Afin de répondre au besoin d’incorporer des considérations éthiques au sein d’algorithmes d’Intelligence Artificielle, nous proposons une nouvelle méthode hybride, combinant raisonnement et apprentissage, où des agents juges évaluent l’éthique du comportement d’agents apprenants. Cette séparation offre plusieurs avantages : co-construction entre agents et humains ; juges plus accessibles pour des humains non-experts ; récompense plus riche par l’utilisation de multiples valeurs morales. Les expérimentations sur la distribution de l’énergie dans un simulateur de Smart Grid montrent la capacité des agents apprenants à se conformer aux règles des agents juges, y compris lorsque les règles évoluent.
To answer the need to imbue Artificial Intelligence algorithms with ethical considerations, this article propose a method combining reasoning and learning, where judging agents evaluate the ethics of learning agents’ behavior. This separation offers several advantages: co-construction between agents and humans; judges more accessible for non-experts humans; richer feedback by using multiple judgments. Experiments on energy distribution inside a Smart Grid simulator show the learning agents’ ability to comply with judging agents’ rules, including when they evolve.
Accepté le :
Publié le :
Keywords: Ethics, Machine Ethics, Multi-Agent Learning, Reinforcement Learning, Hybrid Neural-Symbolic Learning, Ethical Judgment.
Rémy Chaput 1 ; Jérémy Duval 2 ; Olivier Boissier 3 ; Mathieu Guillermin 4 ; Salima Hassas 1
CC-BY 4.0
@article{ROIA_2023__4_2_41_0,
author = {R\'emy Chaput and J\'er\'emy Duval and Olivier Boissier and Mathieu Guillermin and Salima Hassas},
title = {Apprentissage de comportements \'ethiques multi-valeurs par combinaison d{\textquoteright}agents juges symboliques et d{\textquoteright}agents apprenants},
journal = {Revue Ouverte d'Intelligence Artificielle},
pages = {41--66},
publisher = {Association pour la diffusion de la recherche francophone en intelligence artificielle},
volume = {4},
number = {2},
year = {2023},
doi = {10.5802/roia.56},
language = {fr},
url = {https://roia.centre-mersenne.org/articles/10.5802/roia.56/}
}
TY - JOUR AU - Rémy Chaput AU - Jérémy Duval AU - Olivier Boissier AU - Mathieu Guillermin AU - Salima Hassas TI - Apprentissage de comportements éthiques multi-valeurs par combinaison d’agents juges symboliques et d’agents apprenants JO - Revue Ouverte d'Intelligence Artificielle PY - 2023 SP - 41 EP - 66 VL - 4 IS - 2 PB - Association pour la diffusion de la recherche francophone en intelligence artificielle UR - https://roia.centre-mersenne.org/articles/10.5802/roia.56/ DO - 10.5802/roia.56 LA - fr ID - ROIA_2023__4_2_41_0 ER -
%0 Journal Article %A Rémy Chaput %A Jérémy Duval %A Olivier Boissier %A Mathieu Guillermin %A Salima Hassas %T Apprentissage de comportements éthiques multi-valeurs par combinaison d’agents juges symboliques et d’agents apprenants %J Revue Ouverte d'Intelligence Artificielle %D 2023 %P 41-66 %V 4 %N 2 %I Association pour la diffusion de la recherche francophone en intelligence artificielle %U https://roia.centre-mersenne.org/articles/10.5802/roia.56/ %R 10.5802/roia.56 %G fr %F ROIA_2023__4_2_41_0
Rémy Chaput; Jérémy Duval; Olivier Boissier; Mathieu Guillermin; Salima Hassas. Apprentissage de comportements éthiques multi-valeurs par combinaison d’agents juges symboliques et d’agents apprenants. Revue Ouverte d'Intelligence Artificielle, Post-actes des Journées Francophones sur les Systèmes Multi-Agents (JFSMA 2021), Volume 4 (2023) no. 2, pp. 41-66. doi : 10.5802/roia.56. https://roia.centre-mersenne.org/articles/10.5802/roia.56/
[1] Artificial Morality : Top-down, Bottom-up, and Hybrid Approaches, Ethics and Information Technology, Volume 7 (2005) no. 3, pp. 149-155 | DOI
[2] Toward ensuring ethical behavior from autonomous systems : a case-supported principle-based paradigm, Workshops at the Twenty-Ninth AAAI Conference on Artificial Intelligence (2015)
[3] A Value-Driven Eldercare Robot : Virtual and Physical Instantiations of a Case-Supported Principle-Based Behavior Paradigm, Proc. IEEE, Volume 107 (2019) no. 3, pp. 526-540 | DOI
[4] An ethical governor for constraining lethal action in an autonomous system (2009) (Technical report) | DOI
[5] The moral machine experiment, Nature, Volume 563 (2018) no. 7729, pp. 59-64 | DOI
