Date: 28 April 2017
Time: 11:00 - 12:00
Location: Wolfson Lecture Theatre, Computing, Queen Mother Building
Host: Dr Katarzyna Budzynska
Title: The Role of Logic based Argument and Dialogue in Addressing the Value Loading Problem
Abstract: Many of the recently reported successes of Artificial Intelligence have been founded on significant advances in machine learning. It is in this context that concerns have been raised about artificially intelligent, and in particular `superintelligent' machines; the central argument being that in pursuit of their goals such machines may inadvertently realise outcomes that are not aligned with human values, and that may therefore pose serious threats to human well being. In this talk I will summarise and illustrate this argument, and then review current approaches to addressing how one can avert such dangers by ensuring alignment of outcomes with human values; the so called `value loading problem' which is considered by prominent philosophers and AI researchers as amongst the most pressing of research challenges. However, I argue that current approaches, which primarily rely on machine learning techniques, will prove to be inadequate unless machines have the capacity to engage in symbolic joint reasoning with humans about moral and ethical issues. I will then briefly review some of my previous and current research aiming to facilitate such joint reasoning in the form of argumentation based dialogues that capture real-world modes of dialectical reasoning, are provably rational under resource bounds, and facilitate reasoning about values and preferences. I will conclude by pointing to future research directions that may fruitfully be pursued so that logic based argument and dialogue can contribute to solving the value loading problem.
Bio: Sanjay Modgil is a senior lecturer in the agents and intelligent systems group at King's College, and chair of the European Association of Multi agent Systems. His current research focuses on formal logic, argumentation theory and dialogue; in particular the ASPIC+ model of logic-based argumentation, metalevel argumentation, and extensions to argumentation systems to accommodate argumentation about preferences and values. He also works on applications of argumentation to agent reasoning and communication; in particular how rational conclusions are arrived at through the dialectical exchange of argument and counter-argument.