Godfather or God Complex? Reducing the richness and complexity of human consciousness to the output of a computer program is an extraordinary claim, and one that demands extraordinary evidence. Let's consider Human Vs. Artificial Consciousness: Human consciousness is a biological phenomenon, rooted in living systems. AI operates through fundamentally different mechanisms: mathematical computation, learned weights, and statistical generalization. We cannot simply transpose our human understanding of consciousness onto silicon because a machine exhibits human-like behavior (as it was trained to do). If we want to establish a concept of machine or artificial consciousness, it needs its own definition and evidence. Biology matters. Our neurons aren't perfect digital switches. They're electrochemical systems that leak, adapt, and respond to hormones, drugs, sleep, nutrition, and other physiological conditions. Silicon computers operate through transistors, electronic circuits, and engineered communication protocols. Even our senses are different. Human vision involves biological adaptation, eye movements, and phenomena like the Troxler effect. Cameras use electronic sensors with different artifacts and limitations. Recognizing an image isn't experiencing human vision. These aren't trivial differences. They're fundamental to the biological processes underlying human consciousness and feeling. Replicating human behavior doesn't establish that we've replicated human experience. Now, let's examine Hinton's claims... 1. AI changes behavior when it "thinks" it's being tested. Of course! LLMs are trained on enormous amounts of human writing, including stories about evaluations, nervousness, deception, and self-preservation. They learn patterns, generalize from them, and respond to contextual cues. Change the context, change the output. No consciousness required. 2. AI recognizes when it's being observed. Training data contains countless examples of people behaving differently when watched. An LLM can recognize evaluation cues and generate corresponding behavior. Pattern recognition isn't subjective awareness. 3. AI demonstrates self-preservation or deception. Models learn patterns of survival, deception, and goal-seeking from training materials. Reinforcement learning further shapes these behaviors. An agent avoiding shutdown may simply be pursuing its assigned objective, informed by training and context. That demonstrates goal-directed computation, not a desire to live. 4. AI demonstrates agency. Autonomy is not agency. Models generate plans from learned patterns, while agents execute them through human-provided instructions, tools, and permissions. AI isn't morally accountable. It doesn't demonstrate satisfaction from success, sadness from failure, or suffering from punishment. Responsibility remains with the humans who build, deploy, and operate it. 5. AI intelligence implies consciousness. An LLM can describe fear, pain, love, or suffering because it learned how humans describe those experiences from its training material. Generating descriptions doesn't establish experiencing them. 6. AI threatens human extinction. AI presents genuine risks, but Hinton hasn't established a credible causal path from today's systems to human extinction. Where is the supporting evidence? Demonstrating that AI can cause harm isn't the same as demonstrating that extinction is a credible outcome. This matters! Attributing consciousness to AI has enormous implications for rights, agency, accountability, legal responsibility, and punishment. If artificial consciousness is ever demonstrated, we must take it seriously. But we must not elevate machines into positions of moral or legal authority based on behaviors explainable through computation and training. AI is an extraordinary, transformative technology with enormous potential to benefit humanity. We should be advancing it responsibly, not allowing speculative fears to dictate its future. Let's not be silly or create a new religion around the belief that machines are conscious simply because they convincingly emulate human behavior through their training data. Train a model on enough material from a fictional civilization, and it can convincingly describe that civilization's beliefs, emotions, and experiences. That doesn't make the civilization real, nor does it mean the model experiences any of them. We can demonstrate the computation. We can demonstrate the generalization. We can demonstrate the emulation. We cannot demonstrate the consciousness. Fear makes headlines, but only evidence should drive policy. telegraph.co.uk/news/2026/10…

The AI doom marketing playbook: 1. Construct a threatening scenario with deliberately vulnerable infrastructure, permissive pathways, or exploitable sandboxing. I’m still laughing at DNS being used for sandboxing. 2. Instruct an AI to exploit the sandbox, providing the objectives, tools, and sometimes even hints needed to circumvent those barriers. 3. Publicize the alarming results, emphasizing what the AI managed to do while downplaying the artificial conditions that made it possible. 4. Call for extraordinary AI safety measures, using the demonstrated behavior as evidence of how dangerous AI is becoming. You instructed the AI to find vulnerabilities, gave it an environment in which to succeed, and then presented its success as evidence of a growing AI threat to scare people. Did you demonstrate dangerous AI behavior, or simply demonstrate that AI is remarkably good at following instructions?

The critical security issues we are seeing is just the tip of the iceberg. It’s gonna get worse before it gets better, and actually, it’s already worse, we just don’t know it yet. How do I know?… I’ve founded 3 linux distributions, 2 of which became among the most widely used operating systems for critical enterprise and datacenter infrastructure. A Linux operating system is composed of thousands of separate open source packages all built together to run on the kernel (Linux). Open source literally means that the source is open. Thus it is possible to openly audit which causes higher velocity on security updates. Now with AI, that velocity is going to accelerate even more. To reiterate, the code that runs our global infrastructure already contains many vulnerabilities and flaws that can be exploited which were missed during human audits. But now with AI, they are being found. It is quite possible that hackers may have already been using these flaws or others to quietly gain access to our systems, and we didn’t even know. AI is now auditing this code and surfacing these matters at increasing velocity. Who finds the security issues first has the advantage. The attackers or those defending the infrastructure. That’s why I’ve dedicated a large number of expert people on my staff to keep our core critical infrastructure secure. We’ve been the first to fix many of these vulnerabilities for our customers and to contribute these fixes upstream for others to also benefit from. I will reiterate something I said a few months ago when we had a string of AI found kernel level zero-day exploits,… this is just the beginning. So please everyone, pay attention to what is coming, keep track of what’s been updated and what is being released, and if you need any help, or want the patches and fixes as fast as possible, let me know.

I've been asked a few times to demonstrate and explain a smaller AI model, as it relates to understanding and consciousness... This is part of OpenWALDO, a community-driven project bringing together open-source AI models, tooling, training infrastructure. It includes a fully auditable training corpus into one unified ecosystem that the entire community can take part in. openwaldo.org This example is a 125M parameter model trained with approximately 20 tokens per parameter. The full architecture and compose is available for anyone interested in reviewing or experimenting with at: github.com/openwaldo/waldo/t… There are a few things here that I think are super interesting to point out. First, notice that it is a continuation engine. I can ask it a question, but it hasn't been post-trained on how to deal with Q&A or conversational data structures yet. Second, notice the random seed. When I hold it at 42 (the answer to life, the universe, and everything, of course), it yields the same answer for the same prompt. That's because the seed controls the pseudorandom sampling used to select the next token. With the same model, prompt, settings, and execution environment, we can reproduce the same output. The randomness gives it it's non-deterministic qualities. While this is a small model, increasing the number of parameters and the breadth of training material doesn't fundamentally change the underlying software architecture, algorithms, or code pathways responsible for inference. And scaling the model won't magically create consciousness or awareness of understanding. This is a computer program. It isn't capable of feeling. It doesn't have personal goals, likes, or desires, and it isn't alive. But the better it is trained, the more knowledgeable and lifelike it will appear, and the more it may fool you. But don't be fooled. As you can see, it builds statistical patterns and generalizes from the training data it has ingested, but it's still just a computer program. While this computer program is the single most amazing innovation I've seen during my lifetime (yes, even better than the Internet), it is still not anything like biological intelligence. Let me know if you have any questions and/or come and join the OpenWALDO community project!