Summary: New research links cognitive science and information theory, showing that human perception is organized to make the best possible decisions under resource constraints.
Source: Rensselaer Polytechnic Institute.
Imagine driving on a dark country road and spotting a shape by the roadside. Is it a deer or a mailbox? According to new research that brings together cognitive science and information theory — the mathematical foundation of modern communications — our brains are organized to make the best possible decision given limited resources.
This result, published in the journal Science, builds on National Science Foundation-supported work aimed at improving STEM pedagogy, where perceptual skills and expertise are often essential, said Chris Sims, assistant professor of cognitive science at Rensselaer Polytechnic Institute.
“Consider geology, where students must learn to distinguish rocks or formations that look very similar,” Sims said. “Traditionally this learning happens through repetition: you look at many examples until patterns become familiar. But a clearer understanding of how perception works can guide the design of more efficient classroom training that teaches those skills faster and more reliably.”
A foundational principle in cognitive science — the Universal Law of Generalization, first articulated in a 1987 Science article — states that perceptual decisions depend on how similar a new stimulus is to prior experience. The law predicts that the likelihood of applying a past response to a new stimulus decreases exponentially as the perceived similarity declines. This exponential generalization gradient has been observed repeatedly across species, including humans, pigeons, and honeybees.
“The Universal Law describes a robust empirical pattern,” Sims said. “But it left an open question: why does this pattern appear so consistently in nature? My work set out to explain that.”
Sims turned to information theory, a mathematical discipline developed at Bell Labs in the 1940s that quantifies the limits and optimal performance of communication systems. Just as information theory can predict the best achievable voice quality over a noisy telephone line, it can also describe how a biological system like vision transmits information amid internal noise.
“You can think of visual perception as a communication channel,” Sims explained. “There is information in the world that must be transmitted from the eyes to the brain. But biological systems face limits: neural signals are noisy and processing resources are finite. Information theory provides a framework to predict the best performance possible under those constraints.”
Modeling the visual system within this framework produced an unexpected connection: a principle from information theory known as efficient coding yields the same exponential generalization pattern predicted by the Universal Law of Generalization. In other words, if a perceptual system minimizes the cost of error while constrained in its information-processing capacity, the resulting generalization behavior naturally follows the universal exponential form.
This finding links two foundational ideas from different disciplines and suggests that evolutionary pressures have shaped perceptual systems to operate near the theoretical optimum predicted by information theory.
“I set out to explain why the exponential pattern appears in nature,” Sims said. “Information theory offers a simple account: perceptual systems are as efficient as they can be given their limitations. That explanation helps us understand why the same pattern shows up across species and tasks.”
Beyond advancing basic science, the result has practical implications. Quantifying perceptual efficiency could improve how educators measure and train perceptual expertise and progression in fields that rely on fine-grained perceptual judgments.
“I’m pleased that we now have mathematical laws to better describe how the brain processes information and to clarify the nature of intelligence more broadly,” Sims said.
Funding: The study was supported by the National Science Foundation.
Source: Mary Martialay – Rensselaer Polytechnic Institute
Publisher: Organized by NeuroscienceNews.com.
Image source: Image credited to Rensselaer.
Original research: Abstract for “Efficient coding explains the universal law of generalization in human perception” by Chris R. Sims, published in Science on May 11, 2018.
DOI: 10.1126/science.aaq1118
Suggested citation formats and publisher metadata are available from the original institutions and journals. Please cite the original Science article by Chris R. Sims for scholarly use.
Abstract
Efficient coding explains the universal law of generalization in human perception
Perceptual generalization and discrimination are fundamental cognitive abilities. For example, if a bird eats a poisonous butterfly, it will learn to avoid preying on that species again by generalizing that past experience to new visual stimuli. In cognitive science, the “universal law of generalization” states that generalization between stimuli follows an exponential function of their distance in psychological space. This study challenges prior theoretical accounts and proposes an alternative explanation based on the principle of efficient coding. It demonstrates that the universal law arises inevitably in any information-processing system — biological or artificial — that minimizes perceptual error costs while operating under constraints on information processing or transmission.
This summary is provided for educational and informational purposes. For full experimental details, formal proofs, and data, refer to the original research publication in Science.