Hou and Jung (2021) showed that the contradictory literatures on algorithm aversion and algorithm appreciation can be reversed at will: in a judge–advisor task where advice quality was experimentally controlled, changing only how the human and algorithmic advisors were presented — labels, descriptions, exemplar images, and a one-versus-many aggregation contrast — flipped whether participants followed the algorithm's or the human's advice more, and the agent-type effect was mediated by the advisor's perceived competence (expert power). Because each framing bundled several cues at once, and only quantity-estimation and creativity-judgment tasks were studied, the composition and scope of the mechanism remain unknown. This problem asks for a systematic empirical account that (i) isolates the separate and joint contributions of the individual framing elements to perceived expert power and advice uptake, for algorithmic and for human advisors; (ii) establishes whether the framing–expert-power–influence mechanism generalizes to other decision types, notably moral judgments, medical diagnosis, and other high-stakes settings, and to realistic rather than stripped-down laboratory contexts; and (iii) identifies additional factors that explain people whose advice uptake does not follow perceived expert power. The problem remains open: later work has studied isolated cues (terminology, accuracy disclosure, expertise information) and compared agents in moral and medical domains, but no study factorially decomposes the framing elements through expert power or tests the mechanism's generalization.

Background

As algorithms increasingly support human decisions, two contradictory findings have coexisted. Algorithm aversion is the observation that people reject algorithmic advice even after seeing it outperform human judgment, for example after watching the algorithm make occasional errors (Dietvorst, Simmons, and Massey, 2015). Algorithm appreciation is the opposite pattern: people weight identical algorithmic advice more heavily than the same advice attributed to a human (Logg, Minson, and Moore, 2019). In the moral domain, Bigman and Gray (2018) found that people are specifically averse to machines making morally weighty decisions. Much of this evidence comes from the judge–advisor paradigm, a standard advice-taking setup in which a participant (the judge) gives an initial estimate, receives a suggestion from an advisor, and gives a final estimate; influence is measured by how far the final estimate moves toward the suggestion (the influence factor, i.e., the fraction of the initial-to-suggestion gap that the participant closes). Hou and Jung (2021) asked what reconciles the contradictory results. In three experiments using a modified judge–advisor numeric estimation task in which each suggestion was yoked to the participant's own initial answer (initial answer plus or minus a random offset between 6 and 9, so advice quality was equated across conditions), they varied only how the advice source was presented — its label, its description, and an exemplar image. One framing produced algorithm appreciation (mean influence factor 0.57 for algorithmic versus 0.45 for human suggestions), and an alternative framing of the same task reversed the pattern into algorithm aversion (0.58 for human versus 0.45 for algorithmic). A six-condition experiment crossing human framings (Expert, Mturker, Person) and machine framings (AI, Algorithm, Computer) showed how different framing choices could generate the inconsistent outcomes scattered through earlier studies. Crucially, the agent-type effect was mediated by perceived competence: the authors measured each advisor's expert power — the competence-based component of a social-psychological framework that decomposes an influence source's power into bases such as expert power (perceived knowledge), information power, referent power, and legitimate power — using an adapted Interpersonal Power Inventory, and bootstrapped mediation analyses showed perceived expert power accounted for the framing effects. Two limitations define the open question. First, the framings bundled several cues simultaneously: agent labels (such as AI, algorithm, or computer), provenance and performance descriptions, exemplar images, and a one-versus-many contrast (a single named expert versus advice aggregated from a crowd), so no single framing factor was isolated. Second, the tasks came from only two families: an objective quantity-estimation family and a subjective creativity/popularity-judgment family, leaving moral judgments, medical diagnosis, and other high-stakes decisions untested, together with realistic settings in which advisors carry cross-domain reputations. In addition, although expert power explained the main pattern, some participants did not follow it, indicating further unidentified influences on how people respond to algorithmic decision support.

Problem Statement

The overarching objective is a systematic empirical account of the framing-to-expert-power mechanism in advice uptake: which presentation cues of an advice source raise or lower its perceived competence, how those cues combine, whether the same mechanism operates for algorithmic and human advisors, and how far beyond the original laboratory tasks it reaches. Within that single objective, three linked questions must be answered. (1) Decomposition: isolate the separate contributions of the individual framing elements identified by Hou and Jung (2021) — agent names or labels, provenance and performance descriptions, exemplar images, and the one-versus-many contrast between a single advisor and aggregated crowd advice — to perceived expert power and to actual advice uptake (for example, influence factors in a judge–advisor task), and determine how these elements interact with one another and whether their effects differ between algorithmic and human advisors. (2) Scope: establish whether the framing–expert-power–influence mechanism, demonstrated only on an objective quantity-estimation family and a subjective creativity-judgment family, generalizes to other decision types — including moral judgments, medical diagnosis, and other high-stakes decisions — and to realistic settings where advisors carry cross-domain reputations, with perceived expert power measured and tested as mediator in each new domain. (3) Residuals: identify factors beyond expert power that explain the participants whose advice uptake does not track the advisor's perceived competence. The scope of this account is the decision types named above (moral judgments, medical diagnosis, and other high-stakes settings) together with realistic settings, not every conceivable domain.

