In this groundbreaking study, the researchers quantitatively measured the effects of cognitive biases introduced by LLMs on decision-making. They discovered that LLM-generated summaries reframe the original sentiments of reviews in 26.5% of instances. Moreover, the study revealed a staggering figure: LLMs hallucinated — or provided inaccurate information — approximately 60% of the time when respondents questioned them about news stories, especially when these stories deviated from the training data used in training the models. The researchers characterized this tendency toward generating misleading information as a significant limitation, pointing out the difficulty these models face in reliably distilling fact from fiction.

So, how do these biases seep into the outputs of LLMs? The models often lean heavily on the initial segments of the text they summarize, neglecting to capture essential nuances that may emerge later in the review. This over-reliance on the early context, along with diminished performance when challenged with information beyond their training set, cultivates an environment ripe for biased summarization.

In order to deepen the understanding of the impact these biases have on consumer decisions, researchers implemented a study involving 70 participants who were presented with either the original review or summaries generated by LLMs for various products such as headsets, headlamps, and radios. Astonishingly, the results showed that 84% of participants who read the LLM-generated summaries expressed their intention to purchase the product, in stark contrast to only 52% of those who read the original human reviews. This striking difference underscores the profound influence that the framing of information can have on purchasing decisions.

The research team was surprised by the extent of the effect that the summarization had in their low-stakes experimental context. Specifically, Abeer Alessa, the lead author of the study and a master’s student in computer science, acknowledged the potential for an even more significant impact in high-stakes scenarios where major decisions are at play. This revelation raises questions about the ethical implications of using LLMs in contexts where consumer choices can have far-reaching effects.

In search of solutions to mitigate the issues identified, the researchers explored 18 distinct methods to address cognitive biases and hallucinations. They found that while some mitigation strategies proved effective for particular models in specific situations, there was no singular approach that worked universally across all LLMs. Furthermore, some mitigation techniques appeared to introduce new challenges, potentially compromising LLM performance in other critical areas.

Julian McAuley, a senior author of the paper and a professor of computer science at UC San Diego, emphasized the nuanced nature of bias and hallucination in LLMs. He explained that effectively fixing the issues tied to bias and hallucinations is complicated, requiring a contextualized approach rather than blanket solutions. These challenges highlight the intricate interplay between AI-generated content and human understanding.

The study assessed various models, including small open-source configurations like Phi-3-mini-4k-Instruct, Llama-3.2-3B-Instruct, and Qwen3-4B-Instruct. They also evaluated a medium-sized model, Llama-3-8B-Instruct, as well as larger models like Gemma-3-27B-IT and a proprietary model, GPT-3.5-turbo. This diverse array of models provided a fertile ground for examining the effects of LLMs on the generation of potentially biased and misleading content.

The researchers posit that their findings represent a crucial leap toward analyzing and addressing the content alterations induced by LLMs on human decision-making. By shedding light on these biases, the research aims to foster a deeper understanding of how LLMs can influence media, education, and public policy. The study emphasizes the need for ongoing discourse and research to navigate the complexities of AI-generated content and its ramifications on society.

In December 2025, the researchers presented their work at the esteemed International Joint Conference on Natural Language Processing and the Asia-Pacific Chapter of the Association for Computational Linguistics, signaling ongoing interest and inquiry in the field of artificial intelligence. This research holds promise not only for advancing the understanding of language models but also for guiding the ethical application of AI in various domains.

As the conversation around AI ethics continues to grow, the implications of such research become ever more pressing. The findings from UC San Diego emphasize that while LLMs promise efficiency and versatility in content creation, they also carry the potential for unwanted biases that could skew user perceptions and decisions. To harness the power of these technologies responsibly, it is imperative for developers and users alike to be mindful of the subtleties that influence how information is perceived and acted upon.

Given the increasing prevalence of AI in everyday decision-making contexts, the research serves as a vital reminder of the need for caution. As LLMs are integrated into more facets of daily life, from shopping to information dissemination, ensuring the integrity of the content they generate must remain a priority. A commitment to transparency, accountability, and ethical guidelines in deploying LLMs can help mitigate unintended biases and safeguard against their potential consequences.

Subject of Research: People
Article Title: Chatbots’ Bias Makes Consumers More Likely to Buy Products Suggests New Study
News Publication Date: October 2023
Web References:
References:
Image Credits: David Baillot/University of California San Diego

Keywords: Cognitive Bias, Large Language Models, Consumer Behavior, Artificial Intelligence, Product Reviews, Decision Making, Mitigation Strategies, Research Study.

