Abstract
Decision making can be shaped both by trial-and-error experiences and by memory of unique contextual information. Moreover, these types of information can be acquired either by means of active experience or by observing others behave in similar situations. The interactions between reinforcement learning parameters that inform decision updating and memory formation of declarative information in experienced and observational learning settings are, however, unknown. In the current study, participants took part in a probabilistic decision-making task involving situations that either yielded similar outcomes to those of an observed player or opposed them. By fitting alternative reinforcement learning models to each subject, we discerned participants who learned similarly from experience and observation from those who assigned different weights to learning signals from these two sources. Participants who assigned different weights to their own experience versus those of others displayed enhanced memory performance as well as subjective memory strength for episodes involving significant reward prospects. Conversely, memory performance of participants who did not prioritize their own experience over others did not seem to be influenced by reinforcement learning parameters. These findings demonstrate that interactions between implicit and explicit learning systems depend on the means by which individuals weigh relevant information conveyed via experience and observation.
Introduction
Human decision-making is a hybrid process. We learn through Reinforcement Learning (RL)—the trial-and-error process of updating the "value" of choices based on rewards—and through Declarative Memory, which allows us to remember specific, unique contexts. Furthermore, we don't just learn from our own mistakes; we learn by observing others. This study addresses whether we use the same learning rate for our own experiences as we do for others, how decision "value" affects incidental memory, and if this relationship changes based on agency (making the choice vs. watching).
Methods
The researchers designed a probabilistic decision-making task involving 75 healthy volunteers (ages 19–47).
Task Structure
Participants faced "Experienced" (Exp.) trials, where they chose between two fractal cues, and "Observed" (Obs.) trials, where they watched a computerized partner choose.
As shown in the trial sequence, each choice was followed by a unique, neutral picture (from the IAPS database) and then an outcome (reward or no reward).
Experimental Conditions
The task utilized two types of cue pairs: Congruent (identical reward probabilities for both players) and Incongruent (reversed probabilities).
Computational Modeling and Memory Test
Following the task, participants completed a surprise recognition test for the neutral pictures, including confidence ratings. Behavior was analyzed using Bayesian Information Criterion (BIC) to fit RL models, identifying two learner types: "Single Strategy" (same learning rate for self and partner) and "Divided Strategy" (independent learning rates).
Results
Strategic Divergence and Performance
"Divided" learners significantly outperformed "Single" learners on incongruent trials ($F(1,58) = 67.11, p < 0.0001$). This suggests that separating self-experience from observation is more adaptive in complex environments where social information might conflict with personal experience.
The Value-Memory Link
For "Divided" learners, a strong positive correlation emerged between the estimated reward value of a trial and the likelihood of remembering the associated picture ($r = 0.85, p < 0.005$). This memory boost was specific to Experienced trials; high-value observed trials did not yield the same benefit. Additionally, high-value cues led to faster reaction times ($r = -0.91$ to $-0.98$), validating the internal value models.
Discussion
The findings demonstrate that the interaction between RL and memory is gated by personal agency. Only individuals who prioritized their own experience ("Divided" strategy) showed a memory boost for high-reward contexts. Limitations include the study's power to capture more complex "four-rate" models and the use of a computer simulation rather than a real human partner. Future work should explore if social authenticity or competitive vs. collaborative contexts alter how we weigh observational signals and their subsequent impact on declarative memory.
Summary updated