10-15 Year Plan to Artifical Human Intelligence? Do you have any system proposals?

[bkaz]

@Bitking

So GI is currently unmeasurable in living critters?

Not on intra-brain level, obviously. But we have all kinds IQ tests

@Bitking

Predictive coding as implemented in flat (single layer) HTM is a lot closer to pattern completion, with a report if the system is unable to form a match.

I posted on that, it’s actually the reverse of conventional predictive coding. Where prediction is supposed to inhibit the input, while in HTM it reinforces the input to fire the neuron. So, that’s obviously wrong if you want build a hierarchy:

Predictive coding in HTM

The way prediction is supposed to work in HTM seems to be the exact opposite of conventional interpretation, as well that in On Intelligence. Prediction in apical denrites is reinforcing, basically an AND to basal dendrites. In conventional predictive coding, prediction is supposed to suppress the input, with only prediction error / surprise propagating to the next level. That makes a lot more sense because it implements input compression, which I think is a core principle of cognition. I gues…

@Bitking

with sub-cortical supervision to do any sort of matching of sensation to selecting an appropriate action.

We’ll just have to disagree on that. I don’t think I can get you off that subcortical horse :).

@Bitking

are you saying that GI is just selecting the best learned response for the current collection of inputs? Simple pattern matching?

Of course not. Pattern discovery (unsupervised), not pattern matching (supervised). Where pattern is a set of matching inputs, so it all comes down to the definition of match. Mine is different