Using Brainware, a group from Indiana University Bloomington made theory a reality. They attached a wafer-sized chip, tightly packed with electrodes, to a brain organoid that resembled the cortex, the outermost layer of the brain supporting higher cognitive functions.
The mini-brain worked similarly to a supercomputer’s memory and central processor unit. It was fed electrical shocks as input and used brain activity to produce its computations, which an artificial intelligence tool then decrypted.
Eventually, Brainware was able to distinguish between distinct individuals’s “sounds” after being taught on soundbites from a group of people that were converted into electrical shocks. Another test saw the system successfully solve a difficult arithmetic issue that presents a challenge for AI.
Changes in the mini-brains neural network connections allowed the system to learn, much like our brains do daily. Brainoware is merely the beginning; it opens the door to more advanced hybrid biocomputers that may reduce energy consumption and accelerate processing. Additionally, the configuration enables neuroscientists to delve deeper into the inner workings of human brains.
Drs. Lena Smirnova, Brian Caffo, and Erik C. Johnson of Johns Hopkins University, who were not involved in the study, said that while computer scientists are working to create silicon computers that mimic the functioning of the brain, neuroscientists are attempting to comprehend the calculations of brain cell cultures. Brainoware may provide fresh perspectives on learning and brain development, as well as assist in testing novel treatments for neurological disorders.
Perched on top of a series of electrodes is a small ball of brain cells that hums with activity. It is subjected to an electrical zap sequence for two days, with each stimulus encoding the unique speech characteristics of eight persons. It can distinguish different speakers by day three.
The device, known as Brainoware, advances biocomputing by utilizing 3D brain organoids, or “mini-brains.” Typically derived from human stem cells, these models proliferate quickly to become a diverse range of neurons interwoven into neural networks.
The blobs spark electrically, just like their biological counterparts, indicating that they may be able to process, store, and learn information. They have long been considered a viable hardware element for computers inspired by the brain by scientists.
A Different Take on Neuromorphic Computing
As the human brain consists of 200 billion neurons connected by hundreds of trillions of connections, it is possibly the most powerful computing hardware.
Compared to traditional computers, which have distinct units for data processing and storage, its structure is fundamentally different. The computer must transfer data between each task, which significantly increases computing time and energy. In contrast, the same anatomical location in the brain serves as the union of both activities.
These structures, which link neurons into networks, are known as synapses. By altering the strength of their connections with others, synapses can learn by strengthening their bonds with collaborators who aid in problem-solving and by storing information in the same location.
The procedure might sound recognizable. These ideas serve as a rough foundation for artificial neural networks, an AI technique that has gained immense popularity. However, the energy required is very different. The brain consumes 20 watts, or about the same amount of power as a tiny desktop fan. An artificial neural network that is comparable uses eight million watts. While AI is known to rely on enormous datasets, the brain can learn from a small number of examples with ease.
Researchers have attempted to replicate the processing characteristics of the brain in hardware chips. These neuromorphic chips combine computation and storage in one unit, thanks to their unusual construction and components that can alter their capabilities in response to changes in voltage or temperature. These chips are capable of speech recognition and computer vision. However, they are challenging to produce and only partially mimic the inner workings of the brain.
An Intelligent Computer
You may be confident that the team did not attach electrodes to living brains. They resorted to brain organoids instead. The human stem cell-derived mini-brains took only two months to mature into a variety of neuron types that interacted with one another to form electrically active networks.
Each mini-brain was carefully deposited onto a stamp-shaped chip that was filled with tiny electrodes by the researchers. The chip can simultaneously zap the organoids with approximately thirty electrodes and record the signals from over one thousand channels in the brain cells. This enables the mini-brain activity to be recorded and stimulation to be precisely controlled. Abstract neural outputs are converted into responses that are understandable by humans on a standard computer using an AI tool.
The researchers recorded 240 audio samples of eight persons speaking to test speech recognition. Every clip features a single vowel. The dataset was converted into distinctive electrical stimulation patterns, which they then fed into a freshly developed mini-brain. The Brainwave technology demonstrated approximately 80 percent accuracy in speaker discrimination in just two days.
The electrical zaps “trained” the mini-brain to strengthen certain networks while pruning others, according to the team’s use of a widely used neuroscience metric. This suggests the mini-brain rewired its networks to support learning.
In a different experiment, Brainware faced up against AI in a difficult arithmetic assignment that might provide stronger passwords. Brainoware was significantly faster than an AI with short-term memory but was marginally less accurate. It took less than 10% of the AI’s time to reach virtually compatible outcomes when operated without human oversight.
According to study author Dr. Feng Guo, this is the first instance of brain organoids being used for computing, as reported by MIT Technology Review.