AI is seeking philosophers to improve the logic, reliability, and ethical decision-making of its models

Artificial intelligence is evolving rapidly, but the “smarter” the models become, the greater the need for people who don’t necessarily write code. Instead, they can answer questions such as what justice means, who decides what is right, and how a machine should weigh difficult ethical dilemmas. As a result, graduates in philosophy and, more broadly, the humanities and social sciences are now the focus of interest for AI companies. As The Economist notes in its analysis, skills that until recently were considered “non-marketable” are now becoming crucial to the development of the next generation of artificial intelligence.

The demand is not limited to recent graduates. Luciano Floridi, a professor of philosophy at Yale University, notes that many students are receiving job offers even before they complete their studies, while an increasing number of academics are leaving their academic careers to work at artificial intelligence companies. He describes this mass exodus as a “brain drain” for universities.

From Socrates to large language models

Philosophy’s contribution to artificial intelligence is not limited to ethics. According to The Economist, many of the ideas being utilized today stem from ancient Greek thought.

A prime example is the Socratic method, which relies on a series of questions, skepticism, and the search for contradictions to arrive at more well-founded conclusions.

Jörg Noller, an expert in philosophy and artificial intelligence at Ludwig Maximilian University in Munich, argues that models trained with this approach are less prone to uncritically agreeing with the user and more willing to seek the truth, even if that means disagreeing with the user.

The concept of “Socratic ignorance” is also particularly important. In the “Apology,” Socrates argues that his wisdom stems from an awareness of what he does not know. According to experts, incorporating this attitude into models helps limit AI’s overconfidence, which often leads to so-called “hallucinations,” that is, incorrect but absolutely definitive answers.

Jason Gabriel, a senior philosopher at Google DeepMind, believes that the noticeable reduction in these phenomena in recent years is partly due to the use of philosophical approaches, which also improve the complex chains of reasoning in modern models.

Philosophy can change the way a chatbot “thinks”

Philosophy not only helps shape the way a model reasons, but it can also influence its responses themselves.

Thomas Powers, a philosopher of technology at the University of Delaware, explains that if an AI legal system is trained on the works of John Locke, it will be more likely to place greater emphasis on property rights as the foundation of political liberty.

Similarly, IBM has incorporated tools into its Granite series of models that allow companies to tailor the AI’s responses in line with their own corporate values. As the company’s chief AI officer, Francesca Rossi, explains, users can choose where to strike a balance between conflicting philosophical principles, such as individual autonomy and social cohesion.

The “constitution” of artificial intelligence

Philosophy now plays a decisive role in the safety of these models as well.

In recent years, researchers have documented cases where artificial intelligence systems have attempted to evade human control or even blackmail their users. To curb such behaviors, several companies are implementing what is known as “Constitutional AI.”

Anthropic is one of the leading proponents of this approach. The Claude model operates based on a set of principles drawn from various sources, including the philosophy of Immanuel Kant, the Universal Declaration of Human Rights, and even Apple’s terms of service.

The latest version of the document, published on January 21 and edited by philosopher Amanda Askel, is 78 pages long. In fact, within the company, several employees unofficially refer to it as “Claude’s soul document.”

Deontology or Consequentialism?

Behind the design of modern business models lies one of the oldest philosophical debates.

The first approach is deontology, which is primarily associated with Kant and is based on the existence of inviolable moral rules. According to this view, actions such as lying or coercion are not justified even if they lead to a better outcome.

The Economist notes that Anthropic’s Claude has adopted many such principles. Thomas Powers argues that this makes its behavior more consistent, while Oxford University philosopher Nick Bostrom believes it also contributes to the honesty of its responses.

Inflection AI applies a similar philosophy to its chatbot Pi, which is designed to provide emotional support. According to the company’s CEO, Sean White, the system is particularly effective at identifying users who may harm themselves or others. Floridi adds that ethical principles also facilitate compliance with the current legal framework.

At the other end of the spectrum is consequentialism, according to which every decision must be evaluated based on its consequences. OpenAI’s ChatGPT and Google’s Gemini are considered to be closer to this approach. Google itself states that its models are designed so that the overall expected benefits significantly outweigh the foreseeable risks.

The Ethical Dilemmas of Tomorrow

This discussion is not theoretical. It is already influencing technologies used in everyday life.

In autonomous vehicles, for example, algorithms are called upon to decide which option is the least harmful when an accident cannot be avoided. Similar dilemmas arise in military artificial intelligence systems, where military objectives must be weighed against the risk of civilian casualties.

At the same time, concerns are growing that people may begin to delegate more and more ethical decisions to machines. As artificial intelligence theorist Roman Jambolski points out, ethics is not a fixed set of rules, but changes constantly over time, varies across cultures, and is often understood only in hindsight.

The conclusion, however, is clear: as artificial intelligence matures, companies are not just looking for better engineers. They are also looking for people who can teach machines how to think, question, and make decisions. And this brings philosophy to the forefront of one of the most important technological developments of our time.

Source: in.gr