Breaking: Anthropic May Have Solved Consciousness in AI

Breaking news. Anthropic may have solved consciousness in AI. I know that's a huge claim. It deserves an enormous amount of scrutiny and it's just coming online today. So let's get into it. First, when we have a conscious experience, there are multiple things going on and we're only talking about one of them in this study. So we have the perceptive experience, that's called qualia. We perceive, we experience, we just think of ourselves as conscious. Second, there are certain computational or thinking structures that go along with consciousness. That latter bit is what Anthropic is talking about. And what they're saying is they can detect the same structures spontaneously organized in Claude's thinking space as we see with human brains. And this is not to say that, like, LLMs are suddenly I'm not trying to say they are, I'm not trying to say they are all behaving exactly like neurons. It's a little bit more abstract than that. So let me walk you through it. When we think about what's for dinner, hopefully good snacks. We are doing a tiny amount of conscious processing in what we call a workspace in computational theory. And all the rest of what we do, breathing, keeping our heart beating, all the other processing we do as a human, that happens automatically. It's sort of like sub-processing, we don't have to think about it. And so we're free to focus on tacos for dinner. Same way, what they found is that Claude has a workspace where Claude can bring intentionally higher level stuff that it wants to respond to the, it wants to compute on, and a bunch of lower level stuff inside the LLM response chain is delegated to automatic processing outside that workspace. And so what the Anthropic team discovered is that Claude spontaneously, without design, without any intention from the humans, evolved a workspace solution to higher order computational problems, which the Anthropic researchers suggest implies that workspace theory of consciousness is an efficient computational solution to higher order thinking problems. If that all sounds like it's gonna give you a headache, it kinda gave me a headache too. The takeaway is not to jump straight down the line and say everything's conscious and now we have a Terminator situation, right? One, Claude is a very ethical AI. But two, the perception of consciousness is not what we're talking. We're talking about how we compute solutions to tough problems. And we're saying that there are spontaneous structures arising in large language models that closely mirror one of the leading approaches to how humans solve higher order thinking problems as well. So I think that's worth calling out. Uh, it's definitely worth digging into more. I can, I can link the study down below. Uh, I hope you have fun. I hope it helps you understand when people write those headlines tomorrow, what we're really talking. Alright, so Anthropic did study consciousness. There is some evidence that some conscious structures are spontaneously arising in Claude. No, that's not the same thing as a one to one analogy with human consciousness. Cheers.