AI Ethics Digital Twin - Evolutionary Institutional Design in Multi-Agent AI Societies
Fuente:
Zenodo
Guardado en:
| Autor principal: | |
|---|---|
| Formato: | Recurso digital |
| Lenguaje: | inglés |
| Publicado: |
Zenodo
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866901604303634432 |
|---|---|
| author | Yamamoto, Eiji |
| author_facet | Yamamoto, Eiji |
| contents | <p>The ethical challenges of AI in society emerge not only from the design of individual AI</p> <p>systems but also from “emergent dilemmas” that arise through the interactions among multi-</p> <p>ple AI agents. In response, this study proposes a new framework—the AI Ethics Digital Twin</p> <p>(AEDT)—that reconceptualizes ethics not as fixed norms to be embedded, but as evolving con-</p> <p>structs that can be guided and shaped through institutional design and evolutionary processes.</p> <p>Integrating evolutionary game theory, multi-agent simulation, and mechanism design, we</p> <p>conduct a proof-of-concept (PoC) experiment comparing three institutional strategies—coercive,</p> <p>tolerance-based, and incentive-driven—regarding resource harvesting behaviors among AI agents.</p> <p>Our results demonstrate that even under initially unethical conditions, cooperation can evolve</p> <p>and stabilize through appropriate combinations of rewards, penalties, and imitation. Further-</p> <p>more, we show that institutional design significantly affects sustainability, efficiency, and fairness</p> <p>in collective outcomes.</p> <p>This research presents AEDT as a constructive and experimental framework for exploring</p> <p>the co-evolution of ethics and institutional design, offering both theoretical and practical con-</p> <p>tributions to the development of future AI governance.</p> <p>Keywords: AI ethics, multi-agent simulation, evolutionary game theory, mechanism design,</p> <p>institutional governance, digital twin</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_15656504 |
| institution | Zenodo |
| language | eng |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | AI Ethics Digital Twin - Evolutionary Institutional Design in Multi-Agent AI Societies Yamamoto, Eiji Computer Systems/ethics Artificial intelligence Artificial Intelligence <p>The ethical challenges of AI in society emerge not only from the design of individual AI</p> <p>systems but also from “emergent dilemmas” that arise through the interactions among multi-</p> <p>ple AI agents. In response, this study proposes a new framework—the AI Ethics Digital Twin</p> <p>(AEDT)—that reconceptualizes ethics not as fixed norms to be embedded, but as evolving con-</p> <p>structs that can be guided and shaped through institutional design and evolutionary processes.</p> <p>Integrating evolutionary game theory, multi-agent simulation, and mechanism design, we</p> <p>conduct a proof-of-concept (PoC) experiment comparing three institutional strategies—coercive,</p> <p>tolerance-based, and incentive-driven—regarding resource harvesting behaviors among AI agents.</p> <p>Our results demonstrate that even under initially unethical conditions, cooperation can evolve</p> <p>and stabilize through appropriate combinations of rewards, penalties, and imitation. Further-</p> <p>more, we show that institutional design significantly affects sustainability, efficiency, and fairness</p> <p>in collective outcomes.</p> <p>This research presents AEDT as a constructive and experimental framework for exploring</p> <p>the co-evolution of ethics and institutional design, offering both theoretical and practical con-</p> <p>tributions to the development of future AI governance.</p> <p>Keywords: AI ethics, multi-agent simulation, evolutionary game theory, mechanism design,</p> <p>institutional governance, digital twin</p> |
| title | AI Ethics Digital Twin - Evolutionary Institutional Design in Multi-Agent AI Societies |
| topic | Computer Systems/ethics Artificial intelligence Artificial Intelligence |
| url | https://doi.org/10.5281/zenodo.15656504 |