Rebecca Raper

This chapter defines key terms involved in the Machine Ethics debate—including, but not limited to, the fundamental terms ‘morals’ and ‘machines’—so that we can begin to understand what we are trying to achieve when we refer to building ‘moral machines’. It begins with an introduction to the philosophy and psychology of morality, exploring key philosophical ethical theories such as Virtue Ethics and Utilitarianism, then looking at Kohlberg’s theory of moral development. The second part of the chapter defines key technical terms, including ‘machine’, ‘robot’ and ‘Artificial Intelligence’, and gives a brief introduction to Machine Learning.

With the reframing of the problem, as given by the case studies in the previous chapter, and with the Moral Assurance perspective, this chapter begins by advancing a new definition for Machine Ethics—one focused on enabling moral agency, as opposed to constraining moral behaviour. Developing the idea that to enable moral agency we need to at first trust the machine, an approach is outlined that involves eliciting cognitive requirements for morality in a machine, based on The Waterfall Method often used in the development of IT systems. This approach is also used to prescribe a new test procedure for moral agency—one that focuses on assuring certain cognitive features, rather than simply identifying expected behaviours.

This chapter introduces, and gives an overview of, the field of Machine Ethics. Assuming no prior knowledge on the topic by the readers, this chapter can be taken as a standalone chapter for those just interested in learning more about this area. The chapter begins by drawing the distinction between ethical machines and ethical decision machines before going on to examine some earlier attempts to create machines with the ability to tell between right and wrong. After this, some arguments against building moral machines are discussed, before giving an overview of tests for morality—how do we know when we have successfully built a moral machine? The chapter ends with an overview of AI Alignment, by discussing similarities and differences in this pursuit and Machine Ethics.

We have a prescriptive framework for moral agency in machines—the cognitive moral requirements specification—the aim of this chapter is to look at following this new methodology to develop a list of cognitive needs that the new system must have, including a test specification to assure that the new system meets these needs. The moral requirements are elicited by tracing through the three previously mentioned ethical theories: consequentialism, Deontological Ethics and Virtue Ethics to determine what cognitive features moral decision-making typically exhibits. Then, test scenarios and some methods for testing cognitive abilities are introduced.

With a requirements specification from the previous chapter, and test scenarios, this chapter begins to develop a more technical model for morality, first by examining the relationships that are typically demonstrated in human moral developing agents—children. The child-caregiver relationship is examined to develop a model of morality based upon feedback a child might receive from their caregiver to develop their own moral perspective of the world. There is then an attempt to translate this model into a technical approach, akin to ‘raising robots to be good’, through a secondary, ‘meta’, feedback module that guides Reinforcement Learning systems. The chapter ends with an assessment of this model, using the test specification and cognitive requirements developed in the previous chapter.

This chapter provides the backdrop to the book. With an introduction to Asimov’s infamous laws of robotics, it provides a comprehensive dissection of the laws to demonstrate why they might be problematic in practical terms if we really want to create robots with morals. It then goes on to introduce the field of Machine Ethics, before tracing its way through a summary of the various chapters involved in the book. It ends by stating the key thesis behind the book—that if we want to create machines with morals, we need to raise robots to be good.

It’s important to consider the ethical implications associated with this new field of study. Though Machine Ethics considers the ethics relating to how a machine might make ethical decisions, the system in of itself, has ethical ramifications. This chapter aims to consider seven of the main ethical issues. First; whether artificial moral agents are safe is considered, by discussing some previous arguments against Machine Ethics and whether bestowing machines with morals puts humanity at even more risk. Then, Machine Patiency or Robot Rights is discussed—if robots have their own agency, should they have rights? Then, there is consideration about whether such robots are sustainable, before broader ethical implications (an ethical impact assessment for moral machines) are outlined. Finally, new security implications associated with the approach to building moral machines is discussed, before the idea of robot psychology is introduced.