A biocomputing system comprising living brain cells has acquired the ability to discern a specific individual’s voice amidst hundreds of sound clips.
Researchers at Indiana University Bloomington have used brain organoids, clusters of human brain cells linked to a computer, to perform essential speech recognition. Brain organoids are created by growing stem cells under specific conditions, forming lumps of nerve cells resembling mini-brains. The experiment placed the brain organoids on a microelectrode array to send electrical signals and detect nerve cell activity. The researchers believe such systems could offer more energy-efficient alternatives to silicon chips for AI tasks. However, they acknowledge that much work must be done before developing practical applications.
Guo’s team achieved a notable milestone by utilizing a system to train organoids for problem-solving tasks such as the Hénon map equations, as reported by New Scientist in March. The experiment required the organoids to recognize a specific voice among 240 audio clips of individuals pronouncing Japanese vowel sounds. These sequences were transmitted to the organoids as spatially arranged signals. Initially, the organoids exhibited 30 to 40 percent accuracy, which significantly improved to 70 to 80 percent after two days of training sessions.
Guo emphasized the adaptive learning process observed in the organoids’ training. Their performance was not enhanced when exposed to a nerve cell-inhibiting drug, preventing new connections. The training methodology did not involve feedback mechanisms to inform the organoids of their accuracy, aligning with unsupervised learning in AI research.
Guo highlighted two significant challenges in conventional AI: high energy consumption and the limitations of silicon chips concerning information processing.
Various groups are exploring innovative approaches like biocomputing with living nerve cells to tackle these challenges. For instance, Cortical Labs in Australia made headlines in 2021 by teaching brain cells to play Pong, illustrating the potential of biocomputing in overcoming traditional AI limitations.
The potential role of biocomputing in the future isn’t dismissed by Titouan Parcollet from the University of Cambridge, despite his focus on conventional speech recognition. However, he notes that current deep-learning models outperform any brain in specific tasks, highlighting the remarkable efficiency of existing technologies.
Regarding Brainoware’s functionality, Parcollet suggests its current task is relatively simplified, solely identifying speakers rather than comprehending speech content. He expresses skepticism about its promising prospects in terms of speech recognition.
Guo’s team faces a significant hurdle with Brainoware as well—maintaining the organoids for only one or two months limits their long-term potential. Guo emphasizes the necessity of addressing these limitations to harness the computational capacity of organoids for AI computing fully.
FAQs
1. What exactly is a biocomputing system using brain organoids for speech recognition?
This system involves using clusters of human brain cells, called brain organoids, connected to a computer through microelectrode arrays. The organoids, grown from stem cells, were trained to discern specific voices among numerous sound clips.
2. How were the brain organoids trained for speech recognition?
Researchers transmitted 240 audio clips of individuals pronouncing Japanese vowel sounds to the organoids as spatially arranged signals. Initially, the organoids showed 30 to 40 percent accuracy, improving to 70 to 80 percent after two days of training sessions.
3. What are the challenges in conventional AI that biocomputing with living nerve cells aims to address?
Conventional AI faces high energy consumption challenges and information processing limitations using silicon chips. Biocomputing with living nerve cells, like brain organoids, offers potential energy-efficient alternatives.
[To share your insights with us, please write to sghosh@martechseries.com]
