Home Featured The recognition of speech can be achieved by connecting human brain cells with a computer.

The recognition of speech can be achieved by connecting human brain cells with a computer.

The recognition of speech can be achieved by connecting human brain cells with a computer.

The greatest technological companies in the world are all fixated on creating computers that can mimic the biological computers found in our brains. However, the sophisticated artificial intelligence systems of today are expensive to construct and power hungry. Researchers think a novel, more effective approach to working with AI may be provided by organic computers. An Indiana University team has shown how that could function. In the lab, Feng Guo and his colleagues developed small brains, which they then used to train an AI system to identify speech.

While silicon-based computers may do mathematical operations more efficiently than the human brain, the latter is still superior to the former in complicated information processing and comprehension. The Indiana University group set out to investigate the viability of integrating electronic and biological processing in a hybrid system.

They came up with a system called Brainoware. A brain organoid is a tiny, spherical glob of lab-grown neurons that is considered the biological component. An organoid has 100 million brain cells, or roughly 1,000 times fewer than a human brain, and can measure up to several millimeters in diameter. The “mini-brains” were positioned atop a microelectrode array, which allowed the nerve cell clusters to transmit and receive electrical impulses. This made it possible for Brainoware’s technological and biological components to communicate.

The organoids respond to electric stimuli by modifying their neural network architecture. To demonstrate the system’s ability to process data, the team created a basic test. They sent 240 electrical signal recordings of eight persons pronouncing Japanese vowels to Brainoware. Initially, the algorithm was only able to recognize a single voice 30–40% of the time. The network improved with time, achieving 70–80% accuracy.

Since no data was pre-labeled, this is an example of unsupervised learning by the researchers. The tiny brains do seem to be learning, too. According to New Scientist, the accuracy did not increase when medications that prevent the development of new synapses were given to the grown cells.

Although it’s a remarkable outcome, Brainoware is only a proof of concept. For the time being at least, this method of calculating is far slower and less precise than traditional computing. The cells could be able to recognize a voice, for instance, but deciphering what it says will be a another story. This study, which was published in Nature Electronics, may lead to the development of organic computers that are more effective in the future.


Please enter your comment!
Please enter your name here