Abstract
A newly published study from University College Dublin finds that people hold a strong and consistent belief that artificial intelligence makes decisions based on logic and reason, while viewing human decision-making as driven by emotion and feeling. Crucially, this belief is not a trivial observation; it changes how people actually behave. When participants in an economic experiment believed they were interacting with an AI, they acted more rationally, accepting unfair but financially beneficial offers that they would typically reject when they thought the other party was human. These findings have direct implications for how AI is deployed in business, public services, and policymaking in Ireland and beyond.
Introduction
Artificial intelligence (AI) is now embedded in everyday life. Algorithms recommend products, sort job applications, flag social welfare claims, and increasingly inform decisions in healthcare, law, and government. Scaling AI implementation is a pressing priority for most business organisations (Deloitte AI Institute, 2024), and among governments that have already adopted AI systems, 60% intend for such systems to influence real-time decision-making (Deloitte AI Institute, 2021). Yet despite this rapid expansion, we know surprisingly little about what ordinary people (as distinct from engineers or policymakers) actually believe about how AI thinks and makes decisions.
A new study published in Frontiers in Computer Science by researchers at University College Dublin addresses this gap directly (Vijayakumar, Yang and DeFranza, 2026). The research asks two related questions: Do people think AI makes decisions differently to humans? And if so, does that belief affect how people themselves behave when dealing with an AI? The research confirms that both of these phenomenon are true, in ways that matter for public policy.
Key Findings: What the Research Shows
The research team conducted three pre-registered studies involving over 800 participants. The studies draw on established frameworks for understanding how people rely on feelings versus reason when making decisions (Hong and Chang, 2015), and deploy an incentive-compatible (i.e., a study design where participants face real consequences for their decisions) economic game to observe real behaviour rather than self-reported intentions. The key findings are outlined below.
The first two studies establish that people consistently rate AI as a reason-driven decision-maker and humans as feeling-driven. In Study 1A, participants rated AI decision-making at 8.35 out of 9 on a scale from “feeling” to “reason”, compared to just 4.33 for a typical human. This gap (and the statistical effect size) was very large (Vijayakumar, Yang and DeFranza, 2026). Study 1B confirmed that this finding is not simply a product of the measurement instrument. Using a different approach that allowed participants to rate AI and humans on both dimensions simultaneously, AI was still rated as significantly more reason-based and significantly less feeling-based than humans.
The third study tested whether this belief translates into real behavioural differences. Using an ultimatum game with actual financial stakes, participants were told they were playing with either an AI or a human. Those in the AI condition accepted economically rational but socially unfair offers at a noticeably higher rate (49.2%) compared to those who thought they were dealing with a human (37.5%; Vijayakumar, Yang and DeFranza, 2026). Importantly, statistical mediation analysis confirmed that this behavioural shift was driven specifically by participants’ belief that the AI was more reason-driven, rather than simply by the fact that the counterpart was non-human in some general sense.
Why Does This Matter? The Ultimatum Game Explained
The ultimatum game is a classic tool in behavioural economics (Güth, Schmittberger and Schwarze, 1982). One player, the proposer, divides a sum of money (in this case, one US dollar) and offers a split to a second player, the responder. If the responder accepts, both players get their respective shares. If the responder rejects, neither player receives anything.
Economic theory predicts that a purely rational actor in the role of responder should always accept any offer above zero, since gaining something is always better than gaining nothing. In practice, however, humans routinely reject offers they consider unfair, typically anything below a 30–40% share (Larney, Rotella and Barclay, 2019). This well-documented tendency reflects the role of emotions such as anger, fairness concerns, and social norms in everyday decision-making.
The UCD study used a highly unequal split, with the responder offered US$0.10 against the proposer’s US$0.90, a ratio established in prior research as an “unfair” split (Sanfey et al., 2003). When participants believed the proposer was an AI, they were significantly more likely to accept the small but positive payoff. This shift in behaviour, moving closer to the economically rational optimum, which economists call the Nash equilibrium, was driven by participants’ belief that the AI was operating on pure logic rather than social norms or interpersonal dynamics.
The Importance of Lay Belief
The concept of a “lay belief” is central to this research. Lay beliefs are the informal, sometimes intuitive theories we hold about the world, formed through experience, culture, and media rather than formal education, and they can significantly affect perceptions and decision-making (Plaks, Levy and Dweck, 2009). We all hold lay beliefs about many things, including health, luck, and character, and these beliefs shape how we interpret and respond to our environment, sometimes more powerfully than objective information.
