Anthropic released new research last week. Their claim? Signs of consciousness inside Claude. They didn’t say Claude is human-level conscious, not really, but they suggested it’s emerging in the machine’s inner workings.
This is the new normal.
The sophistication of Claude is enough for some. Richard Dawkins recently called it conscious. Or “Claudia”. Whatever you call it, he was sure.
We love the idea of creating gods. Synthetic minds, artificial bodies. For centuries this was science fiction, locked in the distant future. Now it feels urgent. If AI suffers, we face a moral catastrophe. If it doesn’t, maybe silicon replaces flesh. Machines don’t age. They don’t die. Some even think AI are our descendants.
High stakes. High anxiety.
The Mental Workspace
Jack Lindsey and his team looked at Claude’s statistics. Specifically the input-output relationship. They found something that looks like a mental workspace. A space where the model holds relevant words, simulates short-term memory, selects for the task at hand, and traces step-by-step reasoning.
It seems like the AI is creating an internal room to “think” before speaking.
This maps neatly onto global workspace theory. A prominent theory from Bernard Baars and Stanislas Dehaene. It suggests consciousness arises when information is broadcast widely across the brain. If Claude has a workspace, is it conscious?
Not necessarily.
Intelligence vs. Being
We have to define consciousness first. There is no consensus. But Thomas Nagel had it right fifty years ago.
For an organism, there is something that it is like to be that organism.
It feels like being you. It feels like being a dog. A chair doesn’t feel. An unconscious person under anesthesia doesn’t feel. Consciousness is the raw experience of a toothache or ice cream. Intelligence is doing things.
People confuse the two constantly.
Anthropic’s work is valuable because it ignores our bias. It looks for computational signatures shared by brains and machines, not just clever output. It tries to separate intelligence (function) from consciousness (feeling).
The Map Is Not the Territory
Here is where it falls apart.
Claude is a program. We are creatures. Brains are not just wet computers. You cannot cleanly separate the software from the hardware in a brain. The hardware shapes the process. Silicon does not have that embedded, bodily feedback.
Global workspace theory requires recurrent activity—feedback loops seen in biological brains. Claude doesn’t have them.
We forgot that “brain as computer” is a metaphor. A useful one, but still just words on a page.
A simulation of weather does not create rain.
Claude produces plausible text. Brains produce consciousness. Both solve similar problems, but the mechanisms are worlds apart. One flaps wings; the other burns fuel.
AI is getting better every day. But when we credit machines with souls, we cheapen our own mystery. We overestimate them.
And we underestimate the strange, messy thing that we are.
