Meeting Summary: Pattern recognition is a key component of animal cognition, artificial intelligence, and scientific reasoning. For this reason, it is essential to understand what patterns are and how they can be recognized and characterized. One important role for talk of "patterns" is in identifying what it is in general that models model. For example, we might say that an animal's model of dominance relations within its group captures social patterns or that a scientist's model of population dynamics captures biological patterns. In the philosophy of science, this approach is most closely identified with Dennett's (1991) work on "Real Patterns." Dennett's approach has several significant advantages. First, it connects a wide range of phenomena to a rigorous body of work in information theory that seeks to formally characterize patterns. This work suggests objective criteria for the reality of patterns, criteria that are not beholden to the goals or limitations of particular modelers. Second, talk of "patterns" promises to be sufficiently flexible to accommodate the full breadth of phenomena we might hope to model (from mechanisms and processes to protein shapes and crystalline structures). This flexibility also allows us to evaluate patterns at different levels of abstraction and different levels of idealization. For example, we might think that an economic model gets something right about economic patterns even if the agents in the model do not correspond to any real-world agents, either numerically or qualitatively.
That said, there are many open questions about how best to understand the idea that the models employed by agents and scientists capture (or attempt to capture) patterns in the world. These questions include: If patterns can be evaluated at different levels of abstraction, how can we distinguish models that capture a high-level pattern from models that capture a low-level pattern? More concretely, how can we distinguish cases where we should take the details of a model seriously as description of our target system from cases where we can only expect an abstract correspondence? Moving on, which information theoretic tools are best for evaluating the reality of patterns? Are different approaches to patterns appropriate to different modeling contexts (e.g., an agent in the wild vs. scientists in the lab)? What consequences (if any) does a theoretical account of patterns have for pattern recognition as a cognitive process or for cognition in general (esp. where abstraction is involved)? How should we translate abstract criteria for the reality of patterns into practical criteria for the evaluation of real-world models? What role (if any) is there for patterns in understanding emergence and explaining the relationships between levels in science? The proposed workshop will bring together philosophers and scientists to tackle these questions and to improve our understanding of both science and cognition.