Helena Ward

Conversational AI agents designed to simulate interactions with deceased persons—referred to here as Interactive Personality Constructs of the Dead (IPCDs)—offer a novel and potentially highly valuable way to memorialise the deceased. But IPCDs could also undermine the goal of remembrance. This paper argues that IPCDs put users at risk of mnemonic distortion—an epistemic harm in which generated misrepresentations of the deceased become incorporated into our memory of the de-ceased, such that some of what subsequently shapes our conception of the deceased person originates in the IPCD rather than the deceased—reshaping or, at worst, replacing our memories of the person we hoped to remember. Such distortions are importantly distinct from imaginative engagement with the deceased or the testimony of others, which may also alter our memories of the deceased. The risk of mnemonic distortion is not inherent to IPCDs but rather contingent on design; the paper concludes by considering how this concern could inform recommendations for the design of IPCDs.

This thesis is concerned with our interests in self-presentation: in having a measure of control over which aspects of our selves – our bodies, thoughts, how we style our appearances – we allow others to observe, and on what terms. I consider our self-presentational interests in relation to our right to privacy, privacy being one of the main ways in which we retain self-presentational control. And in light of technologies of the self, which have transformed the ways in which we construct our own identities and observe other people’s. The first aim of this thesis is to vindicate a distinction between the synchronic and diachronic aspects of self-presentation. Sociological and ethical accounts of self-presentation adopt a Synchronic Approach, I argue that this approach is incomplete – our self-presentational interests also have a diachronic character. In short, it is not only important that we are able to exercise a measure of control over how we present ourselves to others at any one time, but also that we’re able to exercise a measure of control over other people's access to our past self-presentations. These are our diachronic self-presentational interests (DSPIs). The second aim of this thesis is to show how technologies of the self undermine our DSPIs. While these technologies do not single-handedly thwart our DSPIs, they undermine our diachronic self-presentational autonomy by introducing social customs and individual incentives to act in ways that are negligent, short-sighted, hostile, or in various other ways insufficiently attentive to our DSPI-related responsibilities. I argue that the nature and extent of this threat has been underestimated because of the implicitly Synchronic Approach. Our DSPIs provide us with some grounds for a right to diachronic privacy – in keeping our childhood foibles, jejune opinions, and bygone bodies to ourselves.

Those who worry about a superintelligent AI destroying humanity often appeal to the instrumental convergence thesis—the claim that even if we don’t know what a superintelligence’s ultimate goals will be, we can expect it to pursue various instrumental goals which are useful for achieving most ends. In this paper, we argue that one of these proposed goals is mistaken. We argue that instrumental goal preservation—the claim that a rational agent will tend to preserve its goals—is false on the basis of the timing problem: an agent which abandons or otherwise changes its goal does not thereby fail to take a required means for achieving a goal it has. Our argument draws on the distinction between means-rationality (adopting suitable means to achieve an end) and ends-rationality (choosing one’s ends based on reasons). Because proponents of the instrumental convergence thesis are concerned with means-rationality, we argue, they cannot avoid the timing problem. After defending our argument against several objections, we conclude by considering the implications our argument has for the rest of the instrumental convergence thesis and for AI safety more generally.