A bio-computing system comprised of living human brain cells has successfully undertaken speech recognition. In this proof-of-concept, brain organoids—clusters of nerve cells resembling mini-brains—were cultivated over two to three months and integrated with a microelectrode array, forming the system named "Brainoware."
The organoids, consisting of up to 100 million nerve cells, were trained to identify a specific individual's voice from a set of 240 audio clips. Initial accuracy ranged from 30 to 40 percent, improving to 70 to 80 percent after two days of unsupervised learning through repeated exposure to the audio clips.
Feng Guo from Indiana University Bloomington, leading the research, highlights the potential of such biocomputing systems to address challenges in conventional AI, such as high energy consumption and the limitations of silicon chips.
However, he acknowledges the preliminary nature of their achievement, emphasizing the need for further advancements. While Brainoware's performance is limited to speaker identification rather than speech comprehension, there is optimism about the future role of biocomputing in AI.
Guo's team is actively addressing issues like the short lifespan of organoids, aiming to extend their viability for sustained AI computing.
Comments