Humans account for cognitive costs when finding shortcuts: An information-theoretic analysis of navigation

Abstract

When faced with navigating back somewhere we have been before we might either retrace our steps or seek a shorter path. Both choices have costs. Here, we ask whether it is possible to characterize formally the choice of navigational plans as a bounded rational process that trades off the quality of the plan (e.g., its length) and the cognitive cost required to find and implement it. We analyze the navigation strategies of two groups of people that are firstly trained to follow a "default policy" taking a route in a virtual maze and then asked to navigate to various known goal destinations, either in the way they want ("Go To Goal") or by taking novel shortcuts ("Take Shortcut"). We address these wayfinding problems using InfoRL: an information-theoretic approach that formalizes the cognitive cost of devising a navigational plan, as the informational cost to deviate from a well-learned route (the "default policy"). In InfoRL, optimality refers to finding the best trade-off between route length and the amount of control information required to find it. We report five main findings. First, the navigational strategies automatically identified by InfoRL correspond closely to different routes (optimal or suboptimal) in the virtual reality map, which were annotated by hand in previous research. Second, people deliberate more in places where the value of investing cognitive resources (i.e., relevant goal information) is greater. Third, compared to the group of people who receive the "Go To Goal" instruction, those who receive the "Take Shortcut" instruction find shorter but less optimal solutions, reflecting the intrinsic difficulty of finding optimal shortcuts. Fourth, those who receive the "Go To Goal" instruction modulate flexibly their cognitive resources, depending on the benefits of finding the shortcut. Finally, we found a surprising amount of variability in the choice of navigational strategies and resource investment across participants. Taken together, these results illustrate the benefits of using InfoRL to address navigational planning problems from a bounded rational perspective.

Adaptive planning depth in human problem solving

“…Furthermore, we found a strong similarity between how different participants solve the same problem. This stereotypy was partially unexpected, given that previous studies of navigational planning showed a greater variability of strategies (Krichmar & He, 2021; Lancia et al, 2023). However, please note that our experiment required lower memory demands than the previous studies, since the participants could observe the whole problem graph.…”

“…Such partial observability requires balancing exploration and exploitation during planning (K. Friston et al, 2015; Sutton & Barto, 1998). Future studies might assess whether the results reported in this study generalize to dynamical and partially observable settings, or to settings that present participants with different incentives for planning versus following habits (Lancia et al, 2023).…”

“…Alternatively, before navigation begins, participants might start searching for a solution at low planning depth and then increase the depth progressively, until they find a (satisficing) solution. This procedure, which is akin to “iterative deepening depth-first search” in AI (De Groot, 1946; Korf, 1985) and to algorithms that start from an initial “cheap” plan and progressively refine it (Krichmar & He, 2021; Lancia et al, 2023; Todorov, 2009), would impose a filter on insufficient planning depth, which could result in the over-investing bias apparent in Figure 5, when considering that for most problem groups, the second-best planner has a greater depth than the minimum required depth. In other words, it could be faster to recognize that your strategy is too poor for the problem rather than recognizing that you are investing too much.…”

The relationship between object-based spatial ability and virtual navigation performance

“…It would also be interesting to consider twin study cohorts where the genetic contribution to performance can be explored, and also expand to incorporate a measure of general cognitive ability (Malanchini et al, 2021). Finally, going beyond scores of task performance to modelling the behaviour using reinforcement learning methods (de Cothi et al, 2022; Lancia et al, 2023) holds much promise for helping obtain a more mechanistic understanding of the processes involved in navigation and visuo-spatial cognition.…”

Strategic Task Decomposition in Joint Action

“…A carryover of old strategies to novel situations could be common to both individual and joint action setups. However, while in individual settings, the carryover could be linked to the cost of devising a novel strategy (Lancia et al., 2022; Todorov, 2009), in joint action settings, it could be additionally due to (or amplified by) the communication costs inherent to strategy change. Distinguishing between these possibilities is an intriguing question for future experimental study.…”