A New Account for “Practice Makes Perfect”: Revealing the computational and neural mechanisms of strategy transition

 

Substantial practice often leads to efficient and automatic behaviors, a phenomenon commonly referred to as automaticity. Understanding how this is achieved in the brain has been a central topic in psychology. Previous researchers have proposed that practice-induced automaticity involves a shift from implementing rules to using the memory. For example, when first learning the addition algorithm, solving 7+7+7+7 requires step-by-step addition, following the rule; with practice, one may recognize the pattern that four 7s equal 28, and performance begins to rely on memory of this pattern. While the concept of strategy shifting is well-established, when and how the shift occurs has remained unclear.

Leveraging advanced computational modeling and functional magnetic resonance imaging (fMRI) techniques, we proposed a new account that integrates decision-making into this process. We designed an engaging avatar gear selection task in which participants learned two task sequences and were later tested on recalling the task type at a cued position within the MRI scanner (Fig. A). The hypothesis was that participants initially relied on a rule-based strategy, making step-by-step inferences based on the learned sequences. However, as practice progressed, the associations between cues and responses strengthened, enabling a gradual dominance of memory-based strategies, which eventually replaced the rule-based approach (Fig B). Based on this hypothesis, they developed a computational model that chose the strategy through a cost-benefit tradeoff between the two alternative strategies (Fig. C). The model successfully predicted participants' behavioral patterns, particularly the timing of strategy transitions, which were marked by a noticeable temporary response slowing after the switch—consistent with the mental cost of strategy switch (Fig. D).

Figure. A) Experimental design; B) Model predictions illustrating the dynamic cost-benefit tradeoff across the two strategies; C) Architecture of the computational model; D) Behavioral results showing post-transition slowing; E) Enhanced rule representation in rule trials and stronger memory representation in memory trials; F) Dorsolateral prefrontal cortex signaling a boundary effect at the transition moment; G) Ventromedial prefrontal cortex encoding less similar patterns with practice.

Neuroimaging results highlighted how these transitions are reflected in brain activity. When participants employed the rule-based strategy, specific brain regions exhibited patterns unique to rule implementation. During memory-based strategy use, the same regions adapted their patterns to reflect memory operations (Fig. E). Notably, the dorsolateral prefrontal cortex signaled a boundary effect (i.e., a dramatic brain pattern change) at the moment of strategy shifts (Fig. F), while the ventromedial prefrontal cortex encoded increasingly distinct patterns across different memory associations (Fig. G).

This study sheds light on when and how humans dynamically change strategy through practice. These findings provide valuable insights for optimizing human skill training paradigms and advancing AI model training approaches.

Citation: 

Yang, G., & Jiang, J. (2024). Cost-benefit Tradeoff Mediates the Transition from Rule-based to Memory-based Processing during Practice. PLOS Biology, accepted. https://doi.org/10.1371/journal.pbio.3002987