Peter Kratky

## On the categorical claim of human knowledge being "real" Is the computer knowledge real? Saying it's not real by pointing to mechanisms ("ah, that knowledge isn't real because it works *like this*") is like me saying "the brain just computes electrical/chemical signals and therefore doesn't have real knowledge". But I'd like to think my knowledge is real. At least I hope so. We're clear that a model doesn't always get the right answer, but neither do I 😆 But okay, I can make a well-informed + structured moral judgment about something, because my knowledge of morality is real. But Opus does that too. If what it does were pure memory or shallow correlation, we wouldn't see it generalizing new ideas. And saying it's probability, math, etc, that speaks about mechanisms, not capabilities. Describing the mechanism doesn't settle whether a capacity can emerge from it. Sure, it doesn't sit right with a person to think their computer has real knowledge about morality -and look, to be clear, I'm not necessarily saying it *does*- but how do I define that it *doesn't*? That's where things get interesting. What counts as reasoning that one *can't* reduce to pattern recognition? Because if we try to define "real knowledge" in a way that includes me but excludes the model, we usually fall back on consciousness or subjective experience; and that sidesteps the question of capability entirely and turns into philosophy of mind, I think. Or I can just demand a substrate, or a computational architecture, or some floor that one feels comfortable with. Like, "understanding computed by a neuron *is* understanding, but an understanding computed by sand isn't", but that feels more like stipulating something out of sheer stubbornness than an actual principle. Or you can get vague and draw a spectrum, which is more comfortable sure, but then it stops being categorical (it's no longer "its knowledge isn't real" but something more like "Ahh well, understanding is a spectrum and I'm here and the model is over there") but that feels more like retreating into the empirical than actually getting to the meat of the matter.