Investigating the mind’s toolbox
Research Unit 1, SCIoI Project 22
It is often assumed that to solve a cognitive or behavioral task, the mind has a repertoire of different strategies available: the mind’s adaptive toolbox (e.g., Gigerenzer, Hertwig, & Pachur, 2011; Hertwig, Pleskac, Pachur, and the Center for Adaptive Rationality, 2019). The toolbox confers behavioral flexibility and adaptivity—key ingredients of intelligent behavior. Although the assumption of a strategy repertoire is common both in psychology (Siegler, 1994; Payne, Bettman, & Johnson, 1993) and biology (e.g., Collett, 1996; Wilschko & Wilschko, 2012), relatively little is known about the mechanisms underlying strategy selection. Specifically, which strategies need to be assumed to be in an organism’s toolbox? And how is a strategy selected for a given task and environment? According to the framework of strategy selection as rational metareasoning (Lieder & Griffith, 2017), the selection of a strategy from the toolbox is based on a subjective assessment of two key variables, which are traded off against each other: a strategy’s accuracy for the given task and the effort necessary to implement the strategy. According to the working definition of intelligence within the SCIoI cluster, cost-effectiveness is a hallmark of intelligent behavior. But how exactly is cost-effectiveness achieved?
One problem is that the accuracy and cost of a strategy in a given situation are usually not directly accessible to the organism and instead need to be inferred. The goal of this project is to understand ecologically rational strategy selection by focusing on how these crucial inferences for strategy selection are cognitively or computationally implemented—and how they might be improved.
To that end, cognitive scientists and roboticists will work together to identify
a) the features of the task environment that can be used and are used to predict a strategy’s accuracy in a given situation as well as the costs of strategy implementation; and
b) the mechanisms used by an organism to integrate the features when estimating the anticipated accuracy and costs of a strategy.
These questions will be investigated in two contexts: selection of strategies for risky choice in an economic decision task, and through exploration strategies in an Escape Room task, where the objective is to find effective solutions in the environment. The latter case will also be implemented with a robot, allowing us to engage in an analytic-synthetic loop to determine an appropriate weighing of the costs of strategy implementation in humans and a robot.
The project will not only provide insights into commonalities and differences across different domains (decision making vs. exploration) in the type of mechanisms that guide the accuracy and cost assessments for each strategy (e.g., heuristic vs. non-heuristic); it will also carve out key components of intelligent behavior and how they are and can be implemented both in real and artificial systems.