Human brains are masters at recognizing patterns — even when those patterns are meaningless. This “face” on Mars, captured by NASA’s Viking 1 in 1976, is not evidence of alien life, as some speculated, but rather, simply a Martian mesa. (image: NASA)

To advance research on topics from climate change to machine learning, scientific models are crucial. These models often reveal patterns, but humans also have a tendency to see patterns everywhere, even where there are none. How can researchers recognize which patterns are real and which ones are not? Which kinds of real patterns are most useful to science?

These are some of the questions that philosophers and scientists from various disciplines explored in a virtual SFI workshop on “Real Patterns in Science and Cognition” held February 28 – March 4. The workshop was organized by SFI Postdoctoral Fellow Tyler Millhouse, along with SFI External Professor Daniel Dennett (Tufts University), Don Ross (University College Cork, University of Cape Town, and Georgia State University), and Steve Petersen (Niagara University)

Dennett first introduced the concept of real patterns in 1991. “Since then, it’s sort of slowly been building up steam, and people have been more and more interested in applying it to different areas of research,” says Millhouse. “The workshop was designed to bring together people whose work either does try to extend real patterns in this way or is adjacent to it.”

The researchers came away with fresh insights into the connections between their lines of inquiry and a more nuanced understanding of real-patterns thinking. For example, four participants from different disciplines, including Millhouse all spoke about coarse-graining — finding ways to simplify complex data so that it can be more easily understood — but only one of those talks directly dealt with real patterns. Yet “there was surprising convergence in our talks,” Millhouse says. “In particular, we all saw, in different ways, this process of coarse-graining as vital to revealing important patterns in the world, and we were able to share and learn from quite different examples of how this happens — from patterns of social dominance in non-human primates to the patterns in machine learning datasets. These are examples I will draw on in my own work, and they will add important nuance and breadth to the way I think about real patterns.”


This meeting was supported by two grants from the National Science Foundation — AI Institute: Planning: Foundations of Intelligence in Natural and Artificial Systems, NSF Grant Number 2020103; and EAGER: Developing Data and Evaluation Methods to Assess the Generality and Robustness of AI Systems for Abstraction and Analogy-Making, NSF Grant Number 2139983