For the new study, published Monday (Dec. 11) in the journal Nature Electronics, the researchers used a technique called reservoir computing; in this context, the organoid serves as the "reservoir." In such a system, the reservoir stores information and reacts to information that's inputted. An algorithm learns to recognize changes triggered in the reservoir by different inputs and then translates these changes as its outputs.
Using this framework, the researchers plugged the brain organoid into this system by supplying it with electrical inputs delivered through electrodes.
"Basically, we can encode the information — something like an image or audio information — into the temporal-spatial pattern of electrical stimulation," said study co-author Feng Guo, an associate professor of intelligent systems engineering at Indiana University Bloomington.
In other words, the organoid responds differently depending on the timing and spatial distribution of the electricity from the electrodes. The algorithm learned to interpret the organoid's electrical responses to that stimulation.
Although the brain organoid is much simpler than an actual brain — it's essentially a small sphere of brain cells — it has some ability to adapt and change in response to the stimulation. The response of the different types of brain cells, cells at different stages of development, and brain-like structures in the organoid provide a rough analog to the way our brains change in response to electrical signals. Such changes in the brain fuel our ability to learn.
Using this unconventional hardware, the researchers trained their hybrid algorithm to complete two types of tasks: one related to speech recognition and another to mathematics. In the former, the computer showed about 78% accuracy at recognizing Japanese vowel sounds from hundreds of audio samples. And it was fairly accurate in solving the math task but slightly less so than traditional types of machine learning.
The research marks the first time a brain organoid has been used with AI, but previous studies have used simpler types of lab-grown neural tissue in a similar way. For example, scientists have interwoven brain tissue with a form of reinforcement learning, a type of machine learning that might have more similarities with how humans and other animals learn than reservoir computing.
Future research could attempt to combine brain organoids with reinforcement learning, said Lena Smirnova, an assistant professor of environmental health and engineering at Johns Hopkins University who co-authored a commentary about the new study.
One of the advantages of creating biocomputers would be energy efficiency, since our brains use far less energy than today's advanced computing systems. But Smirnova said it might be decades before technology like this could be used to create a general-use biocomputer.
While organoids aren't close to replicating full-blown human brains, Smirnova hopes the technology will give scientists a better understanding of how the brain works, including in diseases like Alzheimer's. Replicating both the brain's structure (with organoids) and function (with computing) could allow researchers to better understand how the brain's structure is related to learning and cognition, for instance.
