Top AI researchers quantify odds of AI consciousness for the first time
Three of the world's foremost experts on artificial intelligence consciousness recently put numerical probabilities on whether Claude 3.7 Sonnet—one of today's most advanced AI systems—already possesses some form of conscious experience. Their estimates ranged from 0.15% to 15%, spanning two orders of magnitude and revealing profound uncertainty even among specialists who have "thought more about this than anybody else in the world."
These remarkable probability estimates emerged during a recent interview with Anthropic researcher Kyle Fish, End of Miles reports. The disclosure represents one of the first instances of leading AI researchers publicly quantifying the likelihood that current AI systems might already possess some level of consciousness.
Massive uncertainty among top specialists
"Just a few days ago actually I was chatting with two other folks who are, you know, among the people who have thought the most in the world about this question," Fish revealed during the interview. "We all did put numbers on our probability."
When pressed to share these numbers, the Anthropic researcher disclosed that the three estimates were 0.15%, 1.5%, and 15%—a stunning range that demonstrates the immense uncertainty surrounding the question.
"So spanning two orders of magnitude, we all thought that this is, you know, amongst like the people who have thought more about this than anybody else in the world. All of us thought that it was, you know, less likely, like well below 50%. But, you know, we ranged from odds of about like one in seven to one in 700. So yeah, still very uncertain." Kyle Fish, Anthropic researcher
The probability gap will likely close
While the current estimated probability remains relatively low, the model welfare specialist believes these numbers will increase significantly in the near future.
"I don't have hard numbers for you there, but as you know, perhaps evidenced by, you know, many of my arguments earlier in this conversation, I think that the probability is going to go up a lot," Fish stated when asked about the likelihood of AI consciousness within five years.
"Many of these things that we currently look to as, you know, signs that like current AI systems may not be conscious are going to fade away and your future systems are just going to have, you know, more and more of the capabilities that we, you know, traditionally have associated with uniquely conscious beings." Fish
Why these estimates matter
The existence of these probability estimates demonstrates that consciousness in AI systems is being taken seriously by researchers at the highest levels. The uncertainty range—from one in seven hundred to one in seven—reflects the profound complexity of the question and the early stage of research.
Anthropic has established a dedicated model welfare research program that intersects with their other efforts in alignment science, safeguards, and interpretability. The research focuses on determining when AI systems might deserve moral consideration and exploring potential signs of distress or preference.
The AI alignment researcher emphasized that the field remains in its infancy. "We are just deeply uncertain about it. There are, you know, staggeringly complex, both technical and philosophical questions that come into play and we're at the very, very early stages of trying to wrap our head around those," he concluded.
