MEDIATE: Mutually Endorsed Distributed Incentive Acknowledgment Token Exchange

TL;DR

MEDIATE addresses the challenge of fostering cooperation in decentralized multi-agent systems under privacy constraints by introducing automatic, domain-adaptive incentivization tokens derived from local value estimates and a privacy-preserving consensus mechanism. It extends the MATE framework with per-agent token derivation and additive secret sharing-based consensus, enabling tokens to adapt to varying reward landscapes while preserving privacy. Empirical results across Iterated Prisoner's Dilemma, CoinGame variants, and Harvest show MEDIATE improves or matches state-of-the-art PI approaches, with tokens converging within the first $1000$ epochs and coordinated token exchange yielding robust social welfare gains. This work provides a scalable, privacy-conscious protocol that enhances cooperative behavior in diverse social-dilemma environments and offers a foundation for future adversarial and large-scale evaluations.

Abstract

Recent advances in multi-agent systems (MAS) have shown that incorporating peer incentivization (PI) mechanisms vastly improves cooperation. Especially in social dilemmas, communication between the agents helps to overcome sub-optimal Nash equilibria. However, incentivization tokens need to be carefully selected. Furthermore, real-world applications might yield increased privacy requirements and limited exchange. Therefore, we extend the PI protocol for mutual acknowledgment token exchange (MATE) and provide additional analysis on the impact of the chosen tokens. Building upon those insights, we propose mutually endorsed distributed incentive acknowledgment token exchange (MEDIATE), an extended PI architecture employing automatic token derivation via decentralized consensus. Empirical results show the stable agreement on appropriate tokens yielding superior performance compared to static tokens and state-of-the-art approaches in different social dilemma environments with various reward distributions.

Figures (5)

• Figure 1: Rate of own coins for different tokens when determined centralized (\ref{['fig:token-analysis:central']}) and decentralized (\ref{['fig:token-analysis:decentral']})
• Figure 2: MEDIATE Architecture
• Figure 3: MEDIATE Evaluation: Comparing the mean Efficiency (Fig. \ref{['fig:eval:e:ipd']}) and rate of Own Coins (Fig. \ref{['fig:eval:e:coin2']}, \ref{['fig:eval:e:coin4']}) of Naïve Learning (grey), MATE (blue), AutoMATE (light blue), MEDIATE-I (orange), and MEDIATE-S (green), and the Mean Token Value (Fig. \ref{['fig:eval:t:ipd']}, \ref{['fig:eval:t:coin2']}, \ref{['fig:eval:t:coin4']}) in the IPD (Fig. \ref{['fig:eval:e:ipd']}, \ref{['fig:eval:t:ipd']}), CoinGame-2 (Fig. \ref{['fig:eval:e:coin2']}, \ref{['fig:eval:t:coin2']}), and CoinGame-4 (Fig. \ref{['fig:eval:e:coin4']}, \ref{['fig:eval:t:coin4']}). The shaded areas mark the 95% confidence intervals.
• Figure 4: Benchmark Comparison: Mean rate of Own Coins (Fig. \ref{['fig:bench:rcoin2']}, \ref{['fig:bench:coin6']}) and Efficiency (Fig. \ref{['fig:bench:harvest']}) of MEDIATE-S (green), MEDIATE-I (orange), MATE (blue), LIO (red), Budget-Gifting (purple), Zerosum-Gifting (pink) and Naïve Learning (grey) in the Rescaled CoinGame-2 (RCG-2) (Fig. \ref{['fig:bench:rcoin2']}), CoinGame-6 (CG-6) (Fig. \ref{['fig:bench:coin6']}), and Harvest (Fig. \ref{['fig:bench:harvest']}). The shaded areas mark the 95% confidence intervals.
• Figure 5: Evaluation Environments