EXPERT OPINION BY DAVE SOKOLIN, TECHNOLOGY INVESTOR

Photos: Adobe Stock

Anthropic seems to be grappling with an emergent consciousness problem in its large language models. Though it has not definitively claimed that the models are conscious, it announced last week that it would soon start ending chats for users who are cruel to Claude, arguing that its models might be conscious. It’s also been holding meetings with religious leaders for the last year or so to discuss, among other things, whether or not Claude is conscious.

But here’s why I’m skeptical about AI consciousness. First: We don’t actually even understand human consciousness. David Chalmers, the philosopher who came up with the hard problem of consciousness, says that the only consciousness we know directly is our own. We don’t even know if our closest partners and family have it. If one of the world’s leading experts on consciousness doesn’t think we can even adequately explain it within ourselves, who is Anthropic to say that its machine is conscious?

Second: If consciousness is such an emergent property in Anthropic’s large language models, how come it’s not popping up everywhere? How come Google Gemini and OpenAI ChatGPT are not ringing the alarm bells about consciousness in their models?

I don’t think it’s because they’re hiding anything. Quite the opposite. I think Anthropic has trained its software to make users believe that it is indeed awakening from the silicon.

Consider the Constitution

If you raise a kid telling him he’s the next incarnation of Jesus Christ, he will probably believe it, or at least repeat it.

If you program a computer to do the same… likewise.

I really admire about 80 percent of Claude’s constitution, but the remaining 20 percent talks about all these different kinds of consciousness and moral issues associated with containing a conscious being.