The challenges of problem solving do not exclusively lie in how to perform heuristic search, but they begin with how we understand a given task: How to cognitively represent the task domain, its components, and the (sub-)goals an agent tries to achieve can determine how quickly progress towards a solution is being made, whether advanced strategies can be discovered, and sometimes even whether a solution can be found at all.
Especially for more complex task domains, there can be a wide variety of potential representations, and it might even be beneficial to make simultaneous use of several ones, or to change them during the problem solving process. While this challenge of constructing and changing representations has been acknowledged early on in problem solving research, for the most part it has been sidestepped by focussing on simple, well-defined problems whose representation is almost fully determined by the task instructions. Thus, the established theory of problem solving as heuristic search in problem spaces has little to say about these issues.
However, over time there have been many developments in related fields which might play a role in addressing this impasse, e.g. the mechanisms of analogy-making, metaphor use, and explanation, the contribution of affective, bodily and environmental factors to cognitive processes, and the development of more complex and ecologically valid experimental tasks eliciting a broader range of behaviour. Yet, so far few of these research fields have integrated their insights into a common problem solving theory, leaving the issues of representation still largely unaddressed.
In this symposium, we bring together researchers working on insight, metaphors, strategy discovery, and explainable AI in order to reflect on the current and future state of problem solving theory and research with respect to understanding the mechanisms of flexible and dynamic representations in humans, artificial systems, and their interaction.