Human-centred mechanism design with Democratic AI

Köster, Raphael; Jan, Balaguer; Tacchetti, Andrea; Weinstein, Ari; Zhu, Tina; Hauser, Oliver P.; Williams, Duncan; Campbell-Gillingham, Lucy; Thacker, Phoebe; Botvinick, Matthew; Summerfield, Christopher

doi:10.1038/s41562-022-01383-x

Cited by 26 publications

(4 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Considering other models may change the system dynamics, and perhaps improve the PoA. Moreover, designing an optimal fairness rule may be done similarly as for voting mechanisms in the manner of [8].…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Price of Anarchy in a Double-Sided Critical Distribution System

Sychrovský¹,

Černý²,

Lichau³

et al. 2023

Preprint

View full text Add to dashboard Cite

Measures of allocation optimality differ significantly when distributing standard tradable goods in peaceful times and scarce resources in crises. While realistic markets offer asymptotic efficiency, they may not necessarily guarantee fair allocation desirable when distributing the critical resources. To achieve fairness, mechanisms often rely on a central authority, which may act inefficiently in times of need when swiftness and good organization are crucial. In this work, we study a hybrid trading system called Crisdis, introduced by Jedličková et al., which combines fair allocation of buying rights with a market -leveraging the best of both worlds. A frustration of a buyer in Crisdis is defined as a difference between the amount of goods they are entitled to according to the assigned buying rights and the amount of goods they are able to acquire by trading. We define a Price of Anarchy (PoA) in this system as a conceptual analogue of the original definition in the context of frustration. Our main contribution is a study of PoA in realistic complex double-sided market mechanisms for Crisdis. The performed empirical analysis suggests that in contrast to market free of governmental interventions, the PoA in our system decreases.

show abstract

Section: Discussionmentioning

confidence: 99%

“…maximal, amount of good bought by a buyer, and the sense of the objective function ensure that it will be exactly this quantity. The constraint (8) imposes that the buyers pay at most in average 𝑝 𝐺 𝑏 for the Good. The constraint (9) forces that the buyers pay at most in average 𝑝 𝑅 𝑏 for the Right.…”

Section: A Hyperparametersmentioning

confidence: 99%

Price of Anarchy in a Double-Sided Critical Distribution System

Sychrovský¹,

Černý²,

Lichau³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Traditionally, a classical model to study the effect of inequality on cooperation is the linear public goods game [ 26 , 27 ]. In this game, individuals decide how much of their endowment to contribute to a public good.…”

Section: An Overview Of the Previous Literature On Asymmetric Public ...mentioning

confidence: 99%

Cooperation and coordination in heterogeneous populations

Wang

Couto

Wang

et al. 2023

Phil. Trans. R. Soc. B

View full text Add to dashboard Cite

One landmark application of evolutionary game theory is the study of social dilemmas. This literature explores why people cooperate even when there are strong incentives to defect. Much of this literature, however, assumes that interactions are symmetric. Individuals are assumed to have the same strategic options and the same potential pay-offs. Yet many interesting questions arise once individuals are allowed to differ. Here, we study asymmetry in simple coordination games. In our set-up, human participants need to decide how much of their endowment to contribute to a public good. If a group’s collective contribution reaches a pre-defined threshold, all group members receive a reward. To account for possible asymmetries, individuals either differ in their endowments or their productivities. According to a theoretical equilibrium analysis, such games tend to have many possible solutions. In equilibrium, group members may contribute the same amount, different amounts or nothing at all. According to our behavioural experiment, however, humans favour the equilibrium in which everyone contributes the same proportion of their endowment. We use these experimental results to highlight the non-trivial effects of inequality on cooperation, and we discuss to which extent models of evolutionary game theory can account for these effects. This article is part of the theme issue ‘Half a century of evolutionary games: a synthesis of theory, application and future directions'.

show abstract

“…Several recent research efforts propose using machine learning to identify novel solutions to social challenges (e.g., Koster et al, 2022;Zheng et al, 2020). Artificial intelligence and machine learning systems increasingly suffuse everyday social processes (Wagner et al, 2021), so it seems natural to ask how they might support beneficial outcomes for human communities.…”

Section: Introductionmentioning

confidence: 99%

Scaffolding cooperation in human groups with deep reinforcement learning

McKee¹,

Tacchetti²,

Balaguer³

et al. 2022

Preprint

View full text Add to dashboard Cite

Altruism and selfishness are highly transmissible. Either can easily cascade through human communities. Effective approaches to encouraging group cooperation—while also mitigating the risk of spreading defection—are still an open challenge. Here, we apply recent advances in deep reinforcement learning to structure networks of human participants playing a group cooperation game. We leverage deep reinforcement learning and simulation methods to train a "social planner" capable of making recommendations to create or break connections between group members. This social planner learns a strategy from scratch, through repeated trial and error. The strategy that it develops succeeds at encouraging prosociality in networks of human participants (N = 208 participants in 13 sessions) playing the group cooperation game for real monetary stakes. Under the social planner, groups finished the game with an average cooperation rate of 77.7%, compared to 42.8% in static networks (N = 176 participants in 11 sessions). In contrast to prior strategies that separate defectors from cooperators (tested here with N = 384 participants in 24 sessions), the social planner learns to take a conciliatory approach to defectors, encouraging them to act prosocially by moving them to small, highly-cooperative neighborhoods. A subsequent validation study (N = 224 participants in 14 sessions) confirms that this encouraging approach can scaffold group cooperation absent the "black box" of deep learning. Deep learning can help explore and identify new approaches to support human cooperation and prosociality.

show abstract

Human-centred mechanism design with Democratic AI

Cited by 26 publications

References 35 publications

Price of Anarchy in a Double-Sided Critical Distribution System

Price of Anarchy in a Double-Sided Critical Distribution System

Cooperation and coordination in heterogeneous populations

Scaffolding cooperation in human groups with deep reinforcement learning

Contact Info

Product

Resources

About