Disinformation elicits learning biases.
Clinical Snapshot
PICO Framework
| P — Population | Healthy adult participants (young adults, male and female) recruited for two laboratory experiments |
| I — Intervention | Exposure to disinformation — sources of varying credibility who occasionally lied about true choice outcomes in a two-armed bandit reinforcement learning task |
| C — Comparator | Exposure to truthful/credible sources and ideal Bayesian learning benchmarks (computational reference standard) |
| O — Outcomes | Computational estimates of learning rate, positivity bias, and deviations from ideal Bayesian updating (learning biases) as quantified by reinforcement learning model parameters |
Bottom Line
This experimental study from UCL's Max Planck Centre uses a computational reinforcement learning framework to investigate how disinformation affects individual learning processes. Using a two-armed bandit task with sources of varying credibility, the authors demonstrate that while learning broadly tracks source credibility — consistent with Bayesian principles — disinformation exposure elicits specific, measurable learning biases. These include inappropriate learning from fully unreliable sources, exaggerated updating from trustworthy information ('jumping to conclusions'), and amplification of positivity bias. The controlled experimental design affords reasonable internal validity for causal inference. However, the abstract lacks quantitative effect sizes and precision estimates, sample characteristics are unreported, and the ecological validity of a gambling task as a proxy for real-world disinformation exposure is uncertain. Generalisability beyond young, healthy, educated adults is unestablished. For clinicians, the most relevant implication is theoretical: disinformation may exploit pre-existing cognitive learning biases rather than simply introducing false beliefs, with potential relevance to psychiatric populations with aberrant learning (e.g., psychosis, OCD). This is a scientifically interesting proof-of-concept study warranting replication with diverse samples and ecologically valid paradigms before clinical or public health translation.
Key Findings
P Value: Not reported in abstract
Effect Size: Not reported in abstract; qualitatively described as 'striking' biases and 'exaggerated' learning from trustworthy sources
Primary Outcome: Learning rate as a function of source credibility, and deviations from ideal Bayesian updating (learning biases) in the presence of disinformation
Nnt Or Sensitivity: Not applicable to this experimental paradigm; key metrics are computational model parameters (learning rates, positivity bias indices) — specific values not reported in abstract
Confidence Interval: Not reported in abstract
Clinical Application
The reinforcement learning paradigm and computational modelling approach are research tools not currently deployable in routine clinical settings. Findings are at proof-of-concept stage. Translation to clinical screening or intervention would require substantial further development, ecological validation, and clinical trials. Australia's media landscape and disinformation environment (including social media regulation debates, COVID-19 infodemic, and electoral misinformation) make this research topical. The RACGP has no current guidelines addressing disinformation literacy as a clinical concern, though health literacy frameworks are relevant. The TGA and Australian Department of Health have addressed health misinformation specifically. Findings may inform public health communication strategies and mental health literacy programs, but no PBS-listed interventions or TGA-approved tools exist in this space. Australian researchers at institutions such as the ARC Centre of Excellence for Automated Decision-Making and Society are active in adjacent fields. Healthy young adults in controlled laboratory settings; potential relevance to populations vulnerable to disinformation including those with impaired reality monitoring, reduced cognitive flexibility, or compromised trust calibration (e.g., psychosis spectrum, depression, dementia) — though this is speculative and not tested in this study
Abstract
In open societies, disinformation is often considered a threat to the very fabric of democracy. However, we know little about how disinformation exerts its impact, especially its influence on individual learning processes. Guided by the notion that disinformation exerts its pernicious effects by capitalizing on learning biases, we ask which aspects of learning from potential disinformation align with ideal 'Bayesian' principles, and which exhibit biases deviating from these standards. To this end, we harnessed a reinforcement learning framework, offering computationally tractable models capable of estimating latent aspects of a learning process as well as identifying biases in learning. In two experiments, participants completed a two-armed bandit task, where they repeatedly chose between two lotteries and received outcome-feedback from sources of varying credibility, who occasionally disseminated disinformation by lying about true choice outcome (e.g., reporting non-reward when a reward was truly earned or vice versa). Computational modelling indicated that learning increased in tandem with source credibility, consistent with ideal-Bayesian principles. However, we also observed striking biases reflecting divergence from idealized Bayesian learning patterns. Notably, in one experiment individuals learned from sources that should have been ignored, as these were known to be fully unreliable. Additionally, the presence of disinformation elicited exaggerated learning from trustworthy information (akin to jumping to conclusions) and exacerbated a normalized measure of 'positivity bias' whereby individuals self-servingly boost their learning from positive, relative to negative, choice feedback. Thus, in the face of disinformation we identify specific cognitive mechanisms underlying learning biases, with potential implications for societal strategies aimed at mitigating its harmful impacts.
References
- 1.Vidal-Perez, J., Dolan, R. J., & Moran, R. (2026). Disinformation elicits learning biases. eLife. https://doi.org/10.7554/eLife.106073
Related Research
Cognitive, affective & behavioral neuroscience
Emotion detection unveiled: A cognitive-computational synthesis of physiological models, machine learning, and datasets.
3 Aug 2026
Cognitive, affective & behavioral neuroscience
A review of current capabilities and future directions in machine-based emotion recognition
2 Aug 2026
JMIR mHealth and uHealth
Machine Learning Frameworks for Wearable-Based Stress Modeling in Naturalistic Settings: Scoping Review
1 Aug 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service