Decision uncertainty during hypothesis testing enhances memory accuracy for incidental information

Abstract

Humans actively seek information to reduce uncertainty, providing insight on how our decisions causally affect the world. While we know that episodic memories can help support future goal-oriented behaviors, little is known about how hypothesis testing during exploration influences episodic memory. To investigate this question, we designed a hypothesis testing paradigm, in which participants figured out rules to unlock treasure chests. Using this paradigm, we characterized how hypothesis testing during exploration influenced memory for the contents of the treasure chests. We found that there was an inverted U-shaped relationship between decision uncertainty and memory, such that memory was best when decision uncertainty was moderate. An exploratory analysis also showed that surprising outcomes lead to lower memory confidence independent of accuracy. These findings support a model in which moderate decision uncertainty during hypothesis testing enhances incidental information encoding.

Figure 1.

Overview of the task. (A) Hypothesis testing task. Participants were instructed to choose a key from three keys given, with the goal to open treasure chests. Choosing a key with target feature would open the treasure chest and a trial-unique object would be shown inside the treasure chest. Otherwise, the treasure chest stayed closed. (B) Three different dimensions for a key and three different features for a dimension. (C) Control task. Participants were instructed to choose the golden key from three keys given. Outcomes of the control task were yoked to outcomes of the hypothesis testing task. (D) Surprise memory task. Participants were presented with all the images from the hypothesis testing task and control task and the same number of new images. Participants need to indicate whether they saw the image during encoding.

Figure 2.

Hypothesis testing enhances memory encoding. (A) After 24-h of delay, memory was better for objects presented during the hypothesis testing task than the control task. Error bars are 95% confidence interval. (B) Model estimate from mixed effect model, where task condition and counterbalance order and the interaction were submitted as fixed effects and objects and subjects were submitted as random effects. Only the main effect of conditions was significant.

Figure 3.

Predictive accuracy for the feature updating reinforcement learning model across the entire task. Because participants completed different number of trials for the task, we only show learning across the four trials before a feature change. We found that predictive accuracy increased across four trials before a feature change, but no learning across rules. Red vertical lines indicate the fourth trial before a feature change.

Figure 4.

Feature RL outperformed item RL. (A) Item RL only updated value for the chosen key, whereas feature RL updated values for all keys that shared features with the chosen key. (B) Cross-validated model probability suggested that both feature RL and item RL predicted the data significantly better than chance (gray dash line), and feature RL outperformed item updating RL across all three blocks. (C) There is increasing predictive accuracy for the last four trials before a rule switch and the switching trial for both feature RL and item RL. Participants did not know the switching trial until they saw the feedback. At the trial level, both feature RL and item RL still predicted the data significantly better than chance (gray dashed line), and feature RL outperformed item RL in predicting at the trial level.

Figure 5.

Changes of decision uncertainty and surprise across the four trials before a feature change, including the trial at switch. (Blue line) Decision uncertainty decreased across the four trials before a feature change, including the trial at switch. (Orange line) Surprise decreased across the four trials before a feature change, but increased at trial of switch (see the Supplemental Material for detailed results on surprise).

Figure 6.

Relationship between memory accuracy and decision uncertainty. There was a significant linear relationship between memory accuracy and decision uncertainty (red). Memory accuracy decreased as decision uncertainty increased. On top of that, there was also a significant inverted U-shaped relationship between memory accuracy and decision uncertainty (blue). Memory was enhanced at moderate decision uncertainty. Black dots indicate individual subject data.

Figure 7.

Correlation of decision uncertainty and surprise. An individual example of changes of decision uncertainty and surprise in one block. Gray dashed line indicates switching trials.