References

  1. Yoyo Tsung-Yu Hou and Malte F. Jung (2021). Who is the Expert? Reconciling Algorithm Aversion and Algorithm Appreciation in AI-Supported Decision Making. Proceedings of the ACM on Human-Computer Interaction 5(CSCW2). DOI: 10.1145/3479864. https://doi.org/10.1145/3479864
  2. Berkeley J. Dietvorst, Joseph P. Simmons, and Cade Massey (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General. DOI: 10.1037/xge0000033. https://doi.org/10.1037/xge0000033
  3. Jennifer M. Logg, Julia A. Minson, and Don A. Moore (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes. DOI: 10.1016/j.obhdp.2018.12.005. https://doi.org/10.1016/j.obhdp.2018.12.005
  4. Yochanan E. Bigman and Kurt Gray (2018). People are averse to machines making moral decisions. Cognition. DOI: 10.1016/j.cognition.2018.08.003. https://doi.org/10.1016/j.cognition.2018.08.003
  5. Markus Langer, Tim Hunsicker, Tina Feldkamp, Cornelius J. König, and Nina Grgić-Hlača (2022). 'Look! It's a Computer Program! It's an Algorithm! It's AI!': Does Terminology Affect Human Perceptions and Evaluations of Algorithmic Decision-Making Systems? Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems. DOI: 10.1145/3491102.3517527. https://doi.org/10.1145/3491102.3517527
  6. Sander de Jong, Maarten W. Bos, Niels van Berkel, and Maarten H. Lamers (2025). Algorithm appreciation or aversion: the effects of accuracy disclosure on users' reliance on algorithmic suggestions. Behaviour & Information Technology. DOI: 10.1080/0144929X.2025.2535732. https://doi.org/10.1080/0144929X.2025.2535732
  7. Ben Greiner, Philipp Grünwald, Thomas Lindner, Georg Lintner, and Martin Wiernsperger (2026). Incentives, Framing, and Reliance on Algorithmic Advice: An Experimental Study. Management Science 72(1), 302–322. DOI: 10.1287/mnsc.2022.02777. https://doi.org/10.1287/mnsc.2022.02777
  8. Micha Kaiser, Cass R. Sunstein, and Lucia A. Reisch (2025). On preferring people to algorithms. Journal of Risk and Uncertainty. DOI: 10.1007/s11166-025-09466-6. https://doi.org/10.1007/s11166-025-09466-6
  9. Yoyo Tsung-Yu Hou, Wen-Ying Lee, and Malte Jung (2023). 'Should I Follow the Human, or Follow the Robot?' — Robots in Power Can Have More Influence Than Humans on Decision-Making. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. DOI: 10.1145/3544548.3581066. https://doi.org/10.1145/3544548.3581066
  10. Yoyo Tsung-Yu Hou, EunJeong Cheon, and Malte F. Jung (2024). Power in Human-Robot Interaction. Proceedings of the 2024 ACM/IEEE International Conference on Human-Robot Interaction. DOI: 10.1145/3610977.3634949. https://doi.org/10.1145/3610977.3634949
  11. Joshua Mu-En Liu and Yoyo Tsung-Yu Hou (2026). Obedient Machines Make Us Everyday Authoritarians: A Power Framework and Research Agenda for Human–AI Interaction. Proceedings of the 8th ACM Conference on Conversational User Interfaces. DOI: 10.1145/3816046.3816305. https://doi.org/10.1145/3816046.3816305
  12. Bertram F. Malle, Matthias Scheutz, Corey Cusimano, John Voiklis, Takanori Komatsu, Stuti Thapa, and Salomi Aladia (2025). People's judgments of humans and robots in a classic moral dilemma. Cognition. DOI: 10.1016/j.cognition.2024.105958. https://doi.org/10.1016/j.cognition.2024.105958
  13. Irene Locci and Sébastien Massoni (2024). Judged by Robots: Preferences and Perceived Fairness of Algorithmic versus Human Punishments. Review of Law & Economics. DOI: 10.1515/rle-2024-0063. https://doi.org/10.1515/rle-2024-0063
  14. Henrik H. J. Detjen, Lars Densky, Niklas von Kalckreuth, and Marvin Kopka (2025). Who is Trusted for a Second Opinion? Comparing Collective Advice from a Medical AI and Physicians in Biopsy Decisions After Mammography Screening. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. DOI: 10.1145/3706598.3713898. https://doi.org/10.1145/3706598.3713898
  15. Ibrahim Filiz, Jan René Judek, Marco Lorenz, and Markus Spiwoks (2023). The extent of algorithm aversion in decision-making situations with varying gravity. PLOS ONE. DOI: 10.1371/journal.pone.0278751. https://doi.org/10.1371/journal.pone.0278751
  16. Sarah A. Jessup, Gene M. Alarcon, Sasha M. Willis, and Michael A. Lee (2024). A closer look at how experience, task domain, and self-confidence influence reliance towards algorithms. Applied Ergonomics. DOI: 10.1016/j.apergo.2024.104363. https://doi.org/10.1016/j.apergo.2024.104363
  17. Zhuoran Lu, Dakuo Wang, and Ming Yin (2024). Does More Advice Help? The Effects of Second Opinions in AI-Assisted Decision Making. Proceedings of the ACM on Human-Computer Interaction 8(CSCW1). DOI: 10.1145/3653708. https://doi.org/10.1145/3653708
  18. Julia I. Baines, Reeshad S. Dalal, Lida P. Ponce, and Ho-Chun Tsai (2024). Advice from artificial intelligence: a review and practical implications. Frontiers in Psychology. DOI: 10.3389/fpsyg.2024.1390182. https://doi.org/10.3389/fpsyg.2024.1390182
  19. Sven Eckhardt, Niklas Kühl, Mateusz Dolata, and Gerhard Schwabe (2025). A Survey of AI Reliance. ACM Computing Surveys. DOI: 10.1145/3776528. https://doi.org/10.1145/3776528