Learning without limits : from problem solving towards a unified theory of learning

The goal of this thesis, as stated in chapter 1, is the development of a theory of problem solving that is psychologically plausible, and does justice to its complexity. The weak-method theory that stems from artificial intelligence acknowledges this complexity, but shows only very limited correspondence with human data. Theories from experimental psychology, on the other hand, neglect the complexity of problem solving, and theorize about how complex skills can be explained in terms of memory systems and basic information processing. In the course of this thesis, the subject of problem solving has been broadened to cognitive skills in general, since there is no principled distinction between the two. The approach I have chosen is to focus on learning cognitive skills. In chapters 5 to 7, a view of skill learning has taken shape, which I will try to explicate in this final chapter. This view is built upon the foundation the ACT-R architecture offers, ensuring a plausible system of learning and memory, and it exhibits the complexity of behavior that artificial intelligence is interested in. The theory of skill learning that emerges has a number of areas that need more investigation, and a number of possible applications. Both topics will be discussed in the final sections of this chapter. Have I solved the problem of NP-completeness? The answer to this question is of course “no”. But I have tried to draw a picture of how humans can acquire the knowledge to at least partially solve hard problems. This is not a simple picture, and cannot be summarized in a clear-cut algorithm. When people start with a new task, they do not use a general machine learning algorithm to acquire knowledge about this task. Rather, they use a set of strategies, knowledge about other domains and tasks, instructions about the current task to start with, and keep adapting and refining their knowledge while they are working on the task. A better understanding of these processes is the key to understanding complex human problem solving.