Decision-making has long been scrutinized through the lens of rational choice theory, where individuals are presumed to weigh options impartially before committing to a course of action. However, the reality is far more complex, colored by an array of cognitive biases and heuristics. The new research spotlights one such influence: the propensity to repeat prior actions, which operates as an inherent bias shaping future choices. By systematically analyzing this phenomenon, the authors elucidate how repeated behaviors sculpt decision trajectories, subtly contouring preferences and outcomes.

At the core of this inquiry lies the concept of context-dependent decision-making. Unlike isolated scenarios where choices stand independent, many real-world decisions are nested in dynamic environments that shape and constrain possible actions. Wagner and colleagues articulate how the history of actions within such contexts doesn’t merely inform choice outcomes; it actively predisposes individuals toward specific behavioral patterns. This finding suggests a form of cognitive inertia, where the residue of previous decisions biases subsequent actions, potentially circumventing deliberate rational evaluation.

The methodology employed to uncover these insights is as rigorous as it is innovative. The researchers combined behavioral experiments with sophisticated computational models that map the probabilistic influences of action history on choice. Participants engaged in tasks designed to mimic fluctuating decision environments, while analytic tools discerned how prior actions skewed decision probabilities. This integration of behavioral data and mathematical modeling provided a nuanced portrait of bias mechanisms underlying choice repetition.

One of the technical linchpins of the study is the deployment of hierarchical Bayesian models that capture the layered dependencies in decision processes. These models enabled the team to quantify the strength of action repetition bias and to distinguish it from other cognitive drivers such as reward sensitivity or risk aversion. By isolating the unique contribution of prior action history, the analysis reveals that even in the absence of explicit rewards, repetition tendencies persist, hinting at deeper neural substrates at play.

Exploring the neural correlates, Wagner et al. discuss the likely involvement of brain regions associated with habit formation and procedural memory, such as the basal ganglia and supplementary motor area. This connection is consistent with the hypothesis that repeated actions become encoded as cognitive habits, thereby influencing choice beyond conscious deliberation. These neural mechanisms provide a biological framework that explains why repetition bias is so robust and often resistant to conscious override.

The implications of this research extend dramatically into realms where decision-making governs critical outcomes. In economic markets, for example, understanding how action repetition biases investor behavior could refine predictive models and improve interventions aimed at minimizing irrational financial decisions. Similarly, in policy design, acknowledging these biases may guide strategies that foster desirable behaviors, such as promoting sustainable consumption through the reinforcement of positive action cycles.

Moreover, the findings challenge current paradigms in artificial intelligence and machine learning, where decision algorithms often assume independence across sequential choices. Incorporating mechanisms that simulate action repetition biases could enhance the realism and efficacy of AI agents, particularly those designed to interact with humans or operate in complex, context-rich environments. This synergy between cognitive science and technology may pave the way for advanced systems with more human-like decision dynamics.

Further explorations inspired by this study might delve into the variability of repetition bias across individuals and cultures. The authors hint at potential moderators, such as personality traits or social context, which could modulate the degree to which past actions influence choice. Unpacking these nuances would deepen our comprehension of decision-making diversity and inform personalized approaches in behavioral interventions or user-experience design.

An intriguing aspect of the study lies in how repetition bias interacts with uncertainty in decision contexts. When environmental feedback is ambiguous or volatile, reliance on prior actions appears to increase, acting as a cognitive anchor when information is scarce. This adaptive dimension suggests that action repetition serves not only as a bias but also as a heuristic that stabilizes decision-making under uncertainty, balancing exploration and exploitation strategies.

Critically, the research raises questions about the boundaries between adaptive and maladaptive repetition biases. While repeated actions can streamline decision processes and reduce cognitive load, excessive reliance may entrench suboptimal behaviors, leading to persistence in ineffective or harmful choices. This dual nature underscores the importance of context in evaluating whether repetition bias facilitates or undermines optimal decision-making.

Technically, the study also advances methodological frontiers by demonstrating how fine-grained behavioral tracking and computational analytics can converge to decode complex cognitive patterns. The approach championed by Wagner and colleagues sets a precedent for future research aiming to unravel the temporal dynamics of choice and the latent variables influencing human cognition over time. Such frameworks may become essential tools in cognitive science and behavioral neuroscience research.

Finally, these insights carry tangible relevance for education and mental health. Understanding how action repetition biases learning strategies, habit formation, and decision-making pathways can inform the development of pedagogical techniques and therapeutic interventions. Targeting these biases may optimize habit change programs or support recovery processes that hinge on altering maladaptive decision patterns.

