
Biocomputing: The AI Nightmare I Wish I Had Never Discovered?
Now that I’m early retired, I have an increasingly dangerous amount of free time on my hands.
And every once in a while, I find myself falling down some incredibly deep, dark, and technical internet rabbit hole. The kind of rabbit hole where you start out looking for something mildly interesting and, a few hours later, find yourself questioning the future of humanity.
Well, guys, over the past couple of days, I’ve discovered something that I almost wish I had never learned about.
It’s called biocomputing.
And it might be one of the most fascinating, disturbing, and downright unnerving developments in the world of artificial intelligence.
The Search for Something Beyond ChatGPT
Let’s take a step back.
Today’s AI revolution is largely built around Large Language Models, or LLMs. These are the technologies powering systems like ChatGPT and Claude.
And while these models have become extraordinarily capable, they also have some fundamental limitations.
For one thing, once a model has been trained, it doesn’t automatically learn new information or continuously update its knowledge in the same way a human does.
They can also struggle with unfamiliar problems, particularly when they haven’t encountered enough relevant examples during training.
Then there’s the enormous amount of electricity required to operate these systems. Behind every seemingly effortless conversation with ChatGPT is an increasingly massive network of data centers, GPUs, cooling systems, and power infrastructure.
This raises an interesting question.
What if simply making conventional AI models bigger and more powerful isn’t enough to achieve true Artificial General Intelligence?
What if we need an entirely different approach to computing?
Researchers are exploring all kinds of alternatives, from more efficient algorithms and specialized computer chips to entirely new computing architectures.
But one particularly fascinating alternative is to borrow something from Mother Nature herself.
The human brain.
Mother Nature Already Solved the Intelligence Problem
Here’s the basic idea behind biocomputing.
For billions of years, biological evolution has been experimenting with different forms of intelligence. And eventually, it produced something pretty remarkable: the human brain.
Think about how conventional computers work.
They generally have separate components for memory and processing. Data is stored in one place, processed somewhere else, and constantly shuffled between the two.
That movement takes time and energy.
The human brain works very differently.
In biological neural networks, memory and processing are deeply integrated into the same physical structure. Neurons communicate with one another, strengthen or weaken their connections, and continuously reorganize themselves based on experience.
This gives humans some extraordinary capabilities.
We can learn new information on the fly. We can adapt to unfamiliar situations. We can generalize from relatively few examples and solve problems we have never encountered before.
A young child can learn to recognize a dog after seeing just a handful of examples. Modern AI systems may require enormous training datasets to develop comparable capabilities.
And then there’s energy efficiency.
The human brain operates on approximately 20 watts of power. Compared with the enormous electrical demands of modern AI infrastructure, biological intelligence is millions of times more efficient.
So naturally, researchers have started asking an intriguing question:
What happens when we take living biological neurons and use them as the actual computing substrate?
Instead of simulating a brain using silicon chips, what if we could build a computer using real neurons?
Welcome to biocomputing.
From Rat Neurons to Biological Computers
Believe it or not, this isn’t entirely new.
One particularly interesting experiment dates back to 2008, when researchers at the University of Reading developed a system that connected living neurons to a small wheeled robot.
The researchers extracted approximately 300,000 neurons from rat fetuses and placed them inside a nutrient rich dish.
These neurons were connected to an array of electrodes inside a temperature controlled incubator.
The electrodes allowed the neurons to communicate with the outside world.
The robot had ultrasound sensors that detected nearby obstacles. Those signals were translated into electrical pulses and transmitted to the neurons.
The resulting electrical activity from the neurons was then used to control the robot’s wheels.
And here’s where things get interesting.
Over the course of a few months, the biological neural network adapted its activity to help the robot avoid obstacles.
In other words, these living neurons were effectively learning how to control a robot.
Another important breakthrough came in 2013, when scientists demonstrated techniques for reprogramming ordinary human skin cells into induced pluripotent stem cells, which could then be converted into neurons.
And that’s a pretty important development.
It means researchers can produce large quantities of human neurons without having to harvest them directly from someone’s brain.
Which brings us to a much more recent development.
Enter Cortical Labs and the CL1
An Australian company called Cortical Labs has been working on bringing these ideas together into something resembling an actual biological computer.
Their creation is called the CL1.
At its core is a layer of approximately 800,000 living human neurons sitting inside a nutrient rich environment.
Surrounding this biological tissue is a sophisticated life support system that maintains the temperature, nutrients, and other conditions necessary to keep the neurons alive.
The neurons sit on top of a high density electrode array that sends electrical signals to the network and reads its activity.
The result is essentially a computer that uses living human neurons as its processing substrate.
And this thing comes with a Python based software development kit that allows researchers and developers to experiment with training these biological neural networks.
Including teaching them to play Pong.
Now, this is where things start getting a little disturbing.
Teaching Human Neurons to Play Pong
Here’s how the training process works.
The computer translates the virtual environment of Pong into electrical signals that are transmitted to the neurons.
The activity of those neurons is then interpreted as actions within the game, such as moving the paddle.
When the neurons successfully hit the ball, they receive a predictable electrical signal.
When they miss, they’re pushed with a chaotic pattern of electrical stimulation.
Essentially, the system provides feedback that encourages the neural network to reorganize itself in ways that produce better results.
Over time, the neurons adapt their connections and activity patterns, eventually learning how to play the game.
And guys, I don’t know about you, but there’s something deeply unsettling about the idea of training living human neurons using electrical rewards and punishment.
And Pong is apparently just the beginning.
Cortical Labs has also demonstrated biological neural networks playing Doom.
They have experimented with clustering multiple biological computers together, creating what they describe as biological hive minds.
In these experiments, groups of biological neural networks performed tasks such as navigating virtual environments and finding specific targets.
And the applications aren’t limited to video games.
Cortical Labs has also demonstrated a robotic dog controlled by laboratory grown neurons, capable of navigating toward a target in a simple environment.
They have even made this technology available through a cloud based service, allowing researchers and developers to experiment with biological computing remotely.
We’re no longer talking about some distant science fiction concept.
Living neural networks are already being used as experimental computing systems.
But Is Biocomputing Really the Future?
Now, before we get carr…
Get full access to AsianDadEnergy's Newsletter at asiandadenergy.substack.com/subscribe
Więcej odcinków z kanału "AsianDadEnergy's Podcast"



Nie przegap odcinka z kanału “AsianDadEnergy's Podcast”! Subskrybuj bezpłatnie w aplikacji GetPodcast.








