#205 - Prof. Moran Cerf - "Did We Accidentally Build A Conscious AI?"

**Moran Cerf** (0:00)
Every organism that tries to survive, learn to manipulate and essentially find ways to trick nature. And if that's true, then they eyes enemies are us. We're the container that it needs to learn to manipulate and get out of the box so it can do better than us. You prompt Claude to tell you something about document. Like, say summarize document for me while thinking of elephants. In the output, it looks the same. If you ask it to summarize document, or summarize document while thinking of elephants, you get the same output. But when you ask summarize an elephant, you kind of see which sites in the Claude brain are active when it does something in parallel, and then you see a perfect mapping to the human brain.

**Peter McCormack** (0:39)
So have they accidentally replicated the human brain in AI?

**Moran Cerf** (0:44)
I would say they did, but not accidentally. What they did is they took the thing that is the crown of humanity if you want, language, we gave that to AI as a given, and now it's doing the same thing to us. AI in the last two months made a huge leap in ability to manipulate and lie. And that's another extension of it learning to become conscious or to simulate one.

**Peter McCormack** (1:06)
This show is brought to you by my lead sponsor, IRON, the AI cloud for the next big thing. IRON builds and operates next generation data centers and delivers cutting edge GPU infrastructure, all powered by renewable energy. Now, if you need access to scalable GPU clusters or are simply curious about who is powering the future of AI, check out iron.com to learn more, which is iren.com.
All right, man. How are you?

**Moran Cerf** (1:33)
Let the games begin.

**Peter McCormack** (1:33)
Let the games begin. I think it must be the most interesting time alive to be a neuroscientist.

**Moran Cerf** (1:40)
You make my mom happy.
It's a very interesting time to be a neuroscientist. There weren't that many neuroscientists long ago, so it's also one of the fewest times that we exist.

**Peter McCormack** (1:51)
But the AI thing must have just changed everything. Yeah.

**Moran Cerf** (1:55)
I think I was pulled into AI because of neuroscience. AI suddenly becomes a neuroscience thing. We used to have our own thing. There was a computer scientist who did AI, neuroscientists did human brains, and suddenly we are merged and we take over.

**Peter McCormack** (2:10)
Oh, really?
Because I interviewed one guy. We were talking about AI safety and are the AI bots going to kill us all. One of the most interesting things he said to me is like, we don't actually understand about 97 percent of how this AI stuff works. He says, we look in the box, we get the answers we think we want, but we don't know how it works. And he said, the amount we don't understand is increasing. And I was like, that's a bit like the brain. We don't understand how the brain works really.

**Moran Cerf** (2:39)
There's even a new kind of line of research in AI, but it's really neuroscience where they try to ask AI to do something while thinking. So the example would be you prompt Claude to tell you something about document.
So you say summarize document for me while thinking of elephants. So you put this while thinking of elephants. Now, this is unnecessary. It's like still summarize document. But when you say while thinking of elephants, essentially make the network activate different sites that you can only see if you look at the actual cells. So in the output, it looks the same. If you ask it to summarize document or summarize document while thinking of elephants, you get the same output. But when you ask summarize an elephant, you cannot see which sites in the Claude brain are active when it does something in parallel, and then you see a perfect mapping to the human brain. When I talk to you right now, you think about something else, you still respond to me, I can look at your brain and see what is the other thing, so is the same with the lie.

**Peter McCormack** (3:33)
So have they accidentally replicated the human brain in AI?

**Moran Cerf** (3:39)
I would say they did, but not accidentally. So what they did is they took the thing that is the crown of humanity if you want, language, and they taught a machine how to speak.
Evolutionary, if you look at the brain, the five steps that evolutionary neuroscientists say made humans, the crown of creation, involve radial symmetry and reinforcement learning, all kinds of things. Step five is the invention of language. So when our brain created the frontal left sides machinery, it allows us to speak to each other and communicate big ideas and correlate behind those big ideas with things that are unreal and very vague, we got the one leg up over everything else, and we gave that to AI as a given, now it's doing the same thing to us.