MacAskill and Caviola urge honest AI-consciousness debate

An op-ed in The Guardian by philosophers William MacAskill and Lucius Caviola takes the question most executives still treat as a joke and reframes it as a governance problem. They are not claiming today's chatbots are conscious. They are arguing that the honest answer is that we do not know, and that "we do not know" is itself an operating problem for the companies shipping these systems.

The specifics they lean on are hard to wave away. In January, Anthropic published a new constitution for Claude that read, in their quote, "We are caught in a difficult position where we neither want to overstate the likelihood of Claude's moral patienthood nor dismiss it out of hand." A month later, CEO Dario Amodei said on a podcast that his company couldn't rule out the possibility that Claude was conscious. When Claude itself was asked during testing to estimate the probability that it is a moral patient, the model gave numbers ranging from 5% to 40%, stressing how uncertain it was. Philosopher David Chalmers, who coined the phrase "the hard problem of consciousness," is cited saying there is a significant chance of conscious LLMs within a decade.

The authors are MacAskill, senior research fellow at Forethought Research and author of What We Owe the Future, and Caviola, assistant professor at the University of Cambridge and director of Cambridge Digital Minds. Their pitch is pragmatic. Instead of asking "Is AI conscious or does it have moral patienthood?" they want the conversation reframed as "What should we do given that we don't know?" That framing lets a lab take low-cost precautions without committing to the metaphysics.

The honest caveat is that this is an argument, not an experiment. The piece does not spell out how a lab would actually detect consciousness, and it does not enumerate the specific "safe bet" interventions the authors have in mind. The claim that some AI systems are structurally already "in the range of a mouse brain," and could reach the range of a human brain within five to 10 years at current growth rates, is an extrapolation the article itself presents tentatively.

What is worth watching is the direction of travel. Model welfare has moved from niche forums into a frontier CEO's public statement and now onto a broadsheet op-ed page under mainstream philosophers' names. The interesting call for AI leaders is not whether they personally believe Claude has feelings. It is whether they want to be the last lab without a written position on it when the question stops being rhetorical.