Precision and Emergence in the Physics of Life

Information is crucial for life. In many ways, organisms "feed" on information about their surroundings and their internal states just as much as they feed on more conventional nutrients. What we have learned about how living organisms process information has been foundational for the development of many technologies, including the current revolution in artificial intelligence (AI). In this project we will explore two major features of information in living systems. Emergence is the fact that the whole is more than the sum of its parts. Shared information makes it possible for birds to fly together in a flock; for protein molecules in our cells adopt well defined structures; for thousands of neurons in the brain to work together when we think, remember, or imagine; and for our behavior to evolve coherently on time scales much longer than the response times of individual neurons. Precision is the fact that while life may seem random and uncertain, crucial steps are accomplished with extraordinary accuracy and reliability. Every cell in a developing embryo "knows" its unique position, and this knowledge flows through networks of genes that regulate one another; our brains, and those of simpler organisms, draw reliable conclusions from the data collected by our sense organs, allowing us to move comfortably through the world; all of this involves not just responding to what just happened, but making predictions about what will happen next. This project will pursue common physical principles that govern these processes across many scales of biological organization. There are striking connections between these principles and the functioning of the networks that drive modern AI. The current revolution in artificial intelligence (AI) can be traced back to physicists' models for neural networks in the brain. Progress in the experimental exploration of real biological systems has paralleled the growth of AI, and these developments embolden us to ask if the old dreams of general physical principles for living systems might be realizable after all. In this project we explore the emergence and precision of biological functions as a guide to more general principles. The biggest success of AI in science is in protein structure prediction, and we will build a statistical mechanics in the space of structures that allows us to understand how such predictions are possible. We will develop new descriptions for emergent behaviors in networks of real neurons, show how these dynamics can generate long time scales in behavior, and explore commonalities with the sharing of information in flocks of birds. In many cases information processing reaches fundamental physical limits, and there is pressure to squeeze as much as possible out of every molecule in cell, every photon in vision, every spike in the brain. We will explore these examples of optimization in both genetic networks and visual computation, connecting in detail to experiment; running through these examples is the idea that relevant information for the organism is information that helps in predicting the future. Our ideas about optimization and collective behavior have striking parallels in AI systems, most importantly in the landscape for optimization, which allows us to construct a statistical mechanics in parameter space for both AI systems and real biological networks. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.