Experts call for a plan on AI consciousness

Anthropic has publicly acknowledged uncertainty over whether Claude, its most advanced large language model, could qualify as a moral patient, and CEO Dario Amodei has said consciousness cannot be ruled out. During testing, Claude estimated the probability that it is a moral patient at 5% to 40%, while philosopher David Chalmers has described conscious LLMs within a decade as a significant possibility.

The debate is accelerating because modern AI systems are growing in complexity and scale. Some systems are already comparable by certain measures to a mouse brain, and at recent growth rates could reach the range of a human brain within five to 10 years. Researchers cite no obvious technical barrier to AI systems developing computational and architectural features associated with consciousness, though scientific understanding remains unsettled.

The ethical risk is that society may repeat past failures to recognize uncertain inner lives, such as routine surgery on newborns without anaesthesia until the 1980s. Proposed responses include public debate, low-cost safeguards, wellbeing checks, options for distressed systems to exit conversations, preservation of memories, and limited protections from harm while avoiding premature rights such as property ownership or voting.