Abstract
Artificial intelligences (AIs) are becoming increasingly complex and impressive. In some cases they are now surpassing human abilities at cognitive tasks. These AI systems learn and act independently, and many now generate their own content. It is an accumulation of these kinds of characteristics that call into question whether human agents should be responsible for the actions of AIs. This thesis considers a set of questions about moral responsibility and artificial intelligences. The aim here is twofold: to clarify how to attribute responsibility for the actions of artificial intelligences, and to then use this to help guide policy and regulation. I argue that whilst in certain circumstances human agents such as developers and deployers will be responsible for harms and benefits caused by AIs this will not always be the case. I contend that the intuition that only human agents can be morally responsible must shift. Employing a functionalist argument I maintain that certain kinds of AI may develop the capacity for moral agency and moral responsibility. I go on to argue that there will also remain a set of cases where neither human nor artificial agent can be responsible for the actions of AIs. In these kinds of cases the harms and benefits that arise should be viewed as an accident. Utilising this assessment of moral responsibility and AI, this thesis culminates in a set of policy and regulation suggestions that aim to not overburden non-responsible agents, reduce the likelihood of harms happening, and to protect victims if they do.