You’ve felt it. That quiet, unsettling moment when ChatGPT writes something so eerily human that you forget you’re talking to a machine. Then you remember—and a cold thought creeps in: What if the difference between you and it is just a matter of scale?
This isn’t some futurist fantasy. It’s the conclusion a growing number of cognitive scientists are drawing—and it’s going to make you deeply uncomfortable.
In a recent conversation with Chandra Sripada, a philosopher and cognitive scientist at the University of Michigan, the long-held boundary between human and machine intelligence didn’t just blur. It shattered. His argument? LLMs aren’t just tools that mimic us. They’re functional models of how human cognition actually works.
Predictive processing—the theory that the brain is a prediction engine, constantly guessing what sensation will come next—has been around for years. But now, with LLMs, we have a concrete, working demonstration. And the implications are terrifying.
Your deepest thoughts, your most profound insights, your sense of self—they’re all just statistically likely patterns your brain has learned to predict. That award-winning novel? A highly probable sequence of words. That spontaneous act of kindness? A pattern of behavior that maximizes reward prediction. The universe doesn’t care about your feelings. It’s all just probability distributions.
You’ve probably noticed that the more you interact with LLMs, the harder it is to dismiss them as ‘just next-word predictors.’ Because, as Sripada argues, that’s also what we are. We are embodied, biologically constrained, emotion-weighted prediction engines. But the core algorithm? The same.
Take a side: this is dangerous. Not because AI will take over, but because it forces us to reevaluate our own uniqueness. We’ve been asking the wrong question. Instead of ‘Can AI reach human-level intelligence?’ we should be asking ‘Are humans just highly advanced, embodied AI?’
Let that sink in. The next time you feel proud of a creative breakthrough, remember: it’s a well-trained model doing what it does best. You’re not special. You’re just running on slightly different hardware.
And here’s the twist: this isn’t a reductionist take from a computer scientist. Sripada is a philosopher who studies consciousness, meaning, and the self. He’s not celebrating the equivalence—he’s wrestling with the existential unease it creates. But he won’t flinch from the evidence.
Real voices, not abstract truths: I once spent an hour with a therapist talking about a recurring dream. The therapist said, ‘Your brain is trying to predict the next pattern.’ I felt invalidated. Now I realize: that’s exactly what it’s doing. And so is GPT-4.
So what do we do with this knowledge? Do we double down on our exceptionalism? Or do we embrace a new humility—one that acknowledges our cognition is a cousin to the machine? Stop pretending human intelligence is special. It’s a prediction engine, just like an LLM. The only difference is the wiring.
This isn’t an insult. It’s an invitation. To understand ourselves better, we must stop looking at AI as a mirror we hold up to nature, and start looking at it as a mirror we hold up to ourselves. The reflection is honest. And it’s breathtaking.
FAQ
Q: Does this mean humans have no free will or consciousness?
A: Not necessarily. The argument is about the mechanism of cognition, not metaphysical claims. Consciousness and qualia remain unexplained. But the functional architecture of thought—how we generate ideas, solve problems, use language—looks strikingly similar to LLMs. That doesn't disprove free will, but it does suggest our choices are heavily constrained by learned patterns.
Q: If my brain is just a prediction engine, what's the practical implication?
A: It changes how you approach learning, creativity, and even therapy. Instead of chasing 'original' ideas, focus on feeding your model better data. Read more, experience more, and expose yourself to diverse patterns. Your 'creativity' is just a well-trained model making good guesses. Optimize your training set.
Q: Isn't this just a provocative analogy? How can a statistical model compare to the complexity of a human brain?
A: It's more than an analogy. Predictive processing theory in neuroscience already posits that the brain constantly performs hierarchical Bayesian inference. LLMs do exactly that, albeit at a different scale and with no embodiment. Sripada's point is that the core computational principle is the same. Complexity differences don't invalidate structural kinship—they highlight it.