chatbots that use advanced perceptron theory

If you're wondering WTH perceptron theory is, there's a decent intro here (there’s some easy math, hardest bit might be ‘∑’, which means ‘add them all up’ — if you’ve programmed, it’s a ‘for’ loop).

That said, a perceptron models an 'artificial neuron', and IMO does a decent job of explaining how an actual neuron works (think of the body of the cell evaluating inputs in an analog way, then if it passes a threshold, it becomes a single impulse traveling down the long skinny bit). Networks of these can somewhat model a static state biological nervous system. It can predict how a retina works — but other than structures like that, our brains are not at all static.

I'd disagree that networks of perceptrons have imagination at all or the ability to disentangle real-world ‘facts’ from the linguistic data they've been trained on, never mind matching what humans have; overall, they don’t have the general ability to learn arbitrary areas of knowledge on their own.

Maybe when they start modeling artificial cortical stacks, which might include astrocyte interaction, and dendrite network growth (which as of now, appears to be centriole-driven) and so can encompass multiple predictive models in each stack. Jeff Hawkins' "Thousand Brains" hypothesis on how general intelligence might arise from many thousands of cortical stacks fits a lot of the known data.

I find it a particularly interesting area.