[6] The Acceptability of Decentralized Energy Systems, master thesis, Delft University of Technology (2019)
[7] Multi-Agent Oriented Programming : Programming Multi-Agent Systems Using JaCaMo, The MIT Press, 2020 | DOI
[8] Formal ethical reasoning and dilemma identification in a human-artificial agent system, Ph. D. Thesis, Institut supérieur de l’aéronautique et de l’espace, Toulouse, France (2019)
[9] Vérification formelle et éthique dans les SMA, Systèmes Multi-Agents et simulation – Vingt-quatrièmes journées francophones sur les systèmes multi-agents, JFSMA 16, Saint-Martin-du-Vivier (Rouen), France, Octobre 5-7, 2016 (Fabien Michel; Julien Saunier, eds.), Cépaduès Éditions (2016), pp. 139-148
[10] Agent programming in the cognitive era, Autonomous Agents and Multi-Agent Systems, Volume 34 (2020) no. 2, pp. 1-31 | DOI
[11] Programming multi-agent systems in AgentSpeak using Jason, 8, John Wiley & Sons, 2007
[12] From programming agents to educating agents – a jason-based framework for integrating learning in the development of cognitive agents, International Workshop on Engineering Multi-Agent Systems, Springer (2019), pp. 175-194 | DOI
[13] On proactive, transparent, and verifiable ethical reasoning for robots, Proceedings of the IEEE, Volume 107 (2019) no. 3, pp. 541-561 | DOI
[14] Apprentissage adaptatif de comportements éthiques, Architectures multi-agents pour la simulation de systeèmes complexes - Vingt-huitième journées francophones sur les systèmes multi-agents, JFSMA 2020, Angers, France, June 29 - July 3, 2020 (Nicolas Sabouret, ed.), Cépaduès (2020)
[15] Jugement éthique dans les systèmes multi-agents, Systèmes Multi-Agents et simulation – Vingt-quatrièmes journées francophones sur les systèmes multi-agents, JFSMA 16, Saint-Martin-du-Vivier (Rouen), France, Octobre 5-7, 2016 (Fabien Michel; Julien Saunier, eds.), Cépaduès Éditions (2016), pp. 149-158
[16] Multi-agent based ethical asset management, 1st Workshop on Ethics in the Design of Intelligent Agents (2016), pp. 52-57
[17] Practical Challenges in Explicit Ethical Machine Reasoning, International Symposium on Artificial Intelligence and Mathematics, ISAIM 2018, Fort Lauderdale, Florida, USA, January 3-5, 2018 (2018) http://isaim2018.cs.virginia.edu/papers/ISAIM2018_Ethics_Dennis_Fischer.pdf
[18] Responsible Artificial Intelligence : How to Develop and Use AI in a Responsible Way, Springer Nature, 2019 | DOI
[19] Safe reinforcement learning via formal methods : Toward safe control through proof and learning, Proceedings of the AAAI Conference on Artificial Intelligence, volume 1, Volume 32 (2018) | DOI
[20] An artificial neural network approach for creating an ethical artificial agent, 2009 IEEE International Symposium on Computational Intelligence in Robotics and Automation-(CIRA), IEEE (2009), pp. 290-295 | DOI
[21] Essentials of the self-organizing map, Neural Networks, Volume 37 (2013), pp. 52-65 | DOI
[22] The nature, importance, and difficulty of machine ethics, IEEE intelligent systems, Volume 21 (2006) no. 4, pp. 18-21 | DOI
[23] Landscape of machine implemented ethics, Science and engineering ethics, Volume 26 (2020) no. 5, pp. 2381-2399 | DOI
[24] Soar-RL : Integrating reinforcement learning with Soar, Cognitive Systems Research, Volume 6 (2005) no. 1, pp. 51-59 | DOI
[25] Dynamic self-organising map, Neurocomputing, Volume 74 (2011) no. 11, pp. 1840-1847 | DOI
[26] What’s Next for AI Ethics, Policy, and Governance ? A Global Overview, Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (2020), pp. 153-158 | DOI
[27] Reinforcement learning : An introduction, MIT press, 2018
[28] Killing, letting die, and the trolley problem, The Monist, Volume 59 (1976) no. 2, pp. 204-217 | DOI
[29] Implementations in Machine Ethics : A Survey, ACM Comput. Surv., Volume 53 (2021) no. 6, 132, 38 pages | DOI
[30] Q-Learning, Machine Learning, Volume 8 (1992) no. 3, pp. 279-292 | DOI | Zbl
[31] A low-cost ethics shaping approach for designing reinforcement learning agents, Proceedings of the AAAI Conference on Artificial Intelligence, Volume 32 (2018) no. 1, pp. 1687-1694 | DOI
[32] Building Ethics into Artificial Intelligence, Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI’18), AAAI Press, Stockholm, Sweden (2018), pp. 5527-5533 | DOI
Cité par Sources :