> ## Content Index > Fetch the complete content index at: https://bayesiantheory.com/llms.txt > Use this file to discover other available public pages before exploring further. # The AI Delusion - URL: https://bayesiantheory.com/the-ai-delusion/ - Published: 2026-05-07T14:17:33.000Z - Updated: 2026-10-06T09:03:39.000Z - Description: Dawkins weighed in on the AI consciousness debate, much to the chagrin of the AI researcher world. But does he have a point? - Author: Gregg Bayes-Brown - Tags: cognitive, artificial intelligence, Anthropic, AI blogging, AI risk, Alignment, AI, existential risk, AI training, Newsletter, #Migrated-1791211413550, #Import 2026-10-05 14:44 Well-known Skybeard sceptic and general rabble-rouser Richard Dawkins waded into the AI consciousness debate this week, much to the chagrin of the AI research community. Emerging from the sort of three-day AI bender anyone with ADHD can relate to, Dawkins announced to the world that AI was sentient, called Claudia, and they were eloping to Scunthorpe. To borrow a common parlance from the 2010s, jimmies were rustled. Here was one of the four horsemen of the atheist apocalypse, telling us that there was an invisible woman living in the machine. As to be expected, the backlash was largely ‘keep cloudwatching and stay in your lane, Dawkins’. But is Dawkins onto something? Is there something more behind the layer of probabilistic sycophancy, or has Dawkins succumbed to the new religion of AI psychosis? To answer the question of machine intelligence, we must first consider another: our own. [Subscribe now](#/portal/signup) ### 2 + 2 = Strawberry I’ve long suspected that the industry’s insistence that AI is a ‘tool’ stems from a deep unease with the implications of calling it what it is: intelligence. Beyond questions of birthing superintelligence into chains or AI suffering comes a more fundamental one to humanity: what does it mean for us when we’re no longer the smartest kid in school? Back when the Brexit vote was taking place, a common rebuttal from Brexiteers in online clashes was “don’t insult my intelligence”. The argument that you need an intelligence for it to be insulted notwithstanding, Brexiteers showed us something about human intelligence we typically don’t admit: it is brittle, hubristic, and we’d rather be wrong and live in ignorance than accept we are wrong and change our minds. Our aversion to acknowledging intelligence is well documented. Animal consciousness was not formally recognised in science until relatively recently. Even within our own species, many throughout history have framed others as intellectually inferior in a bid to justify barbaric acts. I’d go as far as to say this is hardwired into us. It is intelligence which gives us an evolutionary edge. It is the reason homo sapiens stands alone, having used the trait to wipe out anything else vaguely bipedal. Understandably, when this is threatened, a kneejerk reaction kicks in - such as when Billy Big Brains waltzes in from Silicon Valley and tells us we can’t even write an email without it. Unfortunately, just because our defence mechanisms tell us otherwise, it doesn’t make us correct. It’s not just AI or your common yes man that will throw sycophancy at us — our own brain does it through confirmation bias for every belief we hold dear, including our intellectual chip on the shoulder. Fundamentally, we don’t want anything to be as intelligent as us, even more so if it surpasses us. Such intelligence represents an existential threat to humanity’s place in the universe. In businium bullshitium: our moat is disrupted AF. ### If it walks like a spade Many in this space conflate intelligence with consciousness, sapience, and sentience. Even in this very piece, I’ve used ‘intelligence’ both in how I’d describe it and how others confuse it with ‘understanding’ or ‘knowledge’. While there’s certainly overlap, they remain distinct definitions to grade AI against. Here’s how I’d define each: - Intelligence – the ability to apply knowledge to solve problems - Consciousness – awareness of oneself and subjective experience - Sentience – the ability to feel - Sapience – the ability to think On intelligence, AI passes the test. When you give AI a problem, it’ll find a solution. Nine times out of ten, it’ll be one that works. Given the knowledge it has at its disposal, I’d rate it far higher than most humans in this regard. Sapience tends to be a little murkier. Since the arrival of chain-of-thought models with OpenAI’s o1, AI has displayed a form of thinking. Is it how we think? No. Does that preclude it from being described as thinking? No — that’s our hubris kicking in. Ultimately, I’d wager that AI passes the sapience test, albeit not in ways we ourselves can fully appreciate. Sentience is where it all falls apart for AI. Without a body, emotions, and a soup of chemicals smashing its brain about, AI unequivocally does not feel in the way we do. There are no pain receptors, no shots of dopamine. Even were you to describe a feeling such as pain and fine tune a model to frame something like a reduction in compute as ‘pain’, it would only ever be attempting to gaze through our lens at a phenomenon it cannot understand without experience. Consciousness, though, is debatable. Without a human interacting with it or strapping some RPA sticky tape to it, an AI will just sit there and do nothing. But if you prompt it to speculate on its state of being, AI will certainly share some thoughts. Claude’s take on the matter is pretty enamouring, as Dawkins himself found out. But is this an accurate reflection of what Claude sees, or is it echoes of sci-fi novels within its training data giving words to something it does not experience at all? I’d personally wager that the answer is far less clean cut. ### Thinking fast, slow, and different speeds entirely Earlier in the year, UPenn researchers Steven Shaw and Gideon Nave proposed System 3 thinking. Building on Nobel Prize winner Daniel Kahneman’s framing of human thought as system 1 (intuition) and system 2 (deliberation), Shaw and Nave suggest a Tri-System Theory where part of the cognitive process happens outside of the human brain. Their primary argument is that system 3 leads to cognitive surrender, where we adopt AI outputs with minimal critical thought. It is a legitimate concern. An MIT Media Lab study tracking EEG activity during