Jeff Hawkins - Lessons From The Neocortex For AI - Numenta

[@CreationTribe]

The problem with this world is that this video only has ~8k views and 196 likes. Jeff Hawkins, you're the shit! Not only did you just hit me with a solid 5 epiphanies, but I get it - I mean, I totally get it. Man, you just gave me the means to solve a coding problem/project I've been tossing around for a couple of years. Danke!

[@paxdriver]

Brilliant!!! One of the best neurology lectures out there, thank you much!

[@eastonlee2194]

"I'm not joking, a single column can learn thousands of things", this surprised me

[@oker59]

Jacob Bronowski, in his "Science and Human Values", mentions how Mount Everest hikers shows how perception works. Eric Shipton discovered new routes up Mount Everest even the locals didn't know about. He relates learning what the locals knew about Mount Everest, "On the morning of the 27th we turned into the Lobujya Khola, the valley which contains the Khombu Glacