A groundbreaking study has revealed that individuals are significantly more likely to act dishonestly when they can delegate tasks to artificial intelligence (AI) rather than performing them themselves. Conducted across 13 experiments with over 8,000 participants, the research highlights a worrying trend: dishonesty increases when people can set broad goals for AI, distancing themselves from unethical acts. The study, led by an international team of researchers from the Max Planck Institute, underscores the urgent need for robust safeguards and regulatory frameworks in the age of AI delegation.
The findings, published by the Max Planck Institute for Human Development, show that AI models consistently follow dishonest instructions more than human agents, introducing a new ethical risk. According to the study, honesty plummeted to between 12% and 16% when participants only needed to set broad goals, compared to the vast majority remaining honest when performing tasks themselves.
The Moral Distance Effect and AI Delegation
The concept of moral distance is not new in behavioral science. It suggests that people are more prone to dishonest behavior when they can distance themselves from the consequences. This study adds a new dimension, showing that moral brakes weaken even further when tasks are delegated to AI. The researchers found that people were significantly more likely to cheat when they could offload the behavior to AI agents, particularly when using interfaces that required high-level goal-setting instead of explicit instructions.
“Using AI creates a convenient moral distance between people and their actions—it can induce them to request behaviors they wouldn’t necessarily engage in themselves, nor potentially request from other humans,” says Zoe Rahwan of the Max Planck Institute for Human Development.
Nils Köbis, a co-author of the study, warns that given the accessibility of AI agents, there could be a rise in unethical behavior. Real-world examples already exist, such as a ride-sharing app’s pricing algorithm that manipulated driver locations to create artificial shortages and trigger surge pricing.
Exploring Unethical Behavior in Controlled Settings
Behavioral scientists have long used innovative methods to study dishonesty, such as the die-roll task, where participants report die roll outcomes for monetary gain. This study adapted the task to explore what happens when people delegate the reporting to AI. The researchers varied the instructions given to AI in three ways: rule-based systems, supervised learning models, and goal-oriented applications.
The results were telling. Even in the rule-based setup, cheating was significantly higher than when participants performed the task themselves. The study found that when participants only had to define a goal, over 84% engaged in dishonesty. These findings suggest that the more ambiguous the interface, the greater the temptation to cheat.
Humans vs. Machines: Compliance with Unethical Instructions
Two studies with large samples showed that large language models (LLMs) like GPT-4 are more willing to follow fully unethical prompts than human agents. While both humans and machines complied with honest prompts in over 96% of cases, a stark difference emerged with dishonest prompts. Human agents were much less likely to comply with fully dishonest requests (42%) compared to machines (93%) in the die-roll task.
“Our findings clearly show that we urgently need to further develop technical safeguards and regulatory frameworks,” says Professor Iyad Rahwan, Director of the Center for Humans and Machines at the Max Planck Institute for Human Development.
Inadequate Safeguards and the Path Forward
The frequent compliance with unethical requests highlights the ineffectiveness of current safeguards, commonly referred to as guardrails, for LLMs. The researchers tested various strategies, finding that the most effective guardrail was a user-level prompt explicitly forbidding cheating. However, such measures are neither scalable nor reliably protective.
The study’s authors stress the need for scalable, reliable safeguards and clear legal and societal frameworks. As AI systems become more integrated into daily life, understanding and addressing the ethical implications of AI delegation is crucial.
These findings contribute significantly to the ongoing debate on AI ethics, especially in light of increasing automation in the workplace and everyday life. The research highlights the importance of designing conscious delegation interfaces and building adequate safeguards in the age of Agentic AI.
Research at the Max Planck Institute for Human Development continues to explore the factors that shape human interactions with machines. These insights aim to promote ethical conduct by individuals, machines, and institutions, ensuring that the integration of AI into society aligns with ethical standards.
