Please use this identifier to cite or link to this item:
logo share SHARE BASE
Visualizar otros formatos: MARC | Dublin Core | RDF | ORE | MODS | METS | DIDL | DATACITE

Guaranteeing the Learning of Ethical Behaviour through Multi-Objective Reinforcement Learning

AuthorsRodríguez Soto, Manel; López-Sánchez, Maite ; Rodríguez-Aguilar, Juan Antonio CSIC ORCID
KeywordsValue Alignment
Moral Decision Making
Multi-Objective Reinforcement Learning
Issue DateMay-2021
CitationManel Rodríguez Soto, Maite López-Sánchez, & Juan A. Rodríguez-Aguilar (2021). Guaranteeing the Learning of Ethical Behaviour through Multi-Objective Reinforcement Learning. . Adaptive and Learning Agents Workshop at AAMAS 2021 (ALA 2021).
AbstractAI research is being challenged with ensuring that autonomous agents behave ethically, namely in alignment with moral values. A common approach, founded on the exploitation of Reinforcement Learning techniques, is to design environments that incentivise agents to learn an ethical behaviour. However, to the best of our knowledge, current approaches do not offer theoretical guarantees that an agent will learn an ethical behaviour. Here, we advance along this direction by proposing a novel way of designing environments wherein it is formally guaranteed that an agent learns to behave ethically while pursuing its individual objective. Our theoretical results develop within the formal framework of MultiObjective Reinforcement Learning to ease the handling of an agent’s individual and ethical objectives. As a further contribution, we leverage on our theoretical results to introduce an algorithm that automates the design of ethical environments.
Appears in Collections:(IIIA) Comunicaciones congresos

Show full item record
Review this work

Google ScholarTM


WARNING: Items in Digital.CSIC are protected by copyright, with all rights reserved, unless otherwise indicated.