The human brain uses energy very economically. Silicon processors that try to imitate it already provide savings compared with conventional ones, but scientists are looking for new paths. One of them is biohybrid computing, that is, the combination of living nerve cells with electronics. A group from the University of Bristol has taken the first step: it tested whether organoids - laboratory-grown clusters of nerve cells that reproduce some features of the forebrain - can receive and transmit signals. The work was published in Frontiers in Artificial Intelligence.
Organoids less than a millimeter in size lack many structures of a real brain, but they contain neurons and supporting cells. They are placed on multielectrode arrays - plates with many electrodes that allow both stimulation and reading of electrical activity.
To test whether organoids are capable of perceiving signals, the scientists took a tactile sensor that records touches and passed Braille font letters through it. Braille is a raised-dot font for the blind, in which each letter is encoded by a combination of raised dots. The sensor data were converted into sequences of electrical impulses and fed to the organoids. Then, using a simple machine learning algorithm - a support vector machine, which looks for a boundary between data classes - they checked whether the letter that had been presented could be reconstructed from the cells' response activity.
One organoid produced a signal that could be correctly decoded in approximately 61% of cases. When the readings of three organoids were combined, the accuracy rose to 83%. Why three are better than one, scientists do not yet know: possibly different organoids have their own features of connections and development and complement each other. "Different organoids may have slightly different connections, developmental stages and current activity, so the same input produces similar but not identical responses," explained Tianyi Liu, the first author of the work.
To make the task more difficult, the researchers added artificial noise to the signals. The algorithm that analyzed the organoids' responses handled letter recognition better than the same algorithm working directly with noisy signals without the participation of living cells.
However, the authors emphasize: the experiment does not show that organoids are capable of learning or changing over time. It has only been proven that they receive tactile information, process it and produce a signal from which the letter can be reconstructed. "We showed that tactile information can be encoded, processed by organoids and decoded, but we did not demonstrate learning or long-term changes," Liu noted. The next steps are tests with textures, edges and other types of touch, as well as feedback tasks in which the organoid receives a response to its actions. This will make it possible to test whether it can adapt to experience.
The research is laboratory-based; finished computers with living "brains" are still far away. But it outlines a practical path for connecting artificial sensors with biological neural systems, which could become the basis for more energy-efficient computing.