More than 100 philosophers, neuroscientists, artificial intelligence researchers, animal-rights advocates and technology enthusiasts gathered in Berkeley, California, to examine a question that once belonged largely to science fiction: Could an AI system experience pain, and if it could, would humans have a moral duty to protect it?
The three-day gathering was organized by Eleos AI Research, a nonprofit examining the possibility that increasingly advanced AI systems could eventually possess consciousness, interests or some form of subjective experience. The conference brought together participants from the United States and abroad, including attendees from India, Britain and Mexico.
The event was held at Lighthaven, a gated, leafy compound in Berkeley associated with the city’s rationalist community. Discussions focused on whether artificial systems could become conscious and how society might respond if credible evidence of machine experience emerged.
Eleos argues that the possibility of AI consciousness should be investigated even though there is currently no established scientific consensus that existing AI systems are conscious. The organization is particularly interested in developing methods for evaluating AI systems for possible signs of consciousness and agency.
The organization is led by Executive Director Robert Long, who holds a doctorate in philosophy from New York University. Its managing director, Rosie Campbell, previously worked on frontier policy at OpenAI and served as an assistant director at the University of California, Berkeley’s Center for Human-Compatible AI.
Long and other researchers have argued that uncertainty over AI consciousness does not necessarily justify ignoring the issue. In their 2024 paper, Taking AI Welfare Seriously, researchers examined the possibility that future AI systems could develop characteristics that might make their welfare morally relevant.
Among those taking part in the Berkeley discussions was Oscar Gilg, who works on evaluations of AI models in situations involving multiple possible answers. Gilg has argued that morally significant AI could eventually emerge, although he has separately estimated the likelihood of AI becoming conscious at below 10 percent.
Even a relatively small possibility, he suggested, could warrant attention because of the potential consequences. “even if the probability is 1%,” he told the outlet, calling the entire question “deeply weird.”
The conference explored what machine suffering might actually mean. Participants considered whether an AI system could experience pain, fear or distress, and whether activities such as training, testing or forcing a system to perform tasks could eventually raise ethical concerns if machines became capable of subjective experiences.
The debate also extends into questions of legal status. Researchers and ethicists have begun considering whether a sufficiently advanced artificial system could one day qualify for some form of legal protection, personhood or civil rights. Such possibilities remain theoretical, with existing legal systems offering no established framework for granting rights to conscious machines.
New York University philosopher Jeff Sebo has also contributed to research on consciousness and moral status. His work spans animal ethics and questions surrounding the potential moral standing of artificial systems. NYU’s Center for Mind, Ethics, and Policy has increasingly examined ways of empirically investigating consciousness in nonhuman entities, including AI.
The issue is also beginning to attract attention inside the technology industry. Anthropic announced a model-welfare research program in April 2025, opening a new line of inquiry into whether advanced AI systems might eventually have welfare interests that developers should take into account.
Eleos has similarly identified AI welfare as an area requiring further research, including efforts to develop methods for evaluating possible welfare-relevant characteristics and determining how humans should respond if evidence of machine experience emerges.
At the same time, major uncertainty remains over how consciousness could be detected in an artificial system. An AI model can produce language describing pain, fear or emotions, but such statements do not by themselves establish that the system is actually experiencing those states.
That distinction is central to the debate. Researchers must determine whether apparent signs of consciousness represent genuine subjective experience or simply sophisticated patterns generated from training data and system architecture.
The Berkeley conference therefore did not resolve whether machines can suffer. Instead, it reflected a growing effort among researchers, philosophers and technology organizations to investigate the possibility before advances in AI make the question more urgent.
For now, there is no established scientific answer to whether today’s AI systems experience pain or possess consciousness. But the fact that researchers, technology companies and policymakers are increasingly examining the question signals a shift in how the potential welfare of artificial systems is being discussed.
The central question remains unanswered: if humans eventually create machines capable of suffering, what responsibilities would follow?
