Research into organoids—small clusters of specialized living cells grown in the lab that mimic the functioning of an organ—is moving forward at breakneck speed. Initially, the approach was used primarily to test substances on very specific cell populations, particularly in the development of drug treatments. But recently, there is one specific type that seems to be attracting significantly more attention than others: brain organoids.
These synthetic mini-brains that cannot think or feel anything, but still remain functional at the most basic level, play an important role in fundamental neuroscience, but not only. Today, More and more researchers are also exploring their interest in the context of machine learning.
Brains on chips, the new darling of AI researchers
Indeed, an old adage says that the human brain remains the most powerful computer there is — an interpretation reinforced by the fact that current AI models try as best they can to imitate its architecture, with varying degrees of success.
After all, both technologies are based on the same fundamental mechanisms. A large number of neurons, either virtual or biological, form dynamic networks whose architecture evolves each time they are subjected to a signal.
More and more researchers are therefore starting from a rather pragmatic observation: why bother starting from scratch to produce a pale copy of the human brain, when we could directly use this biological marvel that evolution has taken the trouble to optimize for millions of years?
This idea, long confined to science fiction, is beginning to produce quite spectacular results. In the space of five years, for example, we have seen the emergence of clusters of synthetic neurons capable of playing Pong, recognizing words or solving mathematical equations. More recently, the Swiss startup Final Spark has even designed a platform for ” neuro-cloud computing ” to allow researchers around the world to experiment with this technology.
But we may have just reached a new milestone; in a press release spotted by New Atlasa group of Chinese researchers claimed that an artificial intelligence embodied by a brain organoid was now capable of… control a robot.
Robots with real brains
The Tianjin University press release (translated using DeepL) suggests that this concept, dubbed MetaBOC (for Brain-On-Chip, ” brain on a chip » in English), is still far from being mature. The image presented at the top of the article is not not a finished product; it is only a demonstration object which seeks to show what such an android might look like. But the conceptual foundations are already relatively well in place.
This architecture relies on a specialized microcontroller that is responsible for making the connection between the robot and its synthetic brain. In essence, it is the same concept used by the researchers who made their organoid play Pong, but applied to a mechanical interface. According to the authors, this architecture makes it possible to use the brain organoid as an embedded machine learning model. The robot can thus gradually learn to move, avoid obstacles, and even grasp objects.
Many obstacles to overcome
The researchers’ press release does not, however, specify the exact method by which they intend to carry out this reinforcement learning process. They could rely entirely on electronics, using tiny electrical impulses comparable to signals carried by the nervous system. In parallel, they could also use substances such as dopamine synthesis to encourage the brain-on-a-chip when it performs a satisfying action.
Whatever approach is chosenthere is still a lot of work to be done before we get to a functional robot. Indeed, while organic neurons have a number of advantages over their virtual counterparts, particularly in terms of energy consumption and flexibility, building the machine around a living organelle also has its drawbacks. This involves, in particular, design an entire life support system to provide nutrients and oxygen, maintain a viable temperature, protect all the microorganisms, and so on. And these factors will greatly complicate the integration.
Despite these limitations, it will be very interesting to observe the evolution of this theme. It is probably only a matter of time before a laboratory produces a functional prototype. And the process that will take us there is announced quite fascinating, both from a strictly technological point of view and from an ethical one. Indeed, the arrival of the first mechanical systems equipped with artificial brains, even very rudimentary ones, promises to confront humanity with a a whole bunch of rather uncomfortable existential questions about the nature of intelligence, the notions of consciousness and sentience, and so on.
To not miss any news on the Journal du Geek, subscribe on Google News. And if you love us, we have a newsletter every morning.