In sum, the study by Wagner, Wolf, and Kiebel represents a landmark advancement in decoding the interplay between past actions and present choices. By elucidating the cognitive and neural mechanisms of action repetition bias in context-dependent decision-making, this research not only challenges prevailing theoretical models but also opens new avenues for practical applications across diverse sectors. As science continues to probe the intricate workings of the human mind, such integrative approaches will be crucial in bridging the gap between abstract theory and real-world behavior.

Subject of Research:
The cognitive and neural mechanisms underlying action repetition bias in context-dependent decision-making.

Article Title:
Action repetition biases choice in context-dependent decision-making.

Article References:
Wagner, B.J., Wolf, H.B. & Kiebel, S.J. Action repetition biases choice in context-dependent decision-making. Communications Psychology (2025). https://doi.org/10.1038/s44271-025-00363-x

Image Credits:
AI Generated

The crux of this research lies in parsing the mechanisms underpinning confidence revisions when confronted with evidence that contradicts one’s initial decision—termed disconfirmatory evidence. Typically, when people make decisions, they build an internal metric of confidence based largely on confirmatory information—supporting evidence that aligns with the chosen option. However, the study found that encountering disconfirmatory evidence not only tempers this confidence but does so in a nuanced manner that had previously been underappreciated in cognitive science.

Confidence is not a static judgment; it fluctuates as individuals accumulate information. The researchers employed rigorous experimental paradigms where participants performed perceptual decision tasks with varying degrees of uncertainty. By systematically introducing evidence that either corroborated or undermined the participants’ choices, the team was able to map how confidence levels adapted in real-time. Results showed a marked asymmetry: disconfirmatory evidence exerted a disproportionately stronger influence, often leading to more substantial downward adjustments in confidence than the upward influence invoked by confirmatory cues.

This asymmetry has deep implications for understanding cognitive biases and errors. Traditional models have often assumed that confidence updates depend symmetrically on both types of evidence. The evidence presented by Boldt and colleagues challenges this assumption, indicating that the brain might be wired to prioritize skepticism—a cognitive safeguard against overconfidence, which can be perilous in decision-critical environments such as medical diagnostics, financial forecasting, or legal judgments.

Delving into the neural substrates, the authors propose that specific brain regions implicated in error monitoring and conflict detection, including the anterior cingulate cortex and the prefrontal cortex, play pivotal roles in recalibrating confidence in the face of disconfirmatory information. Neuroimaging data suggest heightened neural activity when participants process contradicting evidence, manifesting as an adaptive cognitive mechanism that prevents the entrenchment of erroneous beliefs.

Moreover, this recalibration process is modulated by metacognitive sensitivity—the capacity to introspect on one’s own decision accuracy. The research revealed that individuals with higher metacognitive awareness show more pronounced, flexible confidence adjustments, reflecting enhanced ability to integrate disconfirmatory evidence effectively. This finding opens avenues for potential interventions to boost metacognitive skills, thus refining decision-making quality across various domains.

One particularly illuminating aspect of the study involves computational modeling, where the researchers developed Bayesian models to simulate confidence updating dynamics. Bayesian frameworks assume that individuals integrate prior beliefs with new evidence optimally. However, the data indicated deviations from perfect Bayesian updating, especially under ambiguous conditions. The model modifications incorporated a bias parameter accounting for the overweighting of disconfirmatory evidence’s impact, which significantly improved the fidelity of predicted behaviors.

The ramifications of these discoveries extend beyond laboratory conditions. In real-world scenarios, decision-makers frequently encounter conflicting information. The tendency for disconfirmatory evidence to disproportionately decrease confidence may serve an evolutionary function, increasing caution and prompting re-evaluation. Nonetheless, it also raises the risk of excessive doubt or indecision in scenarios demanding swift action, highlighting a delicate balance the brain must navigate.

Importantly, this research also addresses paradoxes observed in social cognition, where individuals often exhibit motivated reasoning—discounting disconfirmatory facts to preserve prior beliefs. The results of Boldt et al. suggest that while basic cognitive mechanisms favor strong responses to contradictory evidence, social and emotional contexts may modulate this process, sometimes undermining the adaptive benefits of such recalibration.

The experimental design carefully controlled for confounding variables such as task difficulty, prior biases, and response times, ensuring robust findings. By capturing nuanced confidence fluctuations over milliseconds and correlating them with behavioral choices, the study sets a methodological benchmark for future research in cognitive confidence and decision-making.

