TL;DR
This paper explores how humans use decision heuristics (mental shortcuts) in complex tasks that require both choosing an action and skillfully executing it. They found that heuristics, often thought to be efficient due to resource limitations, are also beneficial in these "selection-execution" contexts, leading to faster learning and better performance, especially in spatial aspects of the task.
What Problem Does This Paper Address?
Traditional studies on decision heuristics often use simplified tasks where the "action" part is trivial (like pressing a button). However, in real-world scenarios, choosing an action is only the first step; executing that action skillfully is crucial for achieving the desired outcome. This paper addresses the gap in understanding whether heuristics remain efficient when the execution of an action is complex and requires fine motor control, and how this relates to individual differences in skill. The authors question if the "less-is-more" principle, which suggests simpler strategies can be more robust in uncertain environments, holds true when physical skill is a significant factor.
Methods
The researchers developed a novel "boat docking" task where participants had to select one of two cursors and then pilot it from a start to a goal. The key innovation was that each cursor had a different "motor execution cost" (one was harder to control due to an incongruent mapping), and rewards were contingent on both the correct cursor choice and skillful execution (spatial precision and temporal control).
To classify participants' strategies, they employed a computational modeling approach using the Drift-Diffusion Model (DDM). This model captures decision-making as a process of accumulating evidence towards a decision boundary. They specifically looked at how "difficulty" in the task (defined by either a simple spatial heuristic or a more complex "route-planning" strategy) modulated the "drift rate" (the speed of evidence accumulation). Participants whose drift rates were best explained by strategy-specific difficulty were classified as either "heuristic" or "route-planners."
Datasets
The study used data collected from 53 healthy human participants performing the novel "boat docking" task. No external datasets were used for training or evaluation. The "boat docking" task itself serves as a newly introduced benchmark for studying decision heuristics in contexts integrating action selection and execution.
Key Insights and Ideas
- Heuristics are efficient even with complex actions: Contrary to what might be expected, simple heuristics don't break down when action execution is demanding. They can lead to faster learning and better outcomes.
- "Less-is-more" extends to skill: The principle that simpler strategies can be more robust in complex environments seems to apply not just to decision-making but also to the integration of motor skill.
- Yoked dimensionality between planning and skill: The study suggests that the dimensions of information used for action selection are closely linked to the dimensions that are learned during skill execution. For example, a spatial heuristic for selection was associated with better spatial skill.
- Skill advantage is specific: Participants using heuristics showed a skill advantage primarily in spatial aspects, which were also the core features of their decision policy. This suggests heuristics don't confer a general skill boost but rather enhance skills relevant to the heuristic's focus.
Key Results
- Faster learning: Heuristic users achieved above-chance performance in making state-appropriate choices significantly earlier (by run 2) compared to route-planners (by run 4).
- Higher reward (with caveats): Heuristic users achieved higher overall reward, particularly when using the more difficult (incongruent) cursor.
- Spatial skill advantage: Heuristic users demonstrated better spatial skill (fewer direction changes) when piloting the incongruent cursor.
- Temporal skill learning difference: While both groups improved in reward and spatial skill over time, only the route-planners showed significant learning in temporal skill. Heuristic users showed no temporal skill learning.
- DDM classification: The DDM successfully distinguished between heuristic and route-planning strategies, with route-planners showing a larger decision boundary (requiring more evidence before deciding).
Limitations and Future Work
- Cross-sectional design: The study classified participants into groups based on their strategy but couldn't track within-subject learning of strategies over time. Future work could investigate if individuals can learn to switch between heuristic and complex planning strategies.
- Robustness over time: While the study identified distinct groups, it's unclear if these strategy preferences are stable traits or if they can evolve with extensive practice.
- Agency and learning: The authors suggest that increased "agency" (control over outcomes) might influence strategy adoption. Future research could explore how manipulating agency affects heuristic use.
- Embodied decision-making: The study proposes that their task is a simplified model of "embodied decisions" and suggests future work could explore more complex action selection and execution scenarios.
- Yoked dimensionality: Further investigation is needed to fully understand the causal relationship between the dimensions of information used in planning and those learned in skill execution.
Summary updated