Everyday life often gives us new problems to solve. For example, imagine flying back from a relaxing vacation and your return trip includes a layover in a new city. You’ve just landed at your connecting destination but learn that the next flight has been canceled. Suddenly, you need to start considering a whole new set of possibilities, constraints, and goals: When is the next flight? Would it be nice to stay a few days and explore? When do I have to be back at work? Could I take a car or train home instead? How long would that take? How much would it cost? Then, you would actually need to do something based on these considerations. Put another way, you need to come up with and then execute a novel plan. 

 

Our capacity to flexibly plan and act in new situations (such as deciding what to do when a connecting flight gets canceled) is both quotidian and remarkable. This is because planning is an exceptionally cognitively demanding process, and yet we are able to plan successfully despite having limited attention, memory, and time. No other species and no existing artificial intelligence system can plan as flexibly, generally, and efficiently as we do on a routine basis. What explains our ability to plan? 

 

To understand processes like planning, cognitive scientists often develop formal, mathematical models that capture general principles underlying how people think and act (similar to how Newton’s laws capture general principles underlying how physical objects interact). One of the earliest and most influential models of planning, proposed by Newell and Simon (1972), is based on the idea that people have a representation of a task (e.g., for chess, an understanding of how the pieces move and how to win) and then search for good actions to take using that representation (e.g., by simulating and evaluating different sequences of chess moves). Recently, we built on this original distinction between representing tasks and searching for actions by formulating a new theory of planning. Specifically, we propose that people not only search for good actions, but also search for good task representations via a process that we refer to as value-guided construal. 

 

Our account of value-guided construal allows us to pose two broad questions—one conceptual and one empirical. First, the conceptual question: What makes a task representation good? We propose that a task representation is good if it is simple and planning with it leads to a successful solution. Put another way, good task representations balance utility and complexity. Our computational model allows us to quantify this tradeoff for a given task. 

 

The second question is empirical: Do people form task representations that balance utility and complexity? Answering this question requires data, so we designed a task in which participants needed to plan how to navigate around obstacles in 2D mazes to reach a goal. We wanted to understand how they represented the mazes to plan, and so in a series of experiments, we assessed their memory of task elements after planning using post-trial questions as well as their thought process during planning using mouse tracking. Across these experiments, we found robust support for the idea that people trade off the utility and complexity of a task representation, as predicted by our account of value-guided construal. 

 

Our results tell us whether people flexibly form task representations that trade-off utility and complexity, but they also raise intriguing questions about how people might do so. The current work does not provide a definitive answer to such questions, but we explore how it might be done in principle. For example, one could start by solving the simplest construal and only add in new details (such as obstacles) as needed. Background knowledge about a domain could also guide which details to include in a construal—for instance, in chess, knowing that the queen is the strongest piece on the board might bias one to consider construals that include it. 

 

The human capacity to plan in novel situations is both familiar and puzzling, and understanding what makes it possible remains an important scientific challenge. The framework of value-guided construal provides a new perspective on this old question by drawing attention to the role that actively simplifying tasks plays in allowing people to more efficiently use limited cognitive resources. Our hope is that this work can serve as a starting point for further examination of the construal process itself, as well as other cognitive mechanisms that interface with planning.