For years we have described artificial intelligence as the attempt to imitate the human brain with silicon. Artificial neural networks, increasingly large models, increasingly powerful GPUs. The implicit goal was simple: to build machines that work like the brain.
But what if the path were exactly the opposite?
What if instead of simulating a brain we started using a real one?
It is the radical question coming from Australia, where the startup Cortical Labs is developing a new class of computers based on human neurons grown in the lab. The system is called CL1 and represents one of the boldest experiments in the history of computing: it integrates about 200,000 living biological neurons on a silicon chip.
The principle is as simple as it is unsettling. The neurons are grown in vitro and connected to microelectrodes that allow the computer to send electrical signals and record the responses of the neural network. In other words, the system communicates directly with real brain cells.
It’s not just biology. It’s a new computing paradigm.
Already in 2022 the same lab had demonstrated something surprising: these biological neurons were able to learn to play Pong, the famous 1970s video game. The system received electrical feedback on success or failure and, after a number of attempts, improved its strategy.
It was not a simulation. The neurons were actually learning.
In recent days a new experiment has emerged, also discussed in online scientific and technological communities: the same architecture was used to interact with Doom, one of the most iconic video games in history. The leap is not trivial. It means that these biological networks can tackle more complex environments compared to the simple feedback systems used in the initial experiments.
The emerging technology is often called biological computing or organoid intelligence. The underlying idea is that the biological brain remains, even today, the most efficient processing system ever observed in nature. A human brain consumes about 20 watts. The data centers powering large artificial intelligence models consume amounts of energy that begin to compete with those of entire countries.
Growing neurons may seem like science fiction, but from an energy perspective it could be surprisingly rational.
Of course, the technology is still in its infancy. We are talking about biological neural networks tiny compared to a human brain, which contains about 86 billion neurons. The 200,000 neurons of the CL1 system are an infinitesimal fraction.
Yet the experiment raises profound questions.
If biological neural networks can learn, adapt, and solve problems, we are witnessing the birth of a new computational architecture. No longer computers that imitate the brain, but computers that directly incorporate brain tissue.
At that point the question ceases to be technological and becomes philosophical.
What are we really building?
A machine?
A controlled biological system?
Or a new hybrid form of intelligence?
The history of computing is made of conceptual leaps: from mechanical calculation to transistors, from transistors to microprocessors, from microprocessors to GPUs that today power generative AI.
The idea that future computers might contain living cells may seem disturbing. But, if it works, it could also represent one of the most radical solutions to the energy problem of artificial intelligence.
For decades we have tried to build machines that mimic the brain.
Now someone is trying to do something much simpler.
Use a piece of it directly.
