AI: Cortical Labs connects human neurons to a chip

AI has just crossed a strange frontier. Cortical Labs claims to have connected living human neurons to a silicon chip, with an ambition to create a new form of calculation, halfway between the classic machine and biological tissue. This is not a simple announcement effect. This is a serious path towards computing that is more sober, more flexible, and potentially more confusing than anything the sector has shown so far.

An AI that no longer relies solely on silicon

Cortical Labs no longer talks about purely software AI after US restrictions on AI chip exports. The company is promoting a system where real human brain cells are grown, connected to a chip, and then stimulated to learn simple tasks. In the story presented, these neurons would have been trained to play Doom in a week. The symbol is strong. It's not about the game itself, but the system's ability to adapt.

What is striking is the change in logic. For years, modern AI has been advancing with more power, more data and more computing centers. Here, the idea is almost the opposite. Instead of stacking GPUs, we seek to exploit the natural efficiency of life. The human brain remains, to this day, a brutal reference in terms of energy consumption and plasticity.

We must therefore read this announcement as a signal. AI is no longer just a battle between models, chips and giant clouds. Another way appears. It remains experimental, limited, and still far from mass use. But it is already changing the debate.

Why is this progress so intriguing?

The first point that fuels interest around this biological AI is energy. According to the elements advanced around the CL1 system, a rack of 30 units would remain in a very low electrical envelope compared to the enormous infrastructures mobilized today to train and run AI models. This is where the subject is getting serious. The energy cost of AI is now an industrial, economic and political problem.

The second point is adaptability. An artificial neural network must be trained computationally. A network of living neurons learns differently. It reacts, reorganizes, stabilizes or disrupts depending on stimulation. This difference can open the door to more economical and flexible hybrid systems, capable of dealing with certain problems without copying the operation of current architectures.

Finally, there is the accessibility of the model. With its idea of ​​“Wetware as a Service”, Cortical Labs wants to offer remote access to this type of biological infrastructure. In other words, the company is trying to transform living neurons into a consumable computing resource like a cloud service. This is precisely what gives this announcement a rocking scent. We no longer only sell calculations. We sell living matter organized to calculate.

Between industrial promise and ethical unease

The unease doesn't come from a science fiction fantasy. It comes from reality. When human cells derived from biological samples are used to perform even rudimentary tasks, a question immediately arises: how far can we go without moving the moral boundary? For the moment, we are talking about very simple systems. But the simple fact that this question already exists shows that the terrain is sensitive.

We must also avoid getting carried away. Playing neurons in Doom does not mean creating consciousness. This also does not mean that a machine “thinks” like a human. The risk here is not only technological. It is narrative. The industry loves spectacular demonstrations. But a demonstration is not yet an industrial revolution.

Still, the concern is not absurd. If this form of Artificial Intelligence progresses, it could force regulators, researchers and companies to review entire frameworks. Status of life, ownership of tissues, commercial use, liability in the event of an error: all of this will eventually come back to the table. And this time, the debate will not be theoretical.