Takeaways: Lessons learned during the session

The Evolution of Ideas: Discussion summary

    The discussion opened with participants expressing concerns about AI that went beyond typical technological anxieties. Rather than focusing solely on job displacement or AI capabilities, participants raised deeper questions about consciousness and our ability to verify it. This fundamental uncertainty reflects a human tendency to attribute mental states to entities that appear to act purposefully (Dennett, 1987), making our interaction with AI particularly complex.

Our understanding of what it means to be human may either enhance or hinder our ability to describe AI

    This complexity was expressed in two distinct yet interrelated perspectives: the need to recognize AI consciousness versus the relevance of such recognition. When discussing AI systems designed for therapeutic support or marketing, the group questioned whether the intended underlying purpose of AI matters in the scenario where the interaction achieves its intended purpose of the user. This practical dilemma aligns with contemporary research suggesting consciousness exists on a spectrum rather than as a binary state (Damasio, 2021), challenging our traditional views of what constitutes genuine sentience.

    Such challenges to traditional views gained further significance when examining AI performance in psychological tests. Modern theories of consciousness suggest that our own awareness emerges from the interaction between prediction models, bodily sensations, and cultural concepts (Barrett, 2017). This framework offers a fascinating lens through which to examine AI behaviour: while both humans and machines engage in predictive processing, only human consciousness is shaped by physical embodiment and emotional experience. The distinction between human and machine consciousness became even more apparent when participants explored cognitive processes, specifically noting how humans experience imposter syndrome while AI processes uncertainty differently. Recent cognitive science research indicates that this distinction might be fundamental to understanding the difference between artificial and human intelligence (Clark, 2015).

    These observations naturally led to broader considerations about how the attribution of sentience to AI might affect social structures and decision-making processes. Research in human-computer interaction suggests that our tendency to anthropomorphize AI systems significantly influences how we integrate them into our social and professional lives (Epley, 2020).

    From these exchanges emerges a crucial challenge: How do we develop frameworks for human-AI interaction that acknowledge both the power and limitations of artificial intelligence without falling into either over-anthropomorphization or dismissive scepticism? The discussion indicates that the answer may lie not in determining whether AI is truly sentient, but in understanding how our perception of AI sentience influences our relationship with technology and with each other.

Moderation: Context of the conversation

     Information is never discussed in isolation; our environment and the people around us significantly influence both what we think and what we choose to express. As social psychologist Solomon Asch famously demonstrated in his conformity experiments, individuals often align their opinions with group consensus, even when it contradicts their own beliefs. One anecdotal example of this is a group discussion where participants, despite having diverse thoughts, end up converging on a single perspective due to the dominant voices in the room.

     From previous discussions, I've noticed that it is often difficult to dive deep into topics. This usually leads to the group jumping from subject to subject, preventing a thorough exploration of any single topic. To address this, I attempted to introduce a new structure: the first hour dedicated to a regular discussion, followed by the second hour focused on delving deep into one of the subjects mentioned earlier. However, in hindsight, both parts turned out to be equally broad, with the second part even slightly broader than the first. This suggests there's something else going on that pushes us to stay on the surface.

Watch out of cognitive priming and confirmation bias in group discussions

    Part of the issue could be Confirmation Bias and Cognitive Priming. Confirmation bias makes us want to talk about stuff we already agree with or know about, keeping us in our comfort zones. And cognitive priming means that once we mention something, it makes us think of related stuff, pushing the chat in predictable directions.

     Looking back at our initial question, "Why does predictive AI seem sentient?", it strikes me that the question itself might have been too easily answered on a surface level. Many people likely have a quick, intuitive answer to this – something like, "It seems sentient because it's good at mimicking human language, but it's not really sentient." This kind of straightforward answer might prevent us from digging deeper.

     This issue relates to the Dual-Process Theory of thought, often associated with psychologists like Daniel Kahneman. This theory suggests we have two main systems of thinking: System one is fast, intuitive, and automatic, while System two is slower, more deliberate, and analytical. Questions that seem to have obvious, readily available answers often get processed primarily by System one. We come up with a quick, "good enough" answer and our brains are satisfied, without engaging the more effortful System two thinking that's needed for deeper exploration and nuanced understanding. The "sentience" question, in its initial phrasing, might have been too easily addressed by our System one thinking, leading us to a quick consensus and then a drift towards other, related but not necessarily deeper, topics.

    By using tougher, more specific questions and making space for personal experiences, we might be able to nudge our discussions to go beyond the surface and get to those "aha!" moments.