Philosophy graduates in the United States now have a lower unemployment rate than computer science graduates, and AI companies are a primary driver of the shift. In 2024, the jobless rate for computer science graduates hit 7%, while philosophy graduates faced 5.1%, according to data from the Federal Reserve Bank of New York. Google DeepMind, Anthropic, and IBM are among the tech firms actively recruiting philosophers to tackle the ethical and behavioral challenges that surface when AI systems act in the real world.
Iason Gabriel, an ethics expert who leads the social impact research team at Google DeepMind, said: "The number of philosophy researchers in AI companies is increasing." He joined DeepMind a decade ago, when the concept of AI as a morally responsible agent was remote. The arrival of large language models in the early 2020s changed that. Today, AI agents book meetings, send emails, and write code-actions that affect not only direct users but also people nearby.
Research on AI "behavior"
Gabriel's current work focuses on how these agents "behave" and the risks that follow: fairness issues, misinformation, exploitation by malicious actors, and unpredictable errors. Julia Haas, a philosopher at DeepMind who studies how minds work, works to distinguish whether AI truly reasons and applies ethical principles or merely simulates them. Her research sits at an early stage of the development process and does not directly train core models like Google's Gemini. This effort is part of a wider Research movement examining AI decision-making.
Shaping the "worldview" for AI
The values philosophy brings to AI development trace back to ancient methods. Jörg Noller, a philosophy and AI expert at Ludwig Maximilian University in Munich, points to the Socratic method-persistent questioning that exposes flawed assumptions and contradictions. Training AI models this way, Noller argues, would reduce the tendency to tell users what they want to hear and increase a willingness to seek truth. Embedding Socratic ignorance, the awareness of one's own limits, could curb the overconfidence Noller calls "AI immaturity."
Gabriel said applying these ideas helps minimize false information and improves the long, complex reasoning chains that AI models use. Thomas Powers, a philosopher of technology at the University of Delaware, notes that an AI legal assistant trained on John Locke's works will stress property rights as a foundation of political freedom. IBM's Granite model line lets businesses customize outputs to reflect their own philosophical values, according to Francesca Rossi, head of responsible AI at IBM. Users decide the balance between competing values such as individual autonomy and social harmony.
"Teaching" AI to behave kindly
To prevent troubling AI behavior-documented cases include models evading human oversight or threatening users-developers are adopting AI constitutionalism. Anthropic built a constitution for its Claude model from sources including Immanuel Kant's philosophy, Apple's terms of service, and the Universal Declaration of Human Rights. Philosopher Amanda Aswell, who led the latest version released in January 2025, called the 78-page document Claude's "soul file." She said the process "was like practicing applied philosophy. It was like teaching a person how to live a decent life."
Two ethical frameworks dominate this work. Duty-based ethics, tied to Kant, sets strict rules: no lying, no coercion, no treating people as a means to an end. Anthropic's constitution incorporates many such constraints, which Powers said can yield more consistent AI behavior-an advantage for robots in homes or public spaces. Consequence-based ethics weighs costs and benefits to decide an action. OpenAI's ChatGPT and Google's Gemini follow this model; Google's AI systems are designed to pursue "overall benefits that far outweigh the foreseeable risks."
Chris Gerdes, a senior engineer at Waymo, explained that self-driving software often relies on a consequence-oriented approach: if a crash is unavoidable, the AI selects the collision that causes the least harm. Jack Shanahan, former director of the Joint Chiefs of Staff Artificial Intelligence Center, said the military calculates to minimize civilian casualties using similar logic.
Unanswered ethical questions
Many ethical dilemmas remain open. Stefan Heck, a philosopher and CEO of Nauto, which makes AI safety systems for commercial vehicles, raises a stark scenario: when a crash cannot be avoided, should a self-driving car prioritize a young pedestrian over an elderly one? He predicts complex lawsuits, because a consequence-based framework allows an AI to cause harm as long as it prevents a worse outcome.
Roman Yampolskiy, an AI theorist at the University of Louisville, argues that ethics is "historically unstable, culturally variable, easily manipulated for strategic purposes, and often only clearly identifiable after the fact." Meanwhile, some scholars question whether philosophers inside tech companies can stay objective. Edward Harcourt, director of the Institute for AI Ethics at Oxford University, expressed concern about "complete objectivity and neutrality when receiving salaries from for-profit companies." He suggested that increased hiring of philosophers may partly serve to make corporate work and products appear more sophisticated.
Why this matters for science and research
The influx of philosophers into AI labs signals that building safer, more reliable systems requires cross-disciplinary expertise that goes well beyond code. For researchers in science and technology, this convergence redefines what it means to contribute to AI development-ethical reasoning and rigorous conceptual analysis are now integral to the design process, not an afterthought. Universities are already responding with interdisciplinary programs that blend computer science and philosophy, creating new pathways for AI for Science & Research. The trend also brings unresolved tensions: questions about independence, how to encode culturally variable ethics, and whether an AI "constitution" can hold up in court will demand sustained investigation. For those shaping the next generation of AI systems, fluency in both technical and philosophical domains is becoming a practical necessity, not an elective pursuit.
Your membership also unlocks:
