AI brain

Mary Shelley, the author of “Frankenstein,” would love this. Scientists at the University of Indiana have developed a hybrid computing system, dubbed Brainoware, that combines electronic hardware with lab-grown human brain cells known as organoids. The goal is to reduce the enormous amounts of energy used by AI systems to an efficiency level used by the human brain, thereby making high-powered AI systems available for a fraction of the cost of their silicon-based counterparts.

While the technology is still in its infancy, the idea is to develop a neuromorphic computing system modeled after the structure of the human brain. “Brainoware could provide  additional insights into AI computing because brain organoids can provide BNN’s (biological neural networks) with complexity, connectivity, neuroplasticity and neurogenesis, as well as low energy consumption and fast learning,” according to paper published in Nature Electronics. The human brain typically runs on about 20 watts of power while a comparable AI would need about 8 million watts.

“Brainoware uses a human brain organoid as an adaptive living reservoir to conduct unsupervised learning by processing spatiotemporal information through the neuroplasticity of the brain organoid,” says team leader Fen Guo, associate professor of intelligent systems engineering at IU’s Luddy School of Informatics, Computing and Engineering.

Brainoware’s “adaptive living reservoir” changes in response to its inputs, giving it features of unsupervised learning. Reservoir computing is a type of artificial neural network that captures and remembers information based on a sequence of electrical stimulations. Like an updated scene from a Frankenstein film, Brainoware was zapped “spatiotemporal sequences” of electricity encoded with different types of information. In one experiment, IU scientists hit the organoid with electric pulses corresponding to 240 audio clips of Japanese vowels. The scientists recorded how the lump of brain cells responded and decoded it with a regression algorithm.

At first, the untrained, organoids identified the vowels only about half the time. But after a couple of days of being zapped by data, the accuracy rate climbed to 75 percent, a clear indication that Brainoware was capable of unsupervised learning. Another experiment involving a complex math problem generated similar results. The 3D brain cell cultures are derived from stem cells and were used to create different types of brain tissue like neurons, glia and ventricular zones.

In addition to its AI applications, Brainoware tech could be used to model and study neurological disorders like Alzheimer’s disease. Brainoware may be able to reveal the effectiveness of different treatments.

Brainoware faces many hurdles before biological computers become a cutting edge technology. Next steps include testing the ability to handle complex tasks and engineering more stable organoids. The brain organoids must be grown in incubators, a task that increases in difficulty as complex tasks require bigger “brains.” Brainoware also gives new expression to the phrase “my computer died.”

Brain tissue can develop necrosis and die in the lab. That alone will likely lead to ethical discussions about whether “It’s Alive!” to borrow a phrase from Frankenstein movies.