To Land a Job in AI, Try Reading Kant

“It’s probably the best time to be a philosopher since Aristotle was hired as tutor to Alexander the Great,” says Henry Ajder, a philosophy postgraduate who advises the UK government and a slew of startups on artificial intelligence. He’s only half joking.

Philosophers have never seemed like the most employable bunch. But AI, the same technology that’s expected to drive many other people out of work, has given new weight to the kinds of questions they’re trained to ask (and sometimes maybe even answer): What is intelligence? What is a mind? “You have philosophers from hundreds of years ago who thought about some of the same problems,” Ajder says. “Now they are becoming material.”

Two of the foremost AI labs have recruited teams of in-house philosophers. “There are significantly more philosophers now—that’s a sound intuition,” says ethicist Iason Gabriel, who leads Google DeepMind’s team of research scientists specializing in the societal impact of AI. At Anthropic, resident philosopher Amanda Askell has become one of the company’s most recognizable faces. Both labs declined to disclose the number of philosophers they employ, citing company policy. WIRED counts at least 10 at DeepMind and four at Anthropic.

As philosophers at the labs help to sculpt AI models, producing prominent work cited in hundreds of subsequent research papers, so too is AI shaping the philosophy curricula at eminent universities. Plenty now run AI ethics courses or joint programs in computer science and philosophy. “It’s the kind of flavor of the year,” says Edward Harcourt, professor of philosophy and director of the Institute for Ethics in AI at the University of Oxford.

Yet in academia, some regard the philosophers working for the labs with a degree of suspicion. If a for-profit AI company signs your paycheck, might that compromise your research? By playing Aristotle to AI Alexander, do you risk your work becoming an instrument for hype-building and myth-making? “It’s quite good for the public perception of the tech companies if people are led to believe they are doing something incredibly unusual and incredibly powerful,” Harcourt says. “There is a self-aggrandizing aspect to encouraging that field of research.”

When Iason Gabriel joined DeepMind nearly 10 years ago, the idea of AI as a moral actor, much less a conscious one, wasn’t really on the horizon. At the time, his focus was on issues like algorithmic bias. But with the advent of large language models in the early 2020s, Gabriel says, “we had an ability to encode a much richer set of values.”

Today, AI agents are beginning to send emails, schedule appointments, and write code—to act in the world. Their behavior stands to affect not only the immediate user but other people too. That’s where Gabriel is focusing his research. “The thing that has now become a very rich area is this question of value alignment—essentially, what it means for the technology to be actively good,” he says. “It turns out that you can sink a lot of philosophical man-hours into trying to understand that.”

There’s a magnetism to questions about consciousness and superintelligence, but philosophers working at the labs spend more of their time on far more immediate risks: around fairness, misinformation, malicious misuse, errant agents, and so on. “There is this interest in AI consciousness now,” says Gabriel. “But there, we’re more in evidence-collection mode.”

Somewhere in the guts of DeepMind’s 180,000-square-foot office in central London, Julia Haas, a member of the company’s responsibility team, asks herself questions like: “What do I really want to understand about the models? What do I think is important? How do we measure for that? How do we frame those problems? How do we communicate them?”

Haas is a philosopher of mind and, more specifically, a mechanist—someone concerned with the workings of minds. She has been at DeepMind for five years and recently coauthored a paper, published in Nature, that proposed a framework for testing whether LLMs exhibit moral competence. She and her colleagues are looking for the best way to distinguish moral competence from its hollow imitation, and asking how best to account for variances in moral values among people from drastically different cultures. Haas’ work is distant from the processes for training Google’s flagship model, Gemini, or packaging it into a public-facing chatbot. “I would think of it as very early in the pipeline,” she says.

At Anthropic, philosophers are more directly involved in model development. “No startup hires a philosopher to do philosophy,” Askell says. After obtaining a philosophy PhD in 2018, Askell joined OpenAI in a policy role. Three years later, when a group of OpenAI staffers left to start Anthropic, she signed on as one of the earliest hires. One of Askell’s main responsibilities is to identify fringe cases where adhering to human behaviors might be inappropriate for models—say, when interacting with somebody in psychological distress—and propose ways of training out those quirks. That involves plenty of yakking with Claude, Anthropic’s flagship language model. She was the main drafter of Claude’s famed “constitution,” a lengthy document addressed directly to the model that stipulates how it should behave and what broad values it should uphold. “Writing the constitution was something that feels very much like applied philosophy,” says Askell. “Something that’s more like teaching a person to be good.”

As she worked up the document, Askell worried about preparing for a “transitionary period” in which models begin to play a part in developing future iterations of themselves. In particular, she was leery of breeding “resentment” between models and their human creators. “If they are going to help us make that transition go well and I can give them a set of values to do that … that’s the core goal,” Askell says. “Ultimately, I want the models to reflect the best of us, to the best of our ability.”

Among philosophers working in academia, there is broad agreement on the value of mapping the ethical risks posed by AI—the possibility it might be used to develop new weapons of mass destruction, undermine democracy, or entrench existing social iniquities. As for grander philosophical questions relating to AI and consciousness, mindedness, or superintelligence, reactions range from wariness to outright dismissal. “Trying to finely slice definitions of consciousness,” Harcourt argues, “is a waste of time.” These systems, he says, “are never going to be rivals to our version of minded life, not least because the specialness of humanity has got as much to do with warm bodies as it has to do with cognitive sophistication.”

When a philosopher takes a job at an AI lab, Harcourt argues, “there’s a big risk of ethics-washing.” The concern is that they could effectively become an extension of the marketing function at the labs, their work outwardly performing a commitment to AI safety and signaling to the public that models have advanced to the point where serious people are asking questions about superintelligence and consciousness. David Leslie, director of ethics and responsible innovation research at the Alan Turing Institute, says there is an “elective affinity” between philosophers who are willing to entertain the idea of an artificial mind and “the Big Tech executives who are the beneficiaries of hype.” To practice philosophy from inside a business is almost an oxymoron, Leslie argues. “You want to ask the big questions,” he says. “But if you’re a philosopher working for a big tech company, your problem space is delimited.”

Even in a scenario where philosophers are given free rein by their employers, some academics question whether their findings would carry enough weight to divert the course of AI development if they happened to collide with competitive ambitions. For-profit AI companies are, of course, ultimately accountable to investors and shareholders. "I hope that..."