Detection of Unethical Intelligent Agents in Ethical Distributed Constraint Satisfaction Problems

Abstract

Ethics has become the most interesting research field in artificial intelligence, it has been considered in several areas such as intelligent military applications, private data systems and autonomous systems (e.g. autonomous vehicle). Another artificial intelligence discipline needs to consider ethics in autonomous multi-agent systems. Designers of these autonomous systems create agents that decide, act and interact in dynamic environments under different constraints, where they may share or execute tasks with other agents and human beings. As a consequence, these intelligent agents gain increased autonomy and human supervision by users decreases. In Distributed Constraint Reasoning framework, the scope of the agents activities magnifies while solving mathematical problems and ensuring that such systems will not make irrelevant or even dangerous decisions is necessary. This paper shows an ethical control system which can be implemented into Distributed Constraint Satisfaction Problem (DisCSP) algorithms by several ways, and which is able to detect the abnormal activities and then the responsable unethical agents in order to regulate the resolution ethically. Experimental results show the feasibility of our contribution.