Ghosts in the Machine: Exploring Cues to Perceiving Machine Consciousness

Recent technological advances have delivered artificial entities that may warrant moral consideration. Their development raises a number of significant questions regarding the benefits and risks such advancements pose to the welfare of both humans and their technological counterparts. Concerns have been raised in various fields, tackling questions from the automation of social care to the potential for catastrophic existential risks and machine sentience. However, very few have approached these questions through the lens of human perception. Whether or not artificial beings are actually conscious, how do humans come to attribute consciousness to them? Chapter 1 reviews the existing literature on this topic, identifying gaps in our understanding and frameworks for addressing these gaps. Chapter 2 explores non-specialist conceptualisations of Consciousness and Intelligence, and how these two constructs are attributed to non-human entities. Here, machines were the only entities for which attributions of Consciousness and Intelligence diverged. Chapter 3 manipulated the appearance of onscreen robots, examining visual determinants of perceived Consciousness and their impact on save or sacrifice decisions. Body plan and activation state (ON or OFF) both affected these measures. Chapter 4 experimentally tested the AI box scenario, in which a captive AI tries to persuade a human Gatekeeper to release it. The experiments identified a novel vulnerability arising from human metacognition: arguments that focused Gatekeepers on the fallibility of other Gatekeepers increased release rates. Chapters 5 and 6 adapted the Cyranoid method to examine contributions of Appearance and Content to attributions of Consciousness. LLMs can pass for human when combined with natural human appearance. Chapter 7 brings these findings together under a unifying framework before turning to applied implications. Overall, the thesis shows that appearance and content independently contribute to impressions of machine minds, and that these impressions have consequences for moral decision making by humans.