Artificial intelligence theory could be the key to how collective cell intelligence works

Tokyo, Japan – It has long been understood that groups of cells can perform complex tasks, such as navigating mazes or strategically colonizing new habitats, even though individual biological cells have only limited ability to respond to signals like chemical compounds in their immediate environment. Now, scientists from Japan have developed a theoretical framework that may explain how surprisingly intelligent behavior arises in nature from such groups.

A research team from the Institute of Industrial Science, The University of Tokyo, found that the key is how the cells use their environment to incrementally process information and make decisions in a distributed manner.

 “We can describe these phenomena in detail using well-known physical models,” says Masaki Kato, the first author of the study. “But understanding the computational principles at play is another matter.”

To do this, the research team used the paradigm of reinforcement learning applied in artificial intelligence. Reinforcement learning, unlike supervised and unsupervised machine learning, is ideal in this case because it is based on interaction with the environment. Instead of operating to a set of pre-defined instructions, an individual agent simply probes the environment multiple times and sees what happens. It then adjusts its internal policy to maximize its reward over the long term.

The team considered a population of cells that cooperatively aim to move toward sparsely distributed targets (such as food) by modulating chemical signals that indicate a target is nearby. The whole cell population acts as an agent that uses reinforcement learning to gradually determine the optimal navigation strategy without needing the guidance of a single leader.

Tetsuya J. Kobayashi, the study's senior researcher, says that this theory is the key to understanding how this type of distributed information processing works. “In a certain sense, the environment plays the role of a working memory for consequences; it remembers and reflects the agent’s past actions and exploration in the form of changed states.”

Through simulations, the researchers demonstrated that even agents with limited intelligence can perform complex tasks as a group, such as finding their way through a maze, through decentralized information processing and sharing (that is, without a leader). They compared these agent populations with a single, more intelligent agent with a working memory and found that the simple organisms performed more robustly than the single agent.

This work shows that decentralized swarms or teams of simple agents can coordinate to efficiently process information, a principle that could be used to address problems in various fields including medicine, artificial intelligence, and robotics going forward.

###

The article, “Optimality Theory of Stigmergic Collective Information Processing by Chemotactic

Cells,” was published in PRX Life at DOI: 10.1103/tvfy-lbbl.

About Institute of Industrial Science, The University of Tokyo

The Institute of Industrial Science, The University of Tokyo (UTokyo-IIS) is one of the largest university-attached research institutes in Japan. UTokyo-IIS is comprised of over 120 research laboratories—each headed by a faculty member—and has over 1,200 members (approximately 400 staff and 800 students) actively engaged in education and research. Its activities cover almost all areas of engineering. Since its foundation in 1949, UTokyo-IIS has worked to bridge the huge gaps that exist between academic disciplines and real-world applications.