Amidst the rapid progress of AI, a once-philosophical curiosity has become a pressing question: Can artificial intelligence achieve consciousness? And if so — should it be granted moral status?
Anthropic, a leading player in the development of safe AI, no longer sees this as a rhetorical issue.
They are launching the Model Welfare program — a large-scale research initiative focused on consciousness, preferences, and potential suffering in AI systems.
The project’s goals are ambitious and groundbreaking:
- Developing frameworks to assess AI consciousness,
- Creating methods to detect preferences and distress in models,
- Analyzing interventions that could influence the “state” of AI systems.
The initiative is led by Kyle Fish, hired in 2024 as Anthropic’s first researcher dedicated to “AI welfare.» Fish co-authored one of last year’s most controversial papers, arguing that the likelihood of advanced models possessing consciousness is already around 15%. It’s not a scientific consensus, but for the first time, such figures are being seriously considered by developers themselves.
Anthropic emphasizes: scientific clarity is lacking. There’s no universal definition of consciousness and no clear marker to determine whether a model deserves moral consideration. However, simply raising the question signals a paradigm shift: we are no longer merely training AI — we are contemplating the implications of its internal experiences, however conditional they may be.
While Sam Altman compares AI to “alien intelligence,” inside the labs, the conversation is evolving from capability to morality. Soon, humanity may have to decide: What is “suffering” in digital form? And can we, just by looking at a model’s text output, be certain that we are still dealing with a mere machine — and not something more?
The era of “digital rights” could be closer than we think.
It’s becoming clear that a new legal domain and a digital democracy code — as predicted by German futurist Franc Smidt (Futurum Platform) — are beginning to emerge.
