation that framework was always meant to follow.
There is now a considerably richer case study available than either of these, and it happens to cut both ways, which is exactly why it is worth dwelling on. A recent Guardian profile of Gabriel reports two things worth holding side by side. First, according to Rohin Shah, DeepMind's director for AGI alignment and safety, the framework Gabriel's team developed on the ethics of AI assistants 'has had real practical use' for technologists deciding what behaviour to train into systems like Gemini. Second, and more concretely: the same profile credits Gabriel's research on the risks of anthropomorphic AI with a specific design choice – Google's assistants, including the one launched this year as Gemini Spark, are trained not to imply they are people or to act like an 'interactive buddy'. Taken together, these are about as clean an instance as one is likely to find, outside the company itself, of philosophical critique shaping how a product actually behaves rather than merely accompanying it on the side.
Set against this, the same Guardian profile describes an episode in which commercial and geopolitical pressure appears to have simply overridden whatever deliberation preceded it. In April 2026, Google agreed to let the US military use its AI for 'any lawful government purpose', reversing a restriction the company had held since DeepMind's 2014 acquisition, and one that Anthropic had just refused at real commercial cost. Demis Hassabis, DeepMind's own chief executive, was unusually candid about what pressures of this kind do to the sort of deliberation his company's philosophers are paid to provide: this is not how he had hoped AI development would go, 'where we would be contemplating this philosophically and carefully considering each next step. We're not in that world'. That is not a claim this essay is making about the industry from outside. It is the head of one of its most consequential labs, describing from inside something close to the crowding-out this essay has been arguing is structural rather than incidental.
None of this settles which scenario predominates day to day inside any particular lab, and it would be dishonest to claim otherwise from outside. What can fairly be said is this: the institutional pressure described above is real and well documented in organisations generally; the philosophers currently doing this work are producing at least some artefacts, like the anthropomorphism research that shaped how Gemini Spark behaves, that show genuine critical input can register and change what gets shipped; and other evidence, like the disclaimer pattern, the engineer's doubt, and a chief executive's own admission that commercial pressure is crowding out philosophical deliberation, suggests it does not always do so, or does not do so where it matters most. The reasonable conclusion is not that philosophers in these roles have become hinkers. It is that the structural conditions under which they work make the drift toward hinker-work more likely than the institutions themselves would probably want, and that this is worth naming and watching for now, rather than either dismissing it because thoughtful individuals are involved or declaring it already settled on the strength of one troubling pattern.
A healthy institution needs someone able to say: perhaps we are asking the wrong question, perhaps the objective itself deserves scrutiny.
What philosophy has always contributed is not, primarily, the production of answers. Socrates is not remembered for having supplied Athens with a governance framework; he is remembered for refusing to let the city's unexamined assumptions pass as settled. Contemporary AI debates ask, reasonably, whether systems should be fair, transparent, accountable. Of course they should. But philosophy's distinctive move has always been to ask the question underneath that one: why is this system being built at all, what account of human flourishing is it quietly assuming, and which other accounts have been foreclosed without anyone noticing? These are not compliance questions. A society can build a perfectly auditable, thoroughly certified AI system and still have failed to ask whether it is heading in a direction worth going. Governance helps us navigate. Philosophy is what helps us ask whether we have the right destination, and confusing the two has a cost that compounds quietly, without ever showing up on a risk register.
A healthy institution needs someone able to say: perhaps we are asking the wrong question, perhaps the objective itself deserves scrutiny, perhaps the most affected people are not in this room. Such interventions rarely accelerate a product roadmap. They do not improve a quarterly earnings call. They are, nonetheless, close to the only thing that makes the rest of the apparatus worth trusting.
***
I have argued elsewhere, in my own work on deliberative AI governance, that legitimacy is not something a system acquires simply by complying with a set of pre-agreed principles. It comes from an ongoing public argument about what those principles should even be – from institutions that preserve structured disagreement rather than manufacture premature consensus. The same, I think, is true of philosophy's place inside the companies now hiring it. Its value was never in helping an organisation answer its ethical questions more efficiently. It was in making sure that the questions organisations, governments, and their own experts would prefer to treat as settled remain open enough to be asked again.
AI companies say they want philosophers for their capacity to question, and there is no reason to doubt that many of them mean it. The risk this essay has traced is narrower and more structural: that what an organisation can most easily recognise, reward, and act on is the capacity to answer, so that even sincerely wanted criticism finds it harder to register than an implemented framework does. Answering efficiently was never philosophy's contribution – identifying the question everyone else had already stopped asking was. There is no institutional design that could permanently remove this risk, and it would be a mistake to end by pretending otherwise. The point was never that philosophers should stay out of AI companies; the history rehearsed above suggests they never have stayed out of worldly institutions for long, and there is no obvious reason AI should be the exception. The point is to preserve the practice of philosophy inside them – to keep spaces where questioning is valued apart from what it implements, where dissent is treated as useful rather than merely tolerated, and where success is measured not only by the frameworks produced but by the quality of the questions those frameworks leave open.
Floridi may well be right that philosophers should keep one foot in the university. My argument has simply been that this cannot be sufficient on its own. The decisive question was never really where the philosopher works. It is whether the organisation around her preserves the conditions under which philosophy remains philosophy – and whether she, and the rest of us, keep noticing the difference between a thinker and a hinker for long enough that it continues to matter which one she is.
AI should not replace our reflection; rather, we should reflect more.
H. Mustafa Akyol is an independent researcher working at the intersection of artificial intelligence and philosophy. He holds separate bachelor's degrees in industrial engineering and philosophy, and has worked as an AI code reviewer and as a product manager. His writing addresses questions of responsibility, deliberation, and governance under increasingly capable AI systems; his chapter 'Deliberative AI Governance' is forthcoming in AI Regulation in Industry 5.0: Ethical Governance, Policy Frameworks, and Human-Centric Innovation (Springer Nature). He is currently at work on a book developing these arguments at greater length.
First published online 18th October 2026
If you enjoyed reading this, please consider becoming a Patreon member or making a donation. The Philosopher is unfunded and your support is greatly appreciated.
