‘Brainoware’: Brain cells integrated with AI can recognize sounds.

Scientists have found a way to integrate living human brain cells with computer systems, and it could age out the “A” in AI.

A research paper published today in the journal Nature Electronics by scientists at Indiana University Bloomington describes a new system called “Brainoware” that uses human brain organoids to perform advanced AI functions. These organoids—artificially grown cells or tissues that resemble organs—are now mounted on massive multielectrode arrays, and are among the oldest today. However, researchers hope that their use will pave the way for biocomputers that can perform the same tasks as computers with minimal energy consumption.

“The human brain typically consumes 20 watts, while current AI hardware consumes about 8 million watts,” the research paper notes. Because it can provide quick learning.

“The human brain uses much less energy and learns faster, so some researchers see biocomputing as the way forward,” Michael Le Page tweeted in March, but he noted that pushing the field to its limits raises thorny questions.

Le Page quoted Cambridge developmental neurobiologist Madeleine Lancaster as saying, “These are things that we definitely want to avoid if they push the ethical boundary, and the scientific and ethical community is coming together to decide where that boundary is.”

Brainware sends and receives information from brain organoids through “adaptive reservoir computation.” This method enables unsupervised learning from training data, which can still model the functional connectivity of the organelle. The system's practical capabilities have been demonstrated in tasks such as speech recognition, which after training can accurately identify the voices of individual speakers.

For example, the organoids were trained to recognize an individual's voice in 240 audio clips of eight people speaking Japanese vowel sounds. After the training, the organoids can complete the task with more than 70% accuracy.

However, science is still a long way from making living robots. Organoids only identify the speaker, not understand the speech, which means there is a long and winding road before the technology finds practical use in medicine or engineering.

Titouan Parcolet of the University of Cambridge told New Scientist magazine that the potential of biocomputing is vast, but “current deep learning models are really different from any brain and better at their target tasks.”

“Existing organoids still suffer from high heterogeneity, low generation output, necrosis/hypoxia, and heterogeneity,” the researchers cautioned.

Parallel to the development of Brainoware, AI has been innovatively applied in fields such as healthcare. All together, these developments highlight the versatile and dynamic nature of AI technologies.

Edited by Ryan Ozawa.