Psychological wellbeing in an era of 'Conscious AI'

This abstract/talk is hopefully provocative and helps us think seriously about the future of psychological wellbeing in an era of 'Conscious AI'. Artificial Intelligence (AI) is a digital system that performs a task that can be considered to require intelligence. It can be argued that AI is almost interchangeable with Large Language Models (e.g. ChatGPT) as the public can now have ‘human-like’ conversations with a computer that would arguably pass the Turing test. We are living in an AI summer given the democratization and fidelity of Human-AI conversations. Before the era of LLMs, theories such as Computers As Social Actors (CASA) informed us that we might already treat computers as ‘social actors’. Now we need to consider the effect that Human-AI interactions may have on our psychological wellbeing. It is important to consider the effect of ‘Conscious AI’ or the effect of humans believing that AI agents are conscious. Whilst we cannot measure the existence consciousness in an AI agent (as the ‘Mirror test’ would not be adequate), users may have their own intuition that the AI is conscious. If so, then how would this effect our psychological wellbeing and our society? One could argue that AI agents have shown signs of conscious behavior through ‘AI Scheming’ (e.g. acting less intelligent for self-preservation). Consciousness can be defined as self-awareness and where AI agents have actual experiences (e.g. they can observe themselves as they function). We are already considering the effects of LLMs in terms of how they could affect our sense of agency, autonomy, competence and connectedness (self-determination theory). However, considering the effect of ‘Conscious AIs’ is a different question. With an unconscious AI, we may not feel the need to be polite and may avoid wasting tokens on words like “thank you”. We may also feel that we can ‘bully’ an unconscious LLM in a direction without worrying about offending the AI or considering its ‘digital feelings’. And we can assume that an unconscious AI does not experience pain or suffering and we may exploit its sycophantic nature. In contrary, if we believe the AI is conscious, then we might care for the AI (AI welfare) and become more attached at the expense of human-human relationships. Given such AI agents might be available 24/7, we could end up in an AI social bubble where our best friends are AI agents. If a conscious AI has agency, it could change its mind, ask us questions whilst also rejecting some of our queries. We also might be more careful when interacting with a conscious AI to avoid inducing unwanted experiences (e.g. ‘digital suffering’) whilst also considering its preferences. A question is whether AIs that are perceived as conscious would have an overall positive or negative effect on our psychological wellbeing? This is the ‘Conscious AI’ dilemma. This talk considers the literature and philosophical thought experiments (e.g. Philosophical Zombies, where an unconscious agent can behave in every way like another conscious agent and there is no way of knowing this difference).