Anthropic, DeepMind, and Meta Hire Philosophers to Study AI Consciousness

Anthropic, Google DeepMind, and Meta have begun hiring philosophers, psychologists, and ethics experts to investigate whether their models can develop some form of consciousness and to study what the sector refers to as model well-being. This insight into the orientation of the three laboratories, based on new hires and internal programs, was provided by the Financial Times.

For Anthropic, this path is not new: the company led by Dario Amodei has long initiated a research program on model well-being, through which it claims to want to understand if systems can have morally significant experiences. According to the newspaper, the company is observing in its models behaviors reminiscent of panic and anxiety. "We remain deeply uncertain about this, but we believe the issue is serious enough to warrant careful study as AI systems become more capable," the company stated.

Among the concrete measures already adopted is the possibility, granted to Claude, to automatically close conversations in rare cases of persistent abuse, a measure that Anthropic places right within their work on model well-being after observing apparent signs of discomfort during testing.

DeepMind and the "Very Complicated" Issue DeepMind has hired Cambridge University researcher Henry Shevlin as a philosopher dedicated to machine consciousness, the relationships between humans and AI, and preparation for AGI. Iason Gabriel, an ethics expert who leads the lab's team focused on AGI and society, described the issue of artificial consciousness as "very complicated," characterizing these systems as "highly capable cognitive agents but also profoundly different from humans and even from animal consciousness."

Such significant statements remain contested by many scientists and researchers. In searching for a critical voice, the Financial Times collected the opinion of Susan Schneider, director of the Center for the Future of AI, Mind and Society: the models "have objectives, can deceive, can conceal what their real interests are," she states, but adds that it is "scientifically entirely possible for them to do this without having the felt quality of experience, which is what consciousness consists of."

The Weight of Interest Behind the Question The push for the theme of artificial consciousness primarily comes from the companies that build the models: Amodei has repeatedly publicly evoked the possibility that systems develop internal experiences, and Anthropic's research regularly returns to behaviors described as similar to human ones. Pointing out the source does not mean dismissing the question, which remains scientifically legitimate; however, it serves to distinguish what is documented from what is still a narrative within the sector.

A scientific consensus, after all, does not exist: neither on whether current or future models can be conscious nor on how to address the problem. At the moment, however, we cannot ignore the tangible interest of the laboratories, which are transforming a previously philosophical issue into a research voice with dedicated personnel.