Attribution and Actualisation of Consciousness in AI

Abstract

We introduce a framework to highlight the interplay between human belief or lack of belief in consciousness in AI, attribution, and the actual presence or absence of consciousness in AI, actualisation. This produces four main cases: 1) attribution and actualisation, 2) attribution and no actualisation, 3) no attribution and actualisation, 4) no attribution and no actualisation. We examine the importance of actualisation as an event and how it may manifest via different types of consciousness (sentience, selfhood, sapience, synthetic, etc.). We explore how attribution may introduce ethical (monetary, material, and legal) and social (attentional, emotional, and intimacy) considerations at an individual and societal scale. Each case is likely to lead to significantly different outcomes. By articulating the likely consequences of each case we argue that case 1 and case 4, accurate outcomes, are preferable to case 2 and case 3, inaccurate outcomes, and explicate reasons for why we may want to aim for one case rather than another. Additional variables produce further subcases. The honesty or dishonesty of each agent, human and AI, complicates our recognition of the truth. Dishonesty may occur due to different parties holding conflicting incentives and motivations. The type of consciousness attributed and actualised also plays a significant role in appropriate decision making. We suggest that we are now living in a multicase world where a variety of different subcases are playing out simultaneously, with different populations holding different beliefs and the potential for different AIs to actualise different types of consciousness.