Abstract
Making accurate decisions in uncertain environments requires identifying the generative cause of sensory cues, but also the expected outcomes of possible actions. Although both cognitive processes can be formalized as Bayesian inference, they are commonly studied using different experimental frameworks, making their formal comparison difficult. Here, by framing a reversal learning task either as cue-based or outcome-based inference, we found that humans perceive the same volatile environment as more stable when inferring its hidden state by interaction with uncertain outcomes than by observation of equally uncertain cues. Multivariate patterns of magnetoencephalographic (MEG) activity reflected this behavioral difference in the neural interaction between inferred beliefs and incoming evidence, an effect originating from associative regions in the temporal lobe. Together, these findings indicate that the degree of control over the sampling of volatile environments shapes human learning and decision-making under uncertainty.
Introduction
In cognitive neuroscience, learning is typically investigated through two distinct lenses: perceptual learning, which involves observing cues to categorize the world, and reward-guided learning, which involves acting to obtain outcomes. Because these fields utilize different experimental paradigms, it has been challenging to determine if the brain employs consistent computational principles across both. Specifically, the role of agency—the ability to control information sampling—remains unclear, as previous studies often confounded agency with improved information gathering. This study aims to isolate the effect of agency by comparing two mathematically matched tasks where the only variable is whether the participant is an observer or an active agent.
Methods
The study employed a "reversal learning" task where participants tracked a hidden state that changed unpredictably. Using a yoked design, the researchers ensured that the stimulus sequences and environmental volatility were identical across two conditions:
- Cue-based (Cb): Participants observed sequences of oriented bars and inferred the underlying category.
- Outcome-based (Ob): Participants chose an action to draw from a target category, thereby interacting with the environment to achieve a goal.
Computational modeling based on Bayesian inference was used to estimate the perceived hazard rate ($h$) and inference noise. Neural activity was recorded using Magnetoencephalography (MEG), and multivariate pattern analysis (MVPA) was used to decode the representation of stimuli and internal beliefs. The study involved a main MEG group (n=24) and a behavioral control group (n=25).
Results
Behavioral Stabilization
The results indicated that agency induces a perception of environmental stability. In the active (Ob) condition, participants behaved as if the environment was more stable than in the observational (Cb) condition, despite identical objective volatility. The perceived hazard rate was significantly lower in the active condition ($h_{Cb} = 0.191$ vs. $h_{Ob} = 0.115$), and the Point of Subjective Equivalence (PSE) increased significantly ($p < 0.001$), indicating that more conflicting evidence was required to trigger a change in belief.
Neural Signatures
MEG decoding revealed a "neural hazard rate" that matched the behavioral findings, showing a drop in the brain's internal representation of volatility during active control. Additionally, the brain exhibited "relational" coding—representing how consistent a stimulus was with current beliefs—which peaked approximately 120ms earlier than the coding of absolute stimulus properties.
Anatomical Localization
While early visual and prefrontal regions processed basic stimulus features, the stabilization effect and relational coding were localized to associative regions in the temporal lobe, specifically the middle temporal gyrus and entorhinal cortex.
Discussion
This study demonstrates that interacting with volatile environments stabilizes hidden-state inference and its corresponding neural signatures. This effect may be driven by a form of hypothesis testing, where agency increases the commitment to internal beliefs. While this stabilization improves accuracy in stable environments, it can lead to slower adaptation in highly volatile ones. These findings have implications for understanding psychiatric disorders like Obsessive-Compulsive Disorder (OCD), where belief stability is often impaired.
Behavioral data is available via Figshare.
Summary updated