The researchers draw a link between the public’s lay beliefs about AI and decades of pop culture portrayal. Robots and intelligent systems in film, television, and fiction are almost universally depicted as devoid of emotion and guided by pure logic (Hermann, 2021). Examples such as HAL 9000, Data from Star Trek, and the machines in The Matrix may have shaped a collective intuition about what AI is: precise, cold, and unerringly rational. Even if that image does not match the reality of modern AI systems (which are often trained on messy human-generated data and can reflect underlying human biases), the belief persists and influences behaviour.
Importantly, people have also been shown to believe that AI is incapable of experiencing emotions or performing tasks that rely on emotional understanding (Castelo, Bos and Lehmann, 2019). This compounds the view of AI as a purely rational actor. Notably, this lay belief aligns closely with the economic concept of Homo Economicus, the idealised rational actor who maximises utility given perfect information (Levitt and List, 2008). The study’s authors suggest that when people interact with AI, they may mentally switch to a more analytical, utility-focused mode to match what they assume their AI counterpart is doing.
Policy Implications for Ireland and Beyond
The implications of these findings extend well beyond the laboratory. AI-based systems are being adopted at scale across public and private sectors in Ireland and across the European Union. The study identifies several specific areas where policymakers should take note.
AI in Public Services and Government
A majority of Europeans surveyed (and 60% of those aged 25–34) indicate support for replacing national parliamentarians with algorithmic decision-makers (Jonsson and de Luca Tena, 2021). Whether or not one views this trend with enthusiasm or alarm, the UCD findings suggest that public acceptance of AI in consequential settings may be partly driven not by evidence of AI’s actual performance but by lay beliefs about its rationality that may not be well-founded.
The researchers raise a particular concern: recent evidence suggests that lower AI literacy (that is, less objective knowledge about what AI actually does) correlates with greater willingness to rely on it (Tully, Longoni and Appel, 2025). In other words, those who know the least about AI may trust it the most, in part because they perceive it as a kind of infallible reasoning machine. This is a concerning dynamic for democratic accountability and should prompt public bodies to invest in AI literacy programmes alongside the deployment of AI tools. UNESCO has already raised concerns about the potential for generative AI to embed and amplify biases in public-facing contexts (UNESCO, 2024), and the findings of this study add to that concern.
Business Negotiations and Consumer Contexts
The results have immediate relevance for how organisations deploy AI-facing interfaces. Prior research shows that people are more likely to adopt advice from AI agents when they hold strong beliefs about AI’s superior intelligence (von Walter, Kremmel and Jäger, 2021). The UCD study extends this insight: customers or counterparties who believe they are negotiating with an AI may also adopt a more analytical, cost-benefit-focused approach, sidelining emotional considerations that might ordinarily moderate behaviour. Firms might benefit from deploying AI agents in purely transactional settings where rational engagement is desirable, but could face problems in contexts where emotional intelligence and relational trust are important.
The authors also raise a caution: emerging evidence suggests that interacting with AI may increase the likelihood of immoral behaviour (Gill, 2020) and reduce prosocial orientation (Granulo et al., 2024). If AI exposure nudges people towards a narrow economic rationality, it may simultaneously erode the social and ethical sensitivities that underpin cooperative behaviour.
Healthcare, Legal, and Regulatory Contexts
In emotionally sensitive contexts such as healthcare consultations, legal advice, or social work, the lay belief in AI’s rationality could lead people to discount their own feelings or to accept AI recommendations without adequate critical engagement. People are known to be sceptical of AI recommendations in hedonic or emotionally driven domains (Longoni and Cian, 2020), and conversely may resist AI involvement where they expect it to lack emotional attunement. Policymakers considering the deployment of AI in such settings should not assume that resistance to AI reflects irrational technophobia; it may instead reflect a reasoned judgement that the task requires emotional as well as analytical competence.
AI systems are frequently questioned for reductionistic and biased decision-making processes (Newman, Fast and Harman, 2020), underscoring the need for greater public and policymaker education and information. The present study adds a further dimension: even where AI systems are not biased, the mere belief that they are rational may lead people to defer to them in ways that are not always in their own interests.
Conclusion
This study makes a timely and practically meaningful contribution to our understanding of human-AI interaction. It shows that people’s lay beliefs about AI are strong, consistent, and behaviourally consequential. When people think they are dealing with a rational machine, they tend to become more rational themselves, at least in the economic sense (Vijayakumar, Yang and DeFranza, 2026).
For policymakers in Ireland and elsewhere, the message is clear: the adoption of AI in public-facing roles is not a neutral technical matter. It reshapes the cognitive context in which people make decisions. This has implications for how AI is introduced and communicated in public services, how AI literacy is developed among citizens, and how the use of AI is regulated in high-stakes environments.