Furthermore, the authors ponder the practical applications of these insights. In domains like education, training individuals to better process and accept disconfirmatory information could enhance learning outcomes and critical thinking skills. In clinical psychology, understanding confidence calibration may help address disorders characterized by impaired decision-making, such as obsessive-compulsive disorder or anxiety, where disproportionate doubt or certainty plays a role.

The research also opens exciting prospects for artificial intelligence and human-computer interaction. Designing algorithms or systems that mimic human-like confidence updating, appropriately weighting disconfirmatory evidence, could enhance AI decision transparency and reliability. Such systems could better assist humans in high-stakes environments by modeling adaptive confidence adjustments.

One of the study’s elegant contributions is its challenge to simplistic views of confidence as merely “feeling sure.” Instead, it frames confidence as a dynamic, evidence-dependent process deeply intertwined with both cognitive and neural mechanisms. It underscores that encountering contradictory information is not merely noise but a critical signal that reshapes our certainty landscapes, often in subtle and sophisticated ways.

In sum, the work by Boldt, Sun, and Desender marks a milestone in the science of confidence and decision-making. The intricate dance between belief, doubt, and evidence is at the heart of rationality, and understanding how disconfirmatory evidence disproportionately informs confidence provides a fresh theoretical and practical framework. As cognitive science advances, such nuanced models of confidence may well transform our approaches to education, mental health, technology, and beyond, enhancing human judgment in an era swamped by information and uncertainty.

As the fields of psychology, neuroscience, and artificial intelligence increasingly intersect, these results highlight the profound importance of confidence calibration as both a cognitive skill and a neural phenomenon. Future research building on this foundation promises to unravel further the mysteries of human reason and pave the way for smarter, more adaptive minds and machines alike.

Subject of Research: The neural and cognitive mechanisms of how disconfirmatory evidence affects confidence during decision-making.

Article Title: How disconfirmatory evidence shapes confidence in decision-making.

Article References:
Boldt, A., Sun, Y. & Desender, K. How disconfirmatory evidence shapes confidence in decision-making.
Commun Psychol 3, 150 (2025). https://doi.org/10.1038/s44271-025-00325-3

Image Credits: AI Generated

Traditional behavioral economics has made impressive strides by incorporating psychological realism into economic models, accounting for cognitive biases, heuristics, and other deviations from classical rationality. However, critiques of behavioral economics often point to its limited normative foundation, primarily focused on correcting biases to improve decision-making under the assumption that individuals seek to maximize subjective utility. The capability approach, originally developed by economist and philosopher Amartya Sen, broadens the evaluative space by prioritizing what people are actually able to do and be, rather than reducing well-being to utility or resources alone. Garces-Velastegui’s synthesis asserts that behavioral interventions should be assessed not only by their effects on choice architecture but also by their capacity to expand or restrict people’s capabilities.

One of the core technical contributions of this research lies in the formal modeling of capabilities within a behavioral framework. Unlike utility, which is often modeled as a scalar and ordinal function over commodity bundles or chosen actions, capabilities are represented as multidimensional vectors capturing various functionings—such as health, education, social participation, and autonomy—that individuals deem valuable. The challenge is translated into designing behavioral economic models that accommodate these multidimensional vectors while preserving testability and empirical applicability. Garces-Velastegui approaches this by proposing a capability-informed utility function, which internally weighs choices not just by immediate preferences but also by their impact on expanding constituencies of freedoms.

The implications of this paradigm shift are substantial, especially in policy design. Behavioral policies informed by capabilitarian economics would steer away from paternalistic nudges that reinforce certain behaviors based solely on efficiency criteria. Instead, they would advocate for interventions that enhance agency and opportunity sets, making welfare enhancement intrinsically tied to expanding capabilities. This may require reconsidering the often criticized “libertarian paternalism” framework and embracing a more pluralistic ethical foundation that accounts for diverse individual aspirations, power asymmetries, and contextual factors influencing capability development.

Technically, integrating the capability approach into behavioral economics necessitates overcoming methodological obstacles. Traditional revealed preference approaches struggle with nondichotomous outcomes and incomplete preferences over complex capability sets. To address this, Garces-Velastegui advances an axiomatic foundation that redefines rationality norms within capability spaces, allowing for incompleteness and context sensitivity. This results in models that better reflect real-world decision-making where individuals’ preferences may depend on socially embedded values and dynamic aspirations, challenging the static assumptions of classical economic agents.

