Iris van Rooij - How Cognitive Scientists are Not Dealing with Intractability (But Could Be)

Many computational models of human cognition are known to be intractable (e.g., NP-hard). This means that, even if the models may be able to describe human behavior in simplified lab settings, the postulated computations do not seem to scale to situations of real-world complexity. This theoretical problem is not specific to any particular cognitive domain or to any particular modeling framework. For instance, intractable models can be found in the domains of vision, decision making, reasoning, analogizing, similarity, action planning and communication; and they can be found in the symbolic, connectionist, probabilistic, and dynamical modeling paradigms.  

Given the pervasiveness of intractability, how are cognitive scientists dealing with it? I will argue in my talk that cognitive scientists are not dealing with it in a coherent and productive way. Moreover, by not facing the problem head-on, cognitive science may be missing out on an opportunity to probe the unique ways in which human minds/brains cope with the complexities of the world that they live in. I present illustrative examples of an approach that utilizes intractability rather than sweep it under the rug. I will explain how this approach may lay the foundations for the important goal of scaling models of cognition from the lab to the real world.