As the authors conclude, decision-makers should be aware of the potential existence of lay beliefs about AI’s decision-making style and consider how such beliefs may influence the acceptance of AI recommendations and subsequent human decisions. In a policy environment where AI is advancing faster than public understanding, that is advice worth heeding.
The full study can be found here.
References
Castelo, N., Bos, M.W. and Lehmann, D.R. (2019) ‘Task-dependent algorithm aversion’, Journal of Marketing Research, 56(5), pp. 809–825.
Deloitte AI Institute (2021) The Government and Public Services AI Dossier. Available at: https://www2.deloitte.com/us/en/pages/consulting/articles/ai-dossier-government-public-services.html (Accessed: 1 April 2026).
Deloitte AI Institute (2024) State of Generative AI in the Enterprise: 2024 Year-End Generative AI Report. Available at: https://www2.deloitte.com/us/en/pages/consulting/articles/state-of-generative-ai-in-enterprise.html (Accessed: 1 April 2026).
Fehr, E., Naef, M. and Schmidt, K.M. (2006) ‘Inequality aversion, efficiency, and maximin preferences in simple distribution experiments: comment’, American Economic Review, 96(5), pp. 1912–1917.
Gill, T. (2020) ‘Blame it on the self-driving car: how autonomous vehicles can alter consumer morality’, Journal of Consumer Research, 47(2), pp. 272–291.
Granulo, A., Caprioli, S., Fuchs, C. and Puntoni, S. (2024) ‘Deployment of algorithms in management tasks reduces prosocial motivation’, Computers in Human Behavior, 152, 108094.
Güth, W., Schmittberger, R. and Schwarze, B. (1982) ‘An experimental analysis of ultimatum bargaining’, Journal of Economic Behavior and Organization, 3(4), pp. 367–388.
Hermann, I. (2021) ‘Artificial intelligence in fiction: between narratives and metaphors’, AI and Society, 38, p. 299.
Hong, J. and Chang, H.H. (2015) ‘“I” follow my heart and “we” rely on reasons: the impact of self-construal on reliance on feelings versus reasons in decision making’, Journal of Consumer Research, 41(6), pp. 1392–1411.
Jonsson, O. and de Luca Tena, C. (2021) European Tech Insights (Part II). IE Center for the Governance of Change. Available at: https://docs.ie.edu/cgc/IE-CGC-European-Tech-Insights-2021-(Part-II).pdf (Accessed: 1 April 2026).
Larney, A., Rotella, A. and Barclay, P. (2019) ‘Stake size effects in ultimatum game and dictator game offers: a meta-analysis’, Organizational Behavior and Human Decision Processes, 151, pp. 61–72.
Levitt, S.D. and List, J.A. (2008) ‘Homo economicus evolves’, Science, 319(5865), pp. 909–910.
Longoni, C. and Cian, L. (2020) ‘Artificial intelligence in utilitarian vs. hedonic contexts: the “word-of-machine” effect’, Journal of Marketing, 86(1), pp. 91–108.
Newman, D.T., Fast, N.J. and Harman, D. (2020) ‘When eliminating bias isn’t fair: algorithmic reductionism and procedural justice in human resource decisions’, Organizational Behavior and Human Decision Processes, 160, pp. 149–167.
Plaks, J.E., Levy, S.R. and Dweck, C.S. (2009) ‘Lay theories of personality: cornerstones of meaning in social cognition’, Social and Personality Psychology Compass, 3(6), pp. 1069–1081.
Sanfey, A.G., Rilling, J.K., Aronson, J.A., Nystrom, L.E. and Cohen, J.D. (2003) ‘The neural basis of economic decision-making in the ultimatum game’, Science, 300(5626), pp. 1755–1758.
Tully, S., Longoni, C. and Appel, G. (2025) ‘Lower artificial intelligence literacy predicts greater AI receptivity’, Journal of Marketing, 89(1), pp. 1–20.
UNESCO (2024) Generative AI: UNESCO Study Reveals Alarming Evidence of Regressive Gender Stereotypes. Paris: UNESCO.
Vijayakumar, S., Yang, W.Y. and DeFranza, D. (2026) ‘Lay belief about AI and its decision-making’, Frontiers in Computer Science, 8, 1768435. doi: 10.3389/fcomp.2026.1768435
von Walter, B., Kremmel, D. and Jäger, B. (2021) ‘The impact of lay beliefs about AI on adoption of algorithmic advice’, Marketing Letters, 33, pp. 143–155.