Computational Models of Insight Problem Solving

People are often fascinated by the sudden, unexpected and apparently unintended solution of a difficult problem which pops into someone’s mind and often is accompanied by an aha!. For more than 100 years psychologists address the nature and underlying mechanisms of insights by behavioral experiments on human problem solving. It was found that re-structuring of the given problem elements or the goal representation provide important cognitive mechanisms which help to overcome an impasse. It is speculated that mainly unconscious processes drive re-structuring to relax self-imposed constraints and extend the search space for candidate solutions. What here exactly happens is still an open and unsolved question. Since behavioral and neuro-scientific studies allow only a coarse and investigation of cognitive mechanisms characterizing insight problem solving, we propose to exploit the potential of computational models. Those models might provide hypotheses and simulations which help to attain a better understanding of insight processes. In particular, all models need to address the questions: How existing information is recombined, in order to serve the overall goal? How the concerted interplay between unconscious and conscious processes contributes to find the solution? How the system realize that an impasse is meet and how to overcome the impasse? And what role has the aha! experience as an emotional correlate of the problem solving process?

The workshop will introduce the state of the art on insight problem solving. We will discuss three different computational approaches (connectionism, evolutionary theory, and rational models) that shed light on these questions from different perspectives. We will discuss strengths and weaknesses of each approach and develop further perspectives.

Workshop presenters

Anna Fedor

Institute of Advanced Studies

Köszeg, Hungary

Hermish Mehta
Electrical Engineering and Computer Science
University of California at Berkeley, USA

Michael Öllinger (chair)

Parmenides Center for the Study of Thinking

Pullach, Germany

Ron Sun

Cognitive Science Department

Rensselaer Polytechnic, New York, USA

Anna Fedor

Unconscious Search Through Evolutionary Processes in the Brain

Darwinian Neurodynamics is a theory that suggests that evolutionary processes in the brain play an important role in human cognition: they can generate new ideas in a matter of a few minutes through dynamic neuronal changes. The theory explains how humans search the problem space when trying to solve a difficult problem and how they come up with new candidate solutions.

These evolutionary processes can be implemented in a connectionist model, which can solve insight problems. Candidate solutions are represented as activation patterns of a population of attractor networks. New variation is generated by error-prone copying of activation patterns and activation patterns are selected based on their fitness, i.e., closeness to the solution. Our hypothesis is that real evolutionary processes come to play a role during the impasse phase of problem solving, when conscious thinking gives way to unconscious thought processes and parallel computing. We are working on a more realistic model too, where the units of evolution are continuous-time recurrent networks and their complex firing rate patterns.

Besides computational modelling, I will explain how the theory of Darwinian Neurodynamics is supported by human behavioral experiments and experiments with in vitro neural cell cultures.

Anna Fedor is a postdoctoral researcher who combines theoretical (mainly connectionist) modelling and human behavioral experiments to investigate various topics in cognitive science. She has an MSc and a PhD in evolutionary and theoretical biology from Eötvös University of Sciences, Budapest; has done postdoctoral research at Birkbeck College, London, and Parmenides Center, Munich. She has published papers on language evolution, linguistic recursion, developmental language disorders and insight problem solving, among others. She has also taken part in the Reproducibility Project: Psychology and ManyLabs 5, and she is an enthusiastic proponent of reproducible science.

Hermish Mehta & Rachit Dubey

Explaining the Emergence of Aha Moments as a Consequence of Resource Constraints

Psychology has long been fascinated with understanding the nature of aha moments, moments when we transition from not knowing to suddenly realizing the solution to a problem. In this talk, I will present a resource-rational model that explains when and why we experience aha moments during problem-solving. Our theory posits that during problem-solving, in addition to solving the problem, people also maintain a meta-cognitive expectation about how quickly they will solve that problem. Crucially, aha moments arise whenever we experience a positive surprise i.e. when we solve a task much faster than we expected to solve. In essence, our theory suggests that due to resource-constraints, people routinely maintain expectations about time to finish any task and aha moments emerge as a consequence of maintaining such expectations. I will also present preliminary behavioral evidence that suggests aha moments are indeed a form of meta-cognitive reward prediction errors.

Hermish Mehta is an Electrical Engineering & Computer Sciences student at the University of California, Berkeley and is part of Tom Griffith’s Computational Cognitive Science lab at Princeton University. His research interests lie at the intersection of theoretical computer science, machine learning, and cognitive science. Prior to his work on insight, he spent time doing research on curiosity and designing approximation algorithms.

Michael Öllinger

From Gestalt to Computational Models

The Gestalt psychologists founded the experimental research on insight problem solving. They introduced the concept of re-structuring as the key concept for solving difficult and unusual problems. Their ideas informed cognitive models up to now. From a computational perspective modeling self-driven re-structuring represents the greatest challenge. We will build on the Gestalt concept and presenting the most important developments and theories. We focus on the interplay of unconscious and conscious cognitive processes during problem solving and summarize accounts and proposals which address re-structuring within computational models. We will discuss implications and open questions, and close with next steps.

Michael Öllinger is research scientist at the Parmenides Center for the Study of Thinking and lecturer at the Ludwig-Maximilians-University, Munich. He is an experimental cognitive psychologist, who has been investigating insight phenomena from several perspectives. He hosted a research topic on insight and intuition, see https://www.frontiersin.org/research-topics/4310/insight-and-intuition---two-sides-of-the-same-coin.

Ron Sun

How Insight Emerges: An Account Based on a Cognitive Architecture

We show computationally how implicit processes lead to the emergence of sudden insight. Human creative problem solving has been tackled using computational modeling and simulation based on the Clarion cognitive architecture. Clarion, in general, attempts to provide unified explanations of a wide range of psychological phenomena using a number of basic principles. With these principles, a framework for understanding creative problem solving was generated, which includes both incubation and emergence of insight. Beyond that, the framework can also account for effects of a number of other factors on insight in creative problem solving (including motivation, personality, emotion, and so on).

Ron Sun is Professor of Cognitive Science at Rensselaer Polytechnic Institute. His research interests center around the study of cognition, especially in the areas of cognitive architectures, human reasoning and learning, cognitive social simulation, and hybrid connectionist-symbolic models. He published many papers and books, including Anatomy of the Mind and Cambridge Handbook of Computational Psychology. For his paper on integrating rule-based and connectionist models for accounting for human everyday reasoning, he received the David Marr Award from the Cognitive Science Society. For his work on human skill learning, he received the Hebb Award from the International Neural Network Society. He is a fellow of IEEE, APS, and other professional societies. He was President of INNS 2011-2012.

His Web URL is http://sites.google.com/site/drronsun