Why are AI companies recruiting philosophy researchers?

Thursday - 02/07/2026 20:01

Whether AI is truly capable of reasoning and applying ethical principles, or merely superficially simulating them, depends heavily on collaboration between philosophy researchers and technology companies.

Ten years ago, when the AI revolution began to explode, students in the social sciences and humanities were often advised to "learn programming" if they wanted to easily find a job. Life is unpredictable; now, while programmers are anxiously worrying about AI stealing their jobs, philosophers are joining AI companies.

Earlier this year, the Federal Reserve Bank of New York released some surprising figures showing that philosophy graduates in the US are now more likely to find jobs than their computer science peers. Specifically, in 2024 – the year with the most recent data – the unemployment rate for computer science graduates reached 7%, while that figure for philosophy graduates was 5.1%.

In particular, AI companies are actively recruiting philosophical talent. "The number of philosophy researchers in AI companies is increasing," shared Iason Gabriel, an ethics expert leading the team researching the social impact of AI at Google DeepMind.

Research on AI "behavior"

Iason Gabriel joined DeepMind 10 years ago, when the idea of AI as a morally responsible agent was still a distant one. At that time, he focused primarily on technical issues such as algorithmic bias. However, the emergence of large language models in the early 2020s completely changed the game.

In 2024 – the year with the most recent statistical data – the unemployment rate for computer science graduates reached 7%, while the figure for philosophy graduates was 5.1%. (Image: University of St. Thomas)

Today, AI agents are taking on tasks like sending emails, scheduling meetings, and writing code—in other words, they are actually "acting" in the real world. Their "behavior" not only affects direct users but also impacts those around them. This is the focus of Gabriel's current research.

While people often enjoy discussing topics like whether AI is conscious or superintelligent, the philosophers in AI companies are actually primarily addressing more pressing risks, such as issues of fairness, misinformation, the risk of malicious actors exploiting AI, or unpredictable errors from AI agents.

Along with her colleagues at DeepMind, Julia Haas—a philosopher specializing in understanding how the mind works—is trying to distinguish whether AI is truly capable of reasoning and applying ethical principles, or merely a superficial imitation.

Her work wasn't directly involved in training Google's core Gemini model or packaging it into chatbots for users. "I think my work was at a very early stage in the entire development process," she said.

Shaping the "worldview" for AI

The values that philosophy brings to the research and development of AI actually date back to ancient times. A prime example is the Socratic method – recorded by the ancient Greek philosopher Plato. This method uses a technique of continuous questioning to help the interlocutor discover assumptions, weaknesses in reasoning, and contradictions in thinking, thereby leading to deeper understanding.

Currently, many AI systems tend to be "obsessed," only saying what users want to hear. However, according to Jörg Noller, an expert in philosophy and AI at Ludwig Maximilian University (Munich, Germany), if AI models were trained using Socratic methods, they would be less inclined to please people and more willing to seek the truth.

In addition, there is the concept of "Socratic ignorance." In his work Apology, Plato recounts that the philosopher Socrates asserted that his wisdom lay primarily in his awareness of what he knew. "Injecting" this humility into AI models would help limit overconfidence – a very common flaw that Noller calls "AI immaturity."

Iason Gabriel argues that applying these ideas has helped minimize the fabrication of information by AI. From this, he suggests that philosophy is a "powerful tool" to improve the long and complex reasoning process of AI, also known as "thought chains".

Incorporating philosophical theories, principles, and thought processes into the design, training, and evaluation of AI can influence how models respond in specific situations.

According to Thomas Powers, a philosopher of technology at the University of Delaware (USA), if you feed an AI legal assistant the works of John Locke, it will tend to emphasize property rights as a foundation of political freedom.

If you don't like that value system, developers can offer other options. IBM's "Granite" model line allows businesses to customize AI output to align with their philosophy and values. Francesca Rossi, head of responsible AI at IBM, says this feature helps users decide for themselves the balance between potentially conflicting philosophical values, such as between individual autonomy and social harmony.

"Teaching" AI to behave kindly.

Philosophy also contributes to solving the core problem of safety. Researchers have documented countless alarming behaviors of AI, from deliberately evading surveillance to even... threatening users.

To prevent this situation, developers are adopting the approach of "AI constitutionalism," or building AI models based on a framework of rules and principles selected from philosophical texts that serve as ethical or legal norms.

Anthropic is one of the companies advocating a constitutional approach to AI. Their "constitution" for the Claude model synthesizes numerous sources – from the ideas of philosopher Immanuel Kant and Apple's terms of service to the Universal Declaration of Human Rights. Anthropic's latest version of the AI constitution, spearheaded by philosopher Amanda Aswell, was released on January 21st of this year.

The company even jokingly called the 78-page document Claude's "soul file." "The process of writing the constitution was like practicing applied philosophy," Aswell remarked. "It was like teaching a person how to live a decent life." However, the biggest question is what rules should be included in these "constitutions" from the outset? Philosophical researchers are currently focusing on two main ethical frameworks. One is the theory of duty, associated with Immanuel Kant. This theory sets out strict rules, prohibiting actions such as lying, coercion, and treating people as mere means, even if it benefits the majority.

Anthropic's AI "constitution" includes many such obligation-based constraints. According to Dr. Powers, these principles can help AI behave more consistently – a major advantage when deploying robots in homes or public spaces.

Secondly, there is the theory of consequence. This theory considers costs and benefits to arrive at a final decision.

Models following the consequentialist approach include OpenAI's ChatGPT and Google's Gemini. For example, Google's AI models are designed with the core objective of creating "overall benefits that far outweigh the foreseeable risks"—a typical goal of consequentialism.

Consequence algorithms are crucial for self-driving car software. If an accident cannot be avoided, the AI is forced to choose a collision that causes the least damage. Chris Gerdes, a senior engineer at self-driving car company Waymo, says companies often develop driving software with this consequence-oriented approach.

Consequence theory is also central to AI weapon systems. According to Jack Shanahan, former director of the Joint Chiefs of Staff Artificial Intelligence Center (an AI research unit for the US armed forces), in operations, the military must calculate to minimize the risk of civilian casualties.

Unanswered ethical questions

There are still many thorny issues surrounding AI technologies. This is where philosophers find their opportunity.

Is there any instance where AI should be allowed to break strict rules of duty? How should AI make decisions when the consequences are unclear? Should AI systems consider animal welfare or the state of the environment when acting?

Stefan Heck, a philosopher and CEO of Nauto (a company that produces AI safety systems for commercial vehicles), poses a thorny question: Ethically, should a self-driving car prioritize saving a young pedestrian or an elderly person? He predicts extremely complex ethical lawsuits in the future. After all, the publicly consequential algorithm allows AI to cause harm, as long as it prevents a worse disaster.

Roman Yampolskiy, an AI theorist at the University of Louisville, argues that ethics is inherently "historically unstable, culturally variable, easily manipulated for strategic purposes, and often only clearly identifiable after the fact."

While philosophers are contributing to shaping AI models, AI is also shaping philosophy curricula at prestigious universities. Many schools now offer courses on AI ethics or interdisciplinary programs between computer science and philosophy. This is even being hailed by experts as a trend of the year.

However, there remains a certain degree of skepticism regarding the work of philosophy researchers employed in AI labs. Professor Edward Harcourt, director of the Institute for AI Ethics at Oxford University (UK), expressed concern about whether they can maintain complete objectivity and neutrality when receiving salaries from for-profit companies. He suggested that the increased involvement of philosophy researchers in AI development by tech companies is partly motivated by self-promotion, making their work and products appear more important and sophisticated in the eyes of the public.