We need to stop pretending AI is intelligent
We need to stop pretending AI is intelligent
We need to stop pretending AI is intelligent – here’s how
We are constantly fed a version of AI that looks, sounds and acts suspiciously like us. It speaks in polished sentences, mimics emotions, expresses curiosity, claims to feel compassion, even dabbles in what it calls creativity.
But what we call AI today is nothing more than a statistical machine: a digital parrot regurgitating patterns mined from oceans of human data (the situation hasn’t changed much since it was discussed here five years ago). When it writes an answer to a question, it literally just guesses which letter and word will come next in a sequence – based on the data it’s been trained on.
This means AI has no understanding. No consciousness. No knowledge in any real, human sense. Just pure probability-driven, engineered brilliance — nothing more, and nothing less.
So why is a real “thinking” AI likely impossible? Because it’s bodiless. It has no senses, no flesh, no nerves, no pain, no pleasure. It doesn’t hunger, desire or fear. And because there is no cognition — not a shred — there’s a fundamental gap between the data it consumes (data born out of human feelings and experience) and what it can do with them.
Philosopher David Chalmers calls the mysterious mechanism underlying the relationship between our physical body and consciousness the “hard problem of consciousness”. Eminent scientists have recently hypothesised that consciousness actually emerges from the integration of internal, mental states with sensory representations (such as changes in heart rate, sweating and much more).
Given the paramount importance of the human senses and emotion for consciousness to “happen”, there is a profound and probably irreconcilable disconnect between general AI, the machine, and consciousness, a human phenomenon.
What? No.
Chatbots can't think because they literally aren't designed to think. If you somehow gave a chatbot a body it would be just as mindless because it's just a probability engine.
Exactly. People see “AI” and think LLMs and diffusion models. Those are both probabilistic translation engines. They’re no more intelligent than an AC/DC converter, just a lot more complex.
However, there are neural networks and sense arrays in the field of AI, and those are designed to replicate the process of thought.
The real route to a thinking AI is likely a combination of the two, where a neural network can call on expert systems including translation engines to do the heavy lifting and then run a more nuanced decision tree over the results.
Thing is, modern LLMs and diffusion models are already more complex than a single human mind can fully comprehend, so we default to internally labelling them as either “like us” or “magic”, even when we theoretically know them to be nothing but really deep predictive models.
The problem is in the definition of intelligence.
To me, intelligence is simply problem-solving ability. It does not necessarily imply consciousness, having self-awareness or anything like that. A simple calculator is already displaying intelligence, even if limited to a very narrow situational set of problems, in the sense that it can resolve mathematical questions.
That doesn't mean the calculator is self aware.. it just means it can resolve problems. Biological systems can also resolve problems without necessarily being aware of what they are doing.. does the fungus actually knows it's solving a maze the scientists prepared for it when it just expands following what is preprogrammed by its biological instincts determined by natural selection? Do the ants really know what they are doing when they find the shortest path just by instinctively following a scent of pheromones left by other ants?
Knowing exactly what causes consciousness is an entirely different problem.. and it's one that has not been resolved by any scientist or philosopher in a satisfactory manner. So we simply do not know that.
All the evidence suggests that our own minds are also nothing more than probability engines. The reason we consider humans to be intelligent is because our brains learn to model the events in the physical world that are fed into our brains by the nervous system. The whole purpose of a brain is to try and keep the body in a state of homeostasis. That's the basis for our volition. The brain gets data about about the state of the organism, and interprets it as hunger, pain, fear, and so on. Then it uses its internal world model to figure out actions that will put the body into a more desirable state. From this perspective, embodiment would indeed be a necessary component of human style intelligence.
While LLMs on their own are unlikely to provide a sufficient basis for a reasoning system, its not strictly impossible that a model trained on sensory data from a robot body it inhabits wouldn't be able to build a representation of the world and its body that could be used as the basis for decision making and volition.
My understanding is that the reason LLMs struggle with solving math and logic problems is that those have certain answers, not probabilistic ones. That seems pretty fundamentally different from humans! In fact, we have a tendency to assign too much certainty to things which are actually probabilistic, which leads to its own reasoning errors. But we can also correctly identify actual truth, prove it through induction and deduction, and then hold onto that truth forever and use it to learn even more things.
We certainly do probabilistic reasoning, but we also do axiomatic reasoning i.e. more than probability engines.
This completely understates the gulf between what we call AI and how the human brain actually works. The difference is so severe that acting as if they’re quantitatively comparable is basically pseudoscience. You might as well start claiming that we’re not far off from building a Dyson sphere just because we invented solar panels.
Most “AI” these days are built using linear feed forward networks. The brain is constructed using nonlinear recurrent networks which are can do far more with less. Now you could theoretically create the same output from a linear feed forward network but it’s way less efficient and would require many more neurons to achieve such a result. Which is wild when you consider that there are orders of magnitude more synapses in just the regions of the brain associated with language than there are parameters used in even today’s most advanced “AI” models. Now consider that human synapses rely on over a hundred qualitatively different neurotransmitters and not just a single 16-bit number. It’s also not just the scale of the signal that transmits information in a human synapse but the pattern too. Would you be surprised to know that there are a whole variety of signaling patterns neurons use? Because that’s true too. I haven’t even gotten into the differences in complexity in terms of how neurons process the information they receive. As of now there is no “AI” system that comes anywhere close to replicating that kind of complexity. It’s absurd to suggest where dealing with qualitatively similar machines here.