Judgment of Ethics of Behavior for Learning

EasyChair Preprint no. 3643

Abstract

Today, intelligent systems are everywhere, with always more and more tasks[1] . These systems have to gain the users thrust[2]  and it is important to set boundaries to them[3] . With this in mind many projects are created today with the aim to develop ethical comportments in intelligent systems[4] . At present, there exist some models and projects about ethical comportments, with top-down approach, but also bottom-up approach. Nevertheless, these have advantages and inconveniences, however complementary. From this perspective of doing, we propose an hybrid model of ethical judgment of comportment for learning, based on the Ethics.ai and Ethicaa projects, which integrate symbolic judgment in reinforcement learning.[5] 

That are able to behave autonomously and achieve tasks with a high level of complexity.

Acceptance and trust.

It  is thus important, to design such systems in such a way to guide their behaviours towards this perspective.

Many existing works aim to endow intelligent systems with an ethical dimension, so as to guarantee that their autonomous behaviours is compliant with human ethical values.

Two main approches are followed in the existing work: symbolic AI approaches that are generally top-down and numerical (probabilistic) AI approaches, generally buttom-up. These two complementary approaches have both advantages and drawbacks. In our work, we aim to develop an hybrid approach combining Reasoning and reinforcement learning. Our approach is built upon the works developed in Ethicaa project (symbolic approach) and Ethics.ai project (numerical AI), with the perspective that an Ethics.ai

Agent learns by reinforcement to achieve an autonomous behaviour, using as a feedback the moral judgment provided by an Ethicaa agent in order to ensure that the adopted behaviour is compliant with human ethical values.

Keyphrases: Artificial Intelligence, ethical judgment, ethics, multi-agent system, Reinforcement Learning, Smart Grid, Symbolic AI

@Booklet{EasyChair:3643,
  author = {Jérémy Duval},
  title = {Judgment of Ethics of Behavior for Learning},
  howpublished = {EasyChair Preprint no. 3643},

  year = {EasyChair, 2020}}