Summary: A new online experiment finds that more than 60% of participants in the UK and Germany preferred an algorithmic decision maker to a human when deciding how to redistribute earnings. At the same time, participants judged AI-made decisions as less satisfying and less fair. The study highlights the importance of transparency, accountability and consistent algorithmic behaviour to improve public trust in AI for morally relevant choices.
As automated systems play an increasing role in public policy, workplace decisions and social sorting, understanding how people view algorithmic versus human decision makers is essential. The research suggests that perceived impartiality drives a general openness to algorithmic decision making, but that acceptance depends on how well those systems explain their choices and adhere to recognised fairness principles.
Key Facts:
- AI Preference: Over 60% of participants preferred an algorithmic decision maker for redistributive choices.
- Perceived Fairness: Despite that preference, participants rated AI decisions as less satisfying and less fair than human decisions.
- Transparency and Consistency: The study emphasises the need for transparent, accountable algorithms and consistent decision rules to boost public acceptance.
Source: University of Portsmouth
Study overview
Researchers from the University of Portsmouth and the Max Planck Institute for Innovation and Competition ran an online decision experiment to compare public attitudes toward algorithmic and human decision makers in a redistributive setting. More than 200 participants from the UK and Germany completed tasks whose outcomes could then be redistributed between two people. Participants were asked to choose whether a human or an algorithm should decide how the earnings would be allocated.
Contrary to some earlier studies, the majority of participants in this experiment chose the algorithmic option for making redistributive decisions. This preference held regardless of whether the experimental scenarios implied potential discrimination. In other words, participants were more willing to delegate redistribution to algorithms even when bias concerns might plausibly arise.
Nevertheless, when asked to evaluate the actual decisions, participants expressed greater satisfaction with choices made by humans. Subjective evaluations of decisions were shaped largely by participants’ personal material interests and their own fairness ideals. Participants were generally tolerant of decisions that deviated moderately from their ideal outcomes when those deviations could be explained within familiar fairness frameworks. By contrast, redistributive choices that could not be reconciled with any established fairness principle provoked strong negative reactions.
Dr Wolfgang Luhan, Associate Professor of Behavioural Economics in the School for Accounting, Economics and Finance at the University of Portsmouth and corresponding author of the study, commented: “Our research suggests that while people are open to the idea of algorithmic decision-makers, especially due to their potential for unbiased decisions, the actual performance and the ability to explain how they decide play crucial roles in acceptance. Especially in moral decision-making contexts, the transparency and accountability of algorithms are vital. Many companies are already using AI for hiring decisions and compensation planning, and public bodies are employing AI in policing and parole strategies. Our findings suggest that, with improvements in algorithm consistency, the public may increasingly support algorithmic decision makers even in morally significant areas. If the right AI approach is taken, this could actually improve the acceptance of policies and managerial choices such as pay rises or bonus payments.”
The study therefore points to a nuanced public stance: people may prefer algorithmic decision makers for their impartiality, but they expect those algorithms to behave in ways that align with understandable notions of fairness and to be able to explain their choices when outcomes diverge from individual expectations. Improving algorithmic consistency, interpretability and accountability may be the key to broader acceptance of AI in morally charged domains.
About this AI research news
Author: Glenn Harris
Source: University of Portsmouth
Contact: Glenn Harris – University of Portsmouth
Image: The image is credited to Neuroscience News
Original Research: Open access. “Ruled by robots: preference for algorithmic decision makers and perceptions of their choices” by Wolfgang Luhan et al., published in Public Choice.
Abstract
Ruled by robots: preference for algorithmic decision makers and perceptions of their choices
As technology-assisted decision-making becomes more widespread, this study examines how the algorithmic nature of a decision maker influences perceptions among those affected. Using an online experiment focused on redistributive choices, the authors investigate whether perceived impartiality makes algorithmic decision makers preferable to humans. The majority of participants in this sample preferred algorithms, yet human-made decisions were judged more favourably. Subjective assessments were primarily driven by individual material interests and fairness ideals, with severe negative responses triggered by decisions that failed to conform to any recognised fairness principle.