Your biological center for thought, comprehension, and learning bears some striking similarities to a data center housing rows upon rows of highly advanced processing units. But unlike those neural network data centers, the human brain runs an electrical energy budget. On average, the organ functions on roughly 12 watts of power, compared with a desktop computer’s 175 watts. For today’s advanced artificial intelligence systems, that wattage figure can easily increase into the millions.
Knowing this, researchers believe the development of cyborg “biocomputers” could eventually usher in a new era of high-powered intelligent systems for a comparative fraction of the energy costs. And they’re already making some huge strides towards engineering such a future.
As detailed in a new study published in Nature Electronics, a team at Indiana University has successfully grown their own nanoscale “brain organoid” in a Petri dish using human stem cells. After connecting the organoid to a silicon chip, the new biocomputer (dubbed “Brainoware”) was quickly trained to accurately recognize speech patterns, as well as perform certain complex math predictions.
As New Atlas explains, researchers treated their Brainoware as what’s known as an “adaptive living reservoir” capable of responding to electrical inputs in a “nonlinear fashion,” while also ensuring it possessed at least some memory. Simply put, the lab-grown brain cells within the silicon-organic chip function as an information transmitter capable of both receiving and transmitting electrical signals. While these feats in no way imply any kind of awareness or consciousness on Brainoware’s part, they do provide enough computational power for some interesting results.
To test out Brainoware’s capabilities, the team converted 240 audio clips of adult male Japanese speakers into electrical signals, and then sent them to the organoid chip. Within two days, the neural network system partially powered by Brainoware could accurately differentiate between the 8 speakers 78 percent of the time using just a single vowel sound.
Next, researchers experimented with their creation’s mathematical knowledge. After a relatively short training time, Brainoware could predict a Hénon map. While one of the most studied examples of dynamical systems exhibiting chaotic behavior, Hénon maps are a lot more complicated than simple arithmetic, to say the least.
In the end, Brainoware’s designers believe such human brain organoid chips can underpin neural network technology, and possibly do so faster, cheaper, and less energy intensive than existing options. There are still a number of hurdles—both logistical and ethical—to clear, but although general biocomputing systems may be years down the line, researchers think such advances are “likely to generate foundational insights into the mechanisms of learning, neural development and the cognitive implications of neurodegenerative diseases.”
But for now, let’s see how Brainoware can do in a game of Pong.