Analyzing Approaches to the Problem of Avoiding Side Effects in Autonomous Agents
The world and the Internet are more and more populated by artificial autonomous agents carrying out tasks on our behalf. Many of these agents nowadays are provided with an objective and they learn their behavior trying to achieve their objective as better as they could. This approach has allowed the development of very efficient agents in many environments. However, this approach can not guarantee that an agent, while learning its behavior, will not undertake actions that may have unforeseen negative side effects. Research into this side-effect problem tries to design autonomous agent that are prevented from taking undesirable actions while learning or when deployed.
The aim of this project is to study, implement and evaluate solutions for the side-effect problems. This would require to develop a solid understanding of the reinforcement learning paradigm and its limitations; to explore the current state of the art of this problem; to implement solutions within the OpenGym or GridWorld framework and compare the results with the community.