This article introduces definitions for direct, means-end, oblique (or indirect) and ulterior intent which can be used to test for intent in an algorithmic actor. These definitions of intent are informed by legal theory from common law jurisdictions. Certain crimes exist where the harm caused is dependent on the reason it was done so. Here the actus reus or performative element of the crime is dependent on the mental state or mens rea of the actor. The ability to prosecute these (...) crimes is dependent on the ability to identify and diagnose intentional states in the accused. A certain class of auto didactic algorithmic actor can be given broad objectives without being told how to meet them. Without a definition of intent, they cannot be told not to engage in certain law breaking behaviour nor can they ever be identified as having done it. This ambiguity is neither positive for the owner of the algorithm or for society. The problem exists over and above more familiar debates concerning the eligibility of algorithms for culpability judgements that mens rea is usually associated with. Aside from inchoate offences, many economic crimes with elements of fraud or deceit fall into this category of crime. Algorithms operate in areas where these crimes could be plausibly undertaken depending on whether the intent existed in the algorithm or not. (shrink)

The dominant practice of AI alignment assumes (1) that preferences are an adequate representation of human values, (2) that human rationality can be understood in terms of maximizing the satisfaction of preferences, and (3) that AI systems should be aligned with the preferences of one or more humans to ensure that they behave safely and in accordance with our values. Whether implicitly followed or explicitly endorsed, these commitments constitute what we term apreferentistapproach to AI alignment. In this paper, we characterize (...) and challenge the preferentist approach, describing conceptual and technical alternatives that are ripe for further research. We first survey the limits of rational choice theory as a descriptive model, explaining how preferences fail to capture the thick semantic content of human values, and how utility representations neglect the possible incommensurability of those values. We then critique the normativity of expected utility theory (EUT) for humans and AI, drawing upon arguments showing how rational agents need not comply with EUT, while highlighting how EUT is silent on which preferences are normatively acceptable. Finally, we argue that these limitations motivate a reframing of the targets of AI alignment: Instead of alignment with the preferences of a human user, developer, or humanity-writ-large, AI systems should be aligned with normative standards appropriate to their social roles, such as the role of a general-purpose assistant. Furthermore, these standards should be negotiated and agreed upon by all relevant stakeholders. On this alternative conception of alignment, a multiplicity of AI systems will be able to serve diverse ends, aligned with normative standards that promote mutual benefit and limit harm despite our plural and divergent values. (shrink)

Do people hold robots responsible for their actions? While Clark and Fischer present a useful framework for interpreting social robots, we argue that they fail to account for people's willingness to assign responsibility to robots in certain contexts, such as when a robot performs actions not predictable by its user or programmer.