Artificial intelligence calculations by Osaka University and the University of Tokyo using water and ions provide path to new devices

Electrochemical currents in solution allow for the implementation of reservoir computing, which has recently attracted attention as a next-generation information processing system. The research group comprising Professor Megumi Akai of Osaka University, Graduate School of Information Science and Technology, doctoral student Kan Shaohua, and Associate Professor Kohei Nakajima of the University of Tokyo, Graduate School of Information Science and Technology, presented these results for the first time. It was demonstrated that solution materials containing ions could be used to develop safe and inexpensive artificial intelligence devices. The study was published in Advanced Science.

Artificial intelligence is expected to play an active role in all strata of society. It is expected to not only solve complex problems but also make accurate judgements that are close to the that of people and nature. Implementation of artificial intelligence is also expected to dramatically reduce the enormous amount of information processed by the modern society. Nonetheless, artificial intelligence based on neural networks requires enormous electric power. This not only hinders the development of artificial intelligence, but also creates a situation where artificial intelligence can be utilized fully only by connecting to a large computer.

To address the power problem and utilize artificial intelligence in various fields, it is necessary to construct a device that can perform nonlinear information processing, such as reservoir computing, and utilize artificial intelligence from the existing computer. The research group has developed a system that automatically and quickly measures the transient response current generated from the electrochemical reaction in a solution and monitors the movement of ions at more than 100 points by combining multi-terminal electrodes in the aqueous solution and the electronic circuit. Nonlinear time series problems can be efficiently solved by reservoir computing. When the measured signal was applied to a benchmark task for such problems, strong computational ability was observed. The acidic molecules in the aqueous solution underwent an efficient redox reaction, which improved the processing capacity of the periodic signals. However, the computational capacity was insufficient because it required a high-order nonlinear memory. Although the processing capacity of the periodic signals was low in the solution containing only distilled water, strong computational power was observed for solving high-order nonlinear memory problems.

This system uses the current generated not only from the electrochemical reaction but also from the charging and discharging of the electric double layer at the interface between the solution and electrode. Therefore, the possibility of realizing a revolutionary ionic-based artificial intelligence device in which electrons hardly flow from the electrode to the material was shown. The group has also shown that a safe and inexpensive artificial intelligence device with low power consumption can be realized using water and ions.

Professor Akai said, "Research always progresses in unexpected directions. This study was also initiated based on the mere idea of a student. As a final outcome, we realized that signals from compounds such as water and ions, which were not expected at all, showed high information processing ability. Research progresses through encounters with people and unexpected encounters with materials. We do not know how many years it might take, but we dream of building a computer that is made of organic materials similar to those in the human brain and that works by transmitting ion information."