Simon Kolker

Research Focus

I am a PhD student in the Department of Computer Science at the University of Manchester. Particularly, my project aims to study the effect of uncertainty on Machine Ethics. As autonomous systems take more responsibility in the real world, the consequences of their actions reach into every facet of our lives.

Autonomous systems are an increasingly prevalent feature of the modern world, gaining responsibility as drivers, medical practitioners and soldiers.

As their decisions affect humans, autonomous agents have an increasing ethical impact. The real world is a complex, unpredictable system, meaning agents must interpret and select actions with uncertainty amongst their outcomes. The real world is `open' to new situations unanticipated by the system's designers, meaning agents could inflict harm on humans.

Machine Ethics is the study of how ethical reasoning can be computationally implemented, but there has been little focus on ethical reasoning under uncertainty. This may be due to a gap in Moral Philosophy for uncertain situations. Historically, there is an assumption that Ethics decides the values and it is up to Decision Theory to maximise those values under uncertainty. This view has materialised in Machine Ethics implementations with expected utility maximisation in Reinforcement Learning, probabilistic planning.

Unfortunately, expected utility as the singular criterion for ethical decision making is prone to criticism. Additionally, I believe if Machine Ethics implementations are to be widely adopted, implementations must be aligned with their stakeholders' personal views on Ethics. An individual's views may draw from various Philosophical traditions (e.g. Deontology, Virtue Ethics, Ethics of Care etc.) and there may be uncertainty or conflict between these moral theories.

For any treatment of uncertainty in Machine Ethics, decisions ought to be made in a planning context: considering future decisions rather than selecting the immediately preferred action in every case. As the Automated Planning community demonstrates, such consideration is required in reaching the most desirable outcomes.

Roles

• Member of the Autonomy and Verification Network
• Publicity Chair for IEEE Space Mission Chalenges for Information Technology IEEE Space Computing Conference

Papers