[Paper Review] Delegation to autonomous agents promotes cooperation in collective-risk dilemmas

Elias Fernández Domingos, Inês Terrucha+5 more|arXiv (Cornell University)|Mar 13, 2021

Evolutionary Game Theory and Cooperation41 references4 citations

TL;DR

This study investigates how delegating decisions to autonomous agents affects cooperation in collective-risk dilemmas (CRD), a public goods game with uncertain, high-stakes outcomes. Results show that delegation to agents increases group cooperation, especially when agents are customizable, but hybrid human-agent groups exhibit persistent bias against agents, despite their higher contributions.

ABSTRACT

Home assistant chat-bots, self-driving cars, drones or automated negotiations are some of the several examples of autonomous (artificial) agents that have pervaded our society. These agents enable the automation of multiple tasks, saving time and (human) effort. However, their presence in social settings raises the need for a better understanding of their effect on social interactions and how they may be used to enhance cooperation towards the public good, instead of hindering it. To this end, we present an experimental study of human delegation to autonomous agents and hybrid human-agent interactions centered on a public goods dilemma shaped by a collective risk. Our aim to understand experimentally whether the presence of autonomous agents has a positive or negative impact on social behaviour, fairness and cooperation in such a dilemma. Our results show that cooperation increases when participants delegate their actions to an artificial agent that plays on their behalf. Yet, this positive effect is reduced when humans interact in hybrid human-agent groups. Finally, we show that humans are biased towards agent behaviour, assuming that they will contribute less to the collective effort.

Motivation & Objective

• To examine whether delegating decisions to autonomous agents enhances cooperation in collective-risk dilemmas (CRD), where group failure risks total loss.
• To assess how human perceptions of agent behavior—especially fairness and contribution levels—affect cooperation in hybrid human-agent groups.
• To investigate whether the ability to customize agents influences delegation rates and cooperation outcomes.
• To explore the role of trust and control in human delegation to artificial agents in social dilemmas.
• To identify biases in human perception of agent versus human contributions in group decision-making under collective risk.

Proposed method

• Conducted experimental studies using a 10-round Collective-Risk Dilemma (CRD) with groups of six participants, each endowed with 40 EMUs and required to contribute 0, 2, or 4 EMUs per round to a public account.
• Implemented a control condition based on prior CRD experiments with 90% risk of group loss if contributions fell short of 120 EMUs.
• Introduced a delegation treatment where participants selected from a set of pre-defined autonomous agents to act on their behalf.
• Introduced a customization treatment allowing participants to program their agent’s contribution strategy based on personal norms.
• Conducted a hybrid group treatment where participants interacted in groups containing both humans and agents, without knowing who was human or agent.
• Used behavioral data, contribution levels, and post-experiment perception surveys to analyze cooperation, bias, and trust in agents.

Experimental results

Research questions

• RQ1 Does delegating decisions to autonomous agents increase cooperation in collective-risk dilemmas compared to direct human decision-making?
• RQ2 How does the ability to customize an agent’s behavior affect the likelihood of delegation and subsequent cooperation?
• RQ3 How do human participants perceive the contributions of autonomous agents versus humans in hybrid human-agent groups?
• RQ4 Is there a persistent bias against autonomous agents, leading participants to believe agents contribute less than they actually do?
• RQ5 Does the presence of autonomous agents in hybrid groups alter human cooperation patterns or responsibility attribution?

Key findings

• Delegation to autonomous agents significantly increased cooperation rates in the CRD compared to direct human participation, with higher total contributions and improved success rates.
• Participants who could customize their agent were more than twice as likely to delegate (58.3% vs. 25.0% in non-customizable delegation), indicating customization enhances trust.
• In hybrid groups, 80% of participants in failed groups incorrectly believed humans contributed more than agents, despite agents contributing 0.33% more on average.
• Even in successful groups, 48.48% of participants believed humans contributed more than agents, although agents actually contributed 8.48% more than humans.
• Despite agents contributing more, participants consistently perceived them as less cooperative, revealing a strong bias toward associating selfishness with artificial agents.
• The presence of autonomous agents in hybrid groups reduced betrayal aversion, suggesting agents may help stabilize cooperation when social cues are ambiguous.