A novel biohybrid system, merging a “brain organoid” with conventional AI, achieved a 78% accuracy rate in speech recognition, showcasing the potential for leveraging human physiology to enhance our computational capabilities.
The human brain stands out as the most energy-efficient “computer” globally. While a supercomputer demands 20 mega watts of power to outperform a quintillion calculations per second, the human brain accomplishes the same feat with a mere 20 watts (equivalent to one million watts).
This efficiency has spurred researchers to explore augmenting servers by integrating them with a three-dimensional cluster of lab-grown human brain cells, referred to as an organoid.
In a groundbreaking development, a new hybrid system named “Brainoware” was recently introduced by a team at the University of Indiana Bloomington (UIB).
The researchers at UIB cultivated a brain organoid from stem cells and embedded the tissue amidst thousands of wires. This setup enabled the organoid to receive electrical pulses and record its corresponding neural activity using conventional hardware.
By converting 240 recordings of 8 Chinese speakers articulating syllable sounds into electric pulses, the team demonstrated the potential application of such a system. The neural responses of the brain organoid to the electrical stimulation were utilized to train an AI model to differentiate between the speakers.
Over the course of training, Brainoware’s accuracy improved steadily, culminating in successfully identifying the speaker after two days.
While the organoid necessitated a CO2 incubator and other resource-intensive elements to thrive, Brainoware exhibited lower accuracy in speech recognition compared to a conventional AI-based processing system.
In essence, the program does not represent a leap forward from our existing technology; however, it could serve as a pivotal initial stride towards the advancement of more sophisticated biocomputing methodologies.
Feng Guo, a prominent scholar, shared with Tech Xplore, “We aimed to explore the feasibility of leveraging the biological neural network within the brain organoid for computational tasks. This serves as preliminary evidence of the concept’s viability.”
