Cognition with 8 Bits and 15 Watts: Thinking Effectively on a Tight Budget Date: February 4, 2026
-----
Human cognition is remarkable both for its powerful, broad, and flexible inferential abilities—and for the fact that the systems supporting these abilities are highly bandwidth-limited. In this talk, Dr. Cheyette argues that two mechanisms help reconcile these facts: adaptive online resource allocation processes and the ability to form and reason over highly compressive symbolic abstractions. He first shows how efficient resource allocation processes baked into our visual system explain key properties of our intuitive numerical abilities, including why we represent small quantities precisely and large quantities increasingly imprecisely. He next describes work showing how people can compress observations into highly compact “program-like” internal models, which enable flexible generalization from sparse data. Finally, he examines how people actively learn in domains that exceed their representational capacity, showing how apparent sub-optimalities in human inquiry emerge naturally from efficient resource allocation operating over abstract symbolic representations. Together, this work identifies principles enabling broad, flexible cognition with limited resources.
-----