Human neurons: The hybrid mini-computer, a mix of neuronal cells assembled in organoids and an electronic chip, is a reality. The new machine, capable of performing speech recognition calculations and solving some computer problems, was made by the research group led by Feng Guo of Bloomington University in Indiana. The results, published in the journal Nature Electronics, demonstrate the potential of these innovative bioinformatics tools that fuse brain organoids with electrical hardware. Advances in Artificial Intelligence, particularly those based on machine-learning processes, require enormous computing power. As systems become more powerful and sophisticated, the importance of significantly improving the energy efficiency of chips and computers emerges.

A new era of computing

Precisely in response to these needs, efforts have been made in recent years to develop neuromorphic computer systems that are somewhat similar to the functioning of our biological brain, which can boast much lower power consumption than electronic analogues. A major breakthrough now comes from American researchers who have succeeded in fusing together an electronic chip with a miniature brain obtained from human neurons. The result is a kind of hybrid mini-computer called Brainoware, capable of processing data very efficiently in terms of energy consumption and comparable to its traditional analogue in terms of performance. At the moment, Brainoware is being used for speech recognition operations, specifically on some Japanese sounds, based on 240 audio tracks recorded by eight different people, and put to the test in computing the evolution of a dynamic system.

An important milestone for the future development of bioinformatics is improving the understanding of how the human brain works. “It may take decades before general bioinformatics systems can be realized, but,” commented Lena Smirnova of Johns Hopkins University in Baltimore, “this research is likely to generate fundamental insights into learning mechanisms, neural development, and cognitive implications of neurodegenerative diseases. A human brain typically consumes about 20 watts, while current AI hardware consumes about 8 million watts to handle a comparative ANN (artificial neural network). Brainoware could provide further insights for AI-based computing because brain organoids can provide BNNs (biological neural networks) with complexity, connectivity, neuroplasticity and neurogenesis, as well as low power consumption and fast learning.

Humanoid robots are far away

Brainoware sends and receives information from the brain organoids through “adaptive reservoir computation”. This method enables unsupervised learning from training data, which can still model the functional connectivity of the organoids. The practical potential of the system was demonstrated through tasks such as speech recognition, in which it distinguished the voices of individual speakers with increasing accuracy after training. For example, organoids were trained to identify an individual’s voice in a series of 240 audio clips of eight people uttering Japanese vowel sounds. After training, organoids could complete the task more than 70 per cent accurately. Science, however, is still far from building living robots. The organoids could only identify speakers, not understand their speech, which means there is a long and winding path before the technology reaches practical use in medicine or engineering.