Furthermore, the paper explores the role of cognitive biases and bounded rationality not as mere obstacles to be corrected but as phenomena that must be understood through the lens of capability restrictions. For example, poverty-induced scarcity may impair cognitive bandwidth, limiting an individual’s capability to make long-term beneficial decisions, thus creating a feedback loop between capability deprivation and suboptimal behavior patterns. This reframing transforms behavioral shortcomings into indicators of structural capability constraints, highlighting the ethical imperative for policy interventions that address underlying capability deficits, rather than only symptomatic behavioral anomalies.

The multi-dimensionality inherent to the capability approach also invites richer empirical strategies. Data collection and econometric techniques must be adapted to capture heterogeneous and interdependent dimensions of functionings. The paper suggests utilizing mixed-method approaches that combine quantitative surveys with ethnographic insights to map local entitlements and aspiration structures. This enriched data environment provides a fertile ground for testing behavioral hypotheses within realistic capability contexts, enabling policymakers to tailor interventions commensurate with lived realities rather than abstracted utility functions.

Incorporating the capability approach also has ramifications for measuring economic development and welfare beyond traditional GDP-centric metrics. Behavioral economics infused with the capability framework advocates for composite indices reflecting individuals’ opportunities to lead lives they value, highlighting disparities obscured by aggregate income statistics. Such indices could incorporate psychological well-being, social inclusion, and empowerment measures, providing a more nuanced evaluation of policy efficacy and societal progress over time.

Moreover, the article critiques current applications of behavioral economics that often inadvertently reinforce inequalities by failing to address differential capability endowments. Nudges designed for middle-class populations may not translate effectively for marginalized groups whose capabilities are constrained by systemic barriers. Capabilitarian behavioral economics thus calls for a more equity-sensitive approach, integrating social justice perspectives into behavioral policy design. This enhances both fairness and efficiency by recognizing that expanding capabilities is essential for enabling meaningful choice.

Garces-Velastegui also conceptualizes a framework for interdisciplinary collaboration, urging economists to dialogue with philosophers, sociologists, and psychologists to enrich models of human behavior with ethical and contextual complexity. This confluence is necessary to navigate the intricate terrains of freedom, agency, and social norms that define capabilities but evade purely economic quantification. The paper suggests that such dialogue can mitigate reductionism and foster conceptual sophistication critical for addressing contemporary challenges like climate change, health disparities, and technological disruptions.

Importantly, the paper discusses the potential for capabilitarian behavioral economics to revolutionize education and health policy design. By emphasizing capabilities, interventions can transcend rote behavioral change and foster environments where individuals can develop skills, resilience, and autonomy. In education, this implies curricula and pedagogies designed not solely for knowledge acquisition but for cultivating diverse functionings aligned with personal and societal values. Similarly, health interventions must account for patients’ capability sets, empowering them to manage their conditions within broader life contexts.

The research also forecasts significant implications for welfare economics and ethical theory. By centering capabilities, welfare analyses can better account for justice, empowerment, and participatory dimensions frequently neglected by utility-based approaches. The paper encourages a reevaluation of welfare criteria, suggesting that societal assessments should focus on collective capability enhancements rather than aggregate utility sums, aligning with a broader human development agenda.

Finally, Garces-Velastegui’s work ignites a call to action for policymakers, scholars, and practitioners. Incorporating capabilities into behavioral economics not only improves descriptive accuracy but also promotes a transformative normative agenda oriented toward human flourishing. The melding of these two intellectual traditions offers a powerful toolkit to tackle entrenched societal challenges, guiding interventions that respect dignity, promote agency, and enable truly meaningful economic decisions.

This pioneering study represents a significant step in redefining behavioral economics through the lens of human capabilities. By questioning established assumptions and proposing a rigorous theoretical and empirical foundation for capabilitarian behavioral economics, it opens new pathways for research and policy reforms. As behavioral science continues to evolve, integrating ethical insights about freedom and opportunity promises richer, more equitable understandings of human behavior in complex social environments.

Subject of Research: Integration of behavioral economics and the capability approach to develop a framework called "capabilitarian behavioral economics," focusing on how behavioral economics can learn from and incorporate the multidimensional concept of capabilities.

Article Title: A capabilitarian behavioral economics: what behavioral economics can learn from the capability approach.

Article References:
Garces-Velastegui, P. A capabilitarian behavioral economics: what behavioral economics can learn from the capability approach. International Review of Economics 71, 667–690 (2024). https://doi.org/10.1007/s12232-024-00457-8

Image Credits: AI Generated

DOI: https://doi.org/10.1007/s12232-024-00457-8