Organisms use a variety of senses to glean information and make predictions that are critical to their survival. A better understanding of sensory prediction in living systems could help improve prediction in other realms. (image: Christian Lue/Unslpash)

Prediction is a key part of complex systems, in a wide variety of fields. Physicists and mathematicians use prediction performance to evaluate their models of mechanical systems; engineering prediction algorithms can inform the design of complicated devices. Prediction is also integral in artificial intelligence, in large language models like ChatGPT, which are designed to predict a word or words that follow from a prompt. 

But living organisms use prediction, too, and it’s critical to life itself. They must predict what actions will lead to food, or how changes in the environment will affect their well-being.

“Prediction is really important for everything organisms do,” says Claremont McKenna College physicist Sarah Marzen, a co-organizer of a July 10–14 workshop at SFI called “Sensory Prediction: Engineered and Evolved.”

“A lot of what biological systems have to do to survive is to predict,” says co-organizer James Crutchfield, an External Professor at SFI (University of California, Davis).

Crutchfield and Marzen both focus on ways to use ideas from physics to predict how organisms make predictions — which is often more complicated than mechanical systems. They note that if researchers can better understand prediction in living systems, they could build models of those decisions and use that knowledge to inform prediction algorithms in other areas. 

Their workshop explored this crossroads. The meeting brought together neuroscientists, physicists, computer scientists, mathematicians, biologists, and others to explore how a better understanding of predictions from biological sensory systems might influence prediction in application to other scientific domains.

Marzen says the workshop’s goal was to develop a unifying framework for sensory prediction, including both a way to measure efficient prediction and some theoretical understanding for how it made predictions. The way the researchers framed the questions that motivated the workshop, says Crutchfield, largely emerged from the physical sciences, but the work could provide new insights in theoretical biology, neuroscience, and other disciplines.