Mini brains from the laboratory decipher letters in Braille

The computing power of mini-brains grown in the laboratory can be used to decipher individual letters of Braille for blind people. This is reported by a research group led by Tianyi Liu from the British University of Bristol in the specialist magazine “Frontiers in Artificial Intelligence”. Several mini-brains connected together achieved better results than a single mini-brain.

The experiment was carried out in such a way that a sensor first scanned a Braille letter and translated it into a specific sequence of electrical impulses. This sensor data created a specific activity pattern for each letter in the mini-brains, which is currently still evaluated and assigned by a conventional computer.

"We can take information from a real touch sensor, put it into a form that a living neural network can receive, and then recover meaningful information from the organoid's response," Liu explained. The Braille codes for 26 letters served as a starting point.

When the letter H was sampled, a mini-brain should show a specific pattern of activity. This was the case in an average of 61 percent of the experiments for a single organoid. However, when three mini-brains were connected together, the accuracy was 83 percent.

Together, the mini-brains are more accurate

The researchers don't know exactly why the three mini-brains perform better than one. “Different organoids may have slightly different network connections, developmental stages, and intrinsic activities, so the same input can elicit similar but not identical responses,” Liu explained. The team suspects that the three organoids could possibly complement the individual reactions and strengthen tendencies.

The scientists also wanted to find out how robust the processing by the brain organoids is. Therefore, they changed the input pulses with various disturbances such as noise, missing data and outliers.

When subjected to noise interference, the ensemble of three organoids maintained near accuracy using a clear signal. Missing data reduced accuracy to 72 percent and outliers to 68 percent.

The work is considered a further step on the way to extremely energy-efficient biocomputers. The researchers' goal is to one day use biological nerve networks for certain tasks in which classic computers with their silicon chips reach their limits.

"The next step is to go beyond Braille; we are interested in aspects such as textures, edges and different types of touch, and eventually closed-loop tasks where the organoid receives feedback on its own actions," Liu said.