Now little brains have learned to play pong

Cortical Labs, a startup out of Melbourne, Australia, calls them “bowl brains.” It consists of 800,000 to 1 million neurons grown in a container, placed on a plate with microelectrodes. The electrodes can stimulate the cells and read the activity of the neurons.

A program used to emulate the computer game Pong without any opponent. Ball position information is provided by activating the electrodes to the right or left of the racket. How close the ball is to the racket is determined by the frequency of the signal.

And what about the racket? Yes, it is the mind of the bowl itself.

We think it’s reasonable to see them as cyborg brains. We often say they live in The Matrix. When they’re in the game, they think they’re the racket, says Brett Kagan, head of research at Cortical, for New Scientist magazine.

If that sounds a little shocking, Cortical has a more poetic description of the same phenomenon. in his statement: “The fusion of silicon and brain cell. An entity born in the digital world lit by Promethean fire is the human brain.”

Learn faster than artificial intelligence

Cortical Grail brains cannot, for now at least, not function at the same level as non-biological AI. But they learn faster. Brett Kagan tells New Scientist that it takes 5,000 rounds of artificial intelligence to reach a level that a Cortical wet product can reach in 10-15 tries.

The Australian company showed its results in a preliminary version, which means that the article has not been subject to peer review. But neurologist Carl Friston still looks like fire and flames.

The authors succeeded in obtaining a biological neural network to create an understanding of their surroundings and at the same time operate in this world. Successfully closing the loop of motor perception is not only an exciting technical advance but takes us one more step closer to artificial brains.

New Scientist’s conversation with Carl Friston in particular, and that he uses big words, may not have been a coincidence. Cortical has taken a lot of inspiration From the principle of free energy, the theory that made Carl Friston famous. The behavior that Cortical neurons exhibit seemed to indicate that there was something in Friston’s thoughts.

One can fill in volumes and dedicate one’s life to trying to understand the principle of free energy. Roughly simplified, however, it means that self-regulating biological systems – such as the human brain – strive to minimize surprises. What the Cortical technique appears to show is that neurons in a petri dish also create an internal model of their external world. “They want to predict what will happen in terms of the input they get, and they don’t like to be surprised,” New Scientist writes.

When it comes to non-biological AI, we often talk about the reward function: an incentive to train an algorithm to achieve a goal. With the Cortical system, there is no need for a reward function, and Director of Research Brett Kagan believes it is based on the principle of free energy.

– That’s why cells “play”. When they play, his input becomes more predictable. If the cells are not turned on, the input becomes more random and this creates an aversion.

The wonderful thing here is that spontaneity appears spontaneously, says Carl Friston.

Cortical Labs used cells taken from mice and humans in their experiments, and it is reassuring that tests on human neurons have shown that they work better. However, no conclusions can be drawn on this topic, because mice cells came from mice during the embryonic period, while human neurons originated from stem cells.

According to New Scientist, Cortical Labs’ vision is to develop cyborg brains that could eventually become smarter than computer-based systems.