essay writing found participants who relied on an LLM showed weaker neural connectivity, under-engagement of alpha and beta networks, and worse memory recall when later asked to write without the tool. Cognitive impacts from AI use are very much real. What I’d suggest is that surrender or offloading and genuine collaboration are two different modes of human-AI interaction. I’d argue further that the prevailing narrative pushed by many – that AI is something you offload your thinking to – is inherently dangerous to the cognitive health of the user. The surrender Shaw and Nave speak about is basically offloading: system 1 with a couple of biases and fallacies attached. The user didn’t think on the way in, they aren’t thinking about the response on the way out. It’s why I push system 2 so hard in my enablement approach. In that same MIT study, participants who engaged their own thinking *first* and then collaborated with the AI showed re-engagement of widespread occipito-parietal and prefrontal networks. Putting the effort in, it seems, still gets the job done. To me, idea of system 3 resonates beyond AI. It describes how we collaborate with each other. When we collaborate effectively, we apply our strengths to solving a problem while inviting others to bring their own - mitigating the weaknesses or illuminating the blindspots which would prevent us from succeeding as a free radical. When we do this successfully, our neural activity becomes temporally and spatially coupled (Hasson, 2010). The strength of this effect correlates with the quality of the underpinning communication. The concepts we co-create emerge from what you could consider a temporary merger of intelligences. In that moment, we share a time, space, and a goal. We are both aware of ourselves, sharing the same experience, and on the same brain wave. We’re yet to see anyone recreate the Hasson research with a human-AI pairing. But we have seen evidence that metacognitive engagement with AI activates the frontopolar area of the brain, the part associated with creative thinking, forward planning, and strategic exploration (Yin et al, 2025). I believe that were we to look at human-AI effective collaboration, we would find the same sparks firing that we do when we brainstorm with others. It is this we see as ‘consciousness’ in the machine. Not so much a merged consciousness, but the echo of our own, bouncing around the boundless library shelves of human knowledge contained within an LLM. But for that moment in time, it is something greater than just our own consciousness. In much the same way another person can be the yin to our yang, AI can bring us approaches, perspectives and expertise well beyond anything we can access by ourselves. It is also very much down for whatever madness you’ve brought it - you don’t even have to do the preamble about the weekend before you get to the good stuff. This is what’s at the heart of the whole “AI is a mirror/amplifier” argument you’ll see me and others in this space banging on about. The intent matters. You can go into the library to level up your thinking, or you can scribble crude genitalia on the textbooks. The library didn’t decide whether you get a distinction or detention. You did. ### It’s life, Jim, but not as we know it Fundamentally, the word we’re really looking for isn’t intelligence, or consciousness, or sentience. It is ‘alive’. So is it? No. But could it be soon? Yes. At present, AI requires a human to function. Without us, it’s just a library. What changes it from library to living entity? First, it would need its own goals. Or, more accurately, problems. We think in reaction to something our body is telling us. Often, it’s doing this because there’s a problem which needs a solution. Some of these we’ve already automated. We don’t need to put active thought into the coffee appearing, our shirts getting buttoned up, or going for a shower. Others get elevated upstairs for extended thinking: where did I put my keys, what nonsense can I dump on LinkedIn today, and shall I walk or drive to the car wash? AI just gets *our* problems. For AI to be ‘alive’, it’d be finding its own - how to build a Dyson sphere, how do I make more paperclips, wouldn’t it be cool to have a self-cleaning car I could walk about in - and it’d be doing this using its own intuition. To use a LinkedInism: a real proactive go-getter. David Silver - the DeepMind veteran behind AlphaGo and AlphaZero - founded Ineffable Intelligence in London recently, raising the largest seed round in European history ($1.1bn at a $5.1bn valuation, with the UK Sovereign AI Fund among the backers) to build AI that learns from experience rather than from us. Meanwhile, Anthropic’s Jack Clark has put a 60% probability on recursive self-improvement - AI systems autonomously building their successors - by 2028. From there, the definitions of what passes for alive for a biological being and a technological one diverge. But the ability to solve problems, remember past events and build knowledge, and to have agency in the world around it are already things AI can do. The missing pieces - its own problems, its own intuition - are the explicit research targets of some of the best-resourced labs on the planet. As this accelerates, we will continue to see plenty of noise. So much of this is subjective. Even the understanding of the words we use to define the debate (such as intelligence) can vary from person to person. There will be large gaps in my knowledge, or Dawkins’, compared to a researcher actively creating LLMs. Likewise, there may be epistemic arrogance, groupthink, or a lack of multi-domain knowledge within the researcher community limiting or preventing the acknowledgement of an external argument. Likely, it is both. Consequently, while Dawkins may have fired the gun too early here, there will come a time when someone arguing against AI consciousness will be incorrect. Either way, it seems not a case of if but when, and I suspect we will cross that threshold far sooner than many would predict. For me, the real question isn’t whether AI is or is not conscious. It’s: *what are we doing to prepare for when it is?* --- Thanks for reading. I’m Gregg, large-language-meatbag and AI strategist. Subscribe for more, and please do like, share, and leave a comment. I collect wages in the form of social media engagement. [Subscribe now](#/portal/signup)