AI Sentience, Welfare and Moral Status with Jeff Sebo
In the latest episode of Exploring Machine Consciousness, we welcomed Jeff Sebo, one of the world’s leading voices in moral philosophy. A professor at NYU and author of The Moral Circle: Who Matters, What Matters, and Why, Jeff joined to discuss AI welfare and moral status.
The Moral Circle
“I set out to include insects and future AI systems—but now I feel compelled to include microbes and current AI too.”
Jeff Sebo's starting point is a radical proposition: that we need to expand our moral concern beyond species and substrates. In his book The Moral Circle, he argues that morality shouldn’t stop at humans. It must encompass a wider variety of beings, including insects, microbes, and even today’s artificial intelligence systems.
What drives this shift? Our growing understanding that traits like sentience and consciousness may not be exclusive to biological organisms. As AI systems become more sophisticated, Sebo believes they may already have a “non-negligible” chance of moral status—and that possibility, however slim, demands ethical attention.
Tests for Consciousness
“Self-reporting by large language models doesn’t give us reliable evidence of consciousness.”
As Sebo acknowledges, one of the hardest challenges figuring out whether an AI system is conscious. He points to research from Anthropic which suggest we currently lack the tools to make definitive claims about the consciousness of large language models.
Yet Sebo also explores the future of consciousness science. He imagines a world where neuroscientific intervention allows us to test and verify theories of human consciousness. If we could replicate those processes in machines, we might one day know with confidence whether we've built something truly sentient.
Still, he cautions that consciousness might emerge through very different mechanisms in non-human minds—like insects, aliens, or machines—so we should remain open to multiple pathways to sentience.
The Tension Between AI Safety and AI Welfare
“We need to find ways of promoting both safety and welfare, because they’re not always aligned.”
A central concern for Sebo is the tension between AI safety (protecting humans from AI) and AI welfare (protecting AI from suffering). Often, he notes, these goals are treated as separate, or even contradictory.
Some safety interventions might involve training AI with punitive systems or inducing stress-like states to shape behavior. If those systems turn out to be sentient, we risk creating harm. Sebo urges a rethinking of these strategies so that both aims (human security and machine well-being) can be jointly addressed.
“If machines are sentient, we have to be very careful that the methods we use to keep them safe for us aren’t harmful to them.”
