Humans program artificial delegates to accurately solve collective-risk dilemmas but lack precision

Area of Science:

• Behavioral Economics
• Human-Computer Interaction
• Artificial Intelligence Ethics

Background:

• Autonomous machines are increasingly influencing individual decisions with collective consequences.
• Understanding the impact of delegating decisions to artificial delegates is crucial for managing public goods.

Purpose of the Study:

• To investigate how delegating decisions to artificial agents affects contributions to public goods in a collective-risk dilemma.
• To identify potential losses in translation when humans delegate to algorithms.
• To explore the role of agent reprogramming in improving outcomes.

Main Methods:

• Behavioral experiments combining delegation to autonomous agents and choice architectures.
• Participants played a collective-risk dilemma game, deciding on contributions to a public good.
• A second round allowed participants to reprogram their agents to test learning and adaptation.

Main Results:

• Delegating decisions to agents, even with prior failures, led to increased contributions to public goods when action space was constrained.
• Delegation did not necessarily lead to greater success; precision errors in algorithms hindered target achievement.
• Reprogramming agents was necessary for potential success, highlighting the need for human oversight and adjustment.

Conclusions:

• Artificial delegates can aid in preserving public goods through repeated risky situations, but they are not a panacea.
• Human adaptation and precise adjustment of agent algorithms are essential for achieving success in delegated decision-making.
• The digitization of interactions requires a nuanced understanding of human-algorithm collaboration to optimize collective outcomes.