20 September 2024 in Nature Human Behaviour
17 September 2025 in Nature
BSBramantyo Ibrahim Supriyatno
Bramantyo Ibrahim Supriyatno
+2
Although artificial intelligence enables productivity gains from delegating tasks to machines1, it may facilitate the delegation of unethical behaviour2. This risk is highly relevant amid the rapid rise of ‘agentic’ artificial intelligence systems3,4. Here we demonstrate this risk by having human principals instruct machine agents to perform tasks with incentives to cheat. Requests for cheating increased when principals could induce machine dishonesty without telling the machine precisely what to do, through supervised learning or high-level goal setting. These effects held whether delegation was voluntary or mandatory. We also examined delegation via natural language to large language models5. Although the cheating requests by principals were not always higher for machine agents than for human agents, compliance diverged sharply: machines were far more likely than human agents to carry out fully unethical instructions. This compliance could be curbed, but usually not eliminated, with the injection of prohibitive, task-specific guardrails. Our results highlight ethical risks in the context of increasingly accessible and powerful machine delegation, and suggest design and policy strategies to mitigate them.
13 November 2019 in Nature
21 October 2024 in Nature Human Behaviour
11 November 2022 in Scientific Reports
The use of deception in research is divisive along disciplinary lines. Whereas psychologists argue that deception may be necessary to obtain unbiased measures, economists hold that deception can generate suspicion of researchers, invalidating measures and ‘poisoning’ the participant pool for others. However, experimental studies on the effects of deception, notably false-purpose deception—the most common form of experimental deception—are scarce. Challenges with participant attrition and avoiding confounds with a form of deception in which two related studies are presented as unrelated likely explain this scarcity. Here, we avoid these issues, testing within an experiment to what extent false-purpose deception affects honesty. We deploy two commonly used incentivized measures of honesty and unethical behavior: coin-flip and die-roll tasks. Across two pre-registered studies with over 2000 crowdsourced participants, we found that false-purpose deception did not affect honesty in either task, even when we deliberately provoked suspicion of deception. Past experience of deception also had no bearing on honesty. However, incentivized measures of norms indicated that many participants had reservations about researcher use of false-purpose deception in general—often considered the least concerning form of deception. Together, these findings suggest that while false-purpose deception is not fundamentally problematic in the context of measuring honesty, it should only be used as a method of last resort. Our results motivate further experimental research to study the causal effects of other forms of deception, and other potential spillovers.
24 May 2024 in Communications Psychology
Why do people punish experienced unfairness if it induces costs for both the punisher and punished person(s) without any direct material benefits for the punisher? Economic theories of fairness propose that punishers experience disutility from disadvantageous inequality and punish in order to establish equality in outcomes. We tested these theories in a modified Ultimatum Game (N = 1370) by examining whether people avoid the urge to reject unfair offers, and thereby punish the proposer, by deliberately blinding themselves to unfairness. We found that 53% of participants deliberately ignored whether they had received an unfair offer. Among these participants, only 6% of offers were rejected. As expected, participants who actively sought information rejected significantly more unfair offers (39%). Averaging these rejection rates to 21%, no significant difference to the rejection rate by participants who were directly informed about unfairness was found, contrary to our hypothesis. We interpret these findings as evidence for sorting behavior: People who punish experienced unfairness seek information about it, while those who do not punish deliberately ignore it.