Clarified Mind · Transcript · · Updated
We're Not Ready for Biocomputing – Transcript
Video by Clarified Mind · Watch the original on YouTube
Summary
AI-generated from the transcript. Verify details against the source video.
- Researchers built computers running on living brain cells because AI uses an enormous and rapidly growing amount of electricity.
- Cortical Labs developed systems like Dishbrain and CL1 using human neurons to play games such as Pong and Doom.
- Neuroscientist Mark Solms argues that feeling comes first in evolution, driven by the basic need to stay alive.
Use this video with Pro
Reading, copying and transcript downloads are free. Create saved AI results for this video with a signed-in Pro account ($9/month).
Transcript
Search ignores accents. All transcript text stays visible.
In a few seconds, I'm going to show you one of the most disturbing things happening in science right now, and almost no one is talking about it. That's why we made this video. You're looking at a computer in a laboratory called Cortical Labs. It's playing Doom. But look closer. There's nobody playing it. Or actually, there is. They just can't get out. They are trapped there because this computer is
running on living human brain cells, and those cells are the ones playing. So, why would anyone build a computer out of brain cells? It basically comes down to power.
AI uses a massive amount of electricity, and that number keeps going up. >> Amazon investing another $15 billion and continuing its massive investments in AI, this time in the form of compute power. >> The computing power it takes to train these models roughly doubles every 6 months. Almost nothing in the real world grows that fast. But you can't build power plants that quickly. So, there's a growing gap between what AI needs and what we can actually
supply, and it keeps getting wider. You can see the panic in what the big companies are doing. They're going after energy now, not just better software. Old nuclear plants that got shut down years ago are being turned back on just to power them.
Companies are signing deals to lock up huge chunks of the grid years ahead of time. It's not really a race for smarter AI anymore. It's turning into a race for electricity. So, a few researchers went a totally different direction. They looked at the brain.
Because the brain figured this out a long time ago. Think about what your brain is doing right now, and how little it runs on. Everything you're thinking runs on about 20 watts. We're using whole power plants just to copy a small piece of what your brain does on basically nothing.
And it learns way better than our machines, too. To teach a computer what a dog is, you have to show it thousands of photos. A little kid sees a dog once or twice and just gets it. So, how did we get from that to a dish playing Doom?
It took three steps. Back in 2008, researchers grew a sheet of rat neurons and hooked it up to a little robot. And the cells actually drove it around. It sounds bigger than it was.
The neurons just sat flat on a grid of electrodes and connected to their neighbors. That was pretty much it. This was nowhere near a real brain, and it wasn't even the first time someone had wired living tissue to a machine. One team even did it with the brain of a lamprey. The footage always looked a little like a horror movie. And from the start, people had one real doubt. Were the cells actually driving, or was the computer just taking their random signals and making it look that
way? Keep that question in mind. We're coming back to it. The second step came in 2013. A scientist named Shinya Yamanaka showed you could reset a grown-up cell all the way back to the start. He found four proteins that turn a regular cell into a blank, brand new one. And from there, you can turn it into almost anything, including a neuron. That was huge. It meant you didn't have to take neurons out of a brain anymore. You
could just take a bit of skin and grow them. There was another big step, too. If you grow neurons on a flat dish, they only connect to the ones right next to them. But if you let them grow in 3D, they start to organize on their own. They spread out, fold into layers, and form the rough beginnings of real brain regions, kind of like an embryo does.
One of those tiny balls of tissue is called an organoid. Then, in 2022, a company called Cortical Labs put all of this together into something they called Dishbrain. They grew about 800,000 neurons, part mouse and part human, and taught them to play Pong.
Two things about this were a big deal. First, a dish of cells was actually working toward a goal. Second, and this is the weird part, they'd figured out how to teach it. They could ask the cells a question and get an answer back.
You're not supposed to be able to do that to a blob of tissue in a dish. So, how do you teach a neuron? You can't give it a treat. So, they used a basic fact about neurons. They hate chaos.
If you give them a signal they can't predict, it stresses them out. So, the setup was simple and kind of mean. When they missed the ball, they got hit with scrambled random noise. When they hit it back, the signal went calm and steady. The neurons have no idea what Pong is. They're not playing to win. They're just trying to make the chaos stop. And they didn't pick that method at random. They based it on an idea of the free energy
principle. Basically, it says every living thing is always trying to avoid surprises and keep its world predictable. Because being surprised usually means something's wrong. So, that urge to calm things down isn't something the scientists added. It might be the most basic thing a living cell does. But, there's a big catch. We can grow brain tissue in 3D, and we can talk to living neurons. We just can't do both at the
same time yet. So, in order to talk to them, Dishbrain had to keep the cells flat because the chip that reads them is a flat grid. If you put a round organoid on it, you only touch the bottom layer. So, they had to flatten it out and give up the brain's real shape just to talk to it. And remember, this is the crude version. Half the design is missing.
Which brings us to a big new. Cortical Labs showed off a computer that runs on brain cells. They call it CL1, and it plays Doom. This one was grown from about 200,000 human neurons. They took skin cells, reset them, turned them into neurons, and spread them across a bed of electrodes. Then, they tied the signals to moving and shooting the same way as before. When it did well, the signal
stayed calm. When it took damage or died, the cells got flooded with chaos. And within about a week, it had learned the basics of moving around and shooting. I'm not going to pretend to be calm about this, and I don't think you should be either. It's amazing, and it kind of makes me sick. And I think both of those reactions are fair, because that computer came from a real person. It's literally made of human cells.
I keep imagining what it would be like to be the thing stuck in there, playing the same little piece of game over and over forever, not remembering anything before it, not knowing that anything else even exists.
For these cells, there's nothing else, just the game. Are we making a piece of a human being whose whole world is getting shocked inside a violent video game just to save some electricity? For years, we worried about the opposite thing, humans turning into machines, chips in our heads, that whole idea.
Almost nobody expected it to happen the other way around, but that's kind of what's going on here. Now, to be fair, there was a big backlash to all of this, and it was kind of fair. When Cortical Labs published the Pong work, the title used a heavy word. It said the neurons showed sentience.
Normally, that word means someone's in there having an experience. But the scientists meant it in a really narrow technical way. Basically, anyway, the story blew up and a lot of scientists were angry, too.
About 30 of them signed a letter saying the company had twisted the word just to get headlines and oversold the whole thing to help the business. The worry was pretty practical. Too much hype leads to backlash and enough backlash could shut down a young field like this.
And this research really matters. Scientists use these same cells to build living models of human diseases. They can watch how a disease develops and test drugs on tissue that's way closer to ours than a mouse's. That could save real lives and replace a lot of animal testing. It would be a real loss if a stunt with Pong and Doom dragged it all down. Doom really was kind of a stunt. The audience asked for it and it
made great marketing. But it brings us right back to that doubt from 2008. Is the brain really playing or is the computer just cleaning up random noise and making it look good? Almost 20 years later, we can finally check. The developer behind the demo, who goes by Sean Cole, released the code so anyone can read it. And what's inside is not just a brain in a box.
There are three parts. First, a normal computer chip running an AI that watches the game and decides when to reward and when to punish. Second, the 200,000 human neurons. And third, a translator in the middle passing messages both ways.
Here's the uncomfortable part. Almost all of the actual smarts are on the chip, not in the cells. Reading the screen, running the learning, telling a win from a loss, that's all the computer. The developer was even worried his AI would get so good it wouldn't need the neurons at all. So, he ran a simple test. He took the cells out and replaced them with pure random noise.
And the learning just stopped. So, the neurons are doing something real, even if it's a small part. So, the truth is somewhere in the middle. It's not a brain playing on its own, but the cells aren't just for show, either. It's mostly a normal computer doing the work with living cells adding a small but real piece.
That's where we are today. And here's where I think it goes next. I don't think that small biological part stay small. Labs all over are racing to grow this tissue in full 3D and connect with properly. And it's already turning into a business, a huge business.