The brain is powerful, and computers are powerful—why not harness their strength combined? Researchers at the University of Illinois (UIUC) have built a “living computer” using mouse brain cells in a proof-of-concept that might someday lead to brain-powered robotics.
Andrew Dou, a graduate student at UIUC’s Holonyak Micro & Nanotechnology Lab, led a team of researchers through reprogramming mouse stem cells to obtain a collection of neurons. The team then grew approximately 80,000 neurons in a petri dish, creating a flat layer that would eventually power the living computer.
The computer consists of a box about the size of one’s palm, which sits inside an incubator to keep living cells alive. Dou’s team placed the neuron layer onto an electrode grid below an optical fiber, both inside the box, to create a space in which the neurons would be stimulated by electricity and light. But it wasn’t enough to make what could pass as a computer; the researchers had to train the system to function like one, too.
Dou and his colleagues wanted their system to mimic a neural network, which can “learn” and recognize relationships between data. To that end, they trained the computer to recognize patterns by flashing light and electrical pulses at various rhythms, then allowing the neurons to rest for 30 minutes. After the rest period, the researchers recorded and analyzed the neurons’ responses using a regular computer chip.
Dou’s team measured the computer’s success on the F1 (0 to 1) scale commonly used to test neural networks. At first, the computer couldn’t score above a 0.6; the neurons kept firing what seemed to be random electrical signals. The team had to devise a “combination of chemical and additional electric impulses” to calm this activity, after which the computer scored a 0.98—nearly a perfect score.
The living computer’s relatively quick success has encouraged the team to continue working toward computers and robots powered by brain cells. Other labs are working to build viable biocomputers, too, the consensus being that brain cell-powered computers could eventually store more information or process data far more quickly than our lousy regular computers. Training Dou’s living computer took far less time than training a conventional neural network; if the team builds on the computer’s size, it could handle more complex training and requests. The researchers also hope that expanding the computer’s size and complexity could result in “some unexpected behavior” that they haven’t trained it to perform. That’s not concerning, right?
