A group of researchers from the University of Illinois has created what can be called a "living computer. The main element of this computer is artificially grown brain tissue cells, and the computer itself is proof of the feasibility of the idea that such "artificial brains" could become controlling components of various robotic devices in the future.
Initially, scientists took the necessary number of stem cells from experimental rodents and programmed them to turn into neural tissue cells. Thus they managed to grow a thin film of about 80 thousand neurons in a Petri dish.
The computer itself has a size comparable to the size of a human palm, it is located inside an incubator in which a certain temperature, humidity and other conditions favorable for the vital activity of nerve cells are maintained. A thin film of nerve tissue lies on a grid of electrodes, and even lower are the outputs of optical fibers, which all together allow the cells to be stimulated simultaneously by electricity and light.
However, assembling all of this alone was not enough to make the computer work, the scientists had to "teach" the nerve tissues to work with the rest of the components like a single unit. To do this, they trained the system the same way they train neural networks, only in this case they used current pulses and flashes of light, alternating in the right sequences. After that, the computer was left alone for 30 minutes, which was enough to fix the formed neural connections.
The researchers then performed the F1 test, which is used to test neural networks and which produces a result in the range of 0 to 1. At first, the live computer could not score above 0.6 because the neurons were randomly generating random electrical signals. Scientists had to develop and use a special combination of chemical stimuli and additional electrical impulses to sort of "calm down" this mini-brain, which stopped generating random signals. And already after that, the computer passed the test with a score of 0.98, almost a perfect score.
And in conclusion, it should be noted that this live computer demonstrated very high teachability, it took many times less time to train it than a standard neural network. And in their future work, scientists from Illinois plan to create a more complex biocomputer with more neurons. They hope that this increase in complexity can lead to the demonstration of "some unexpected behavior" that the neural network was not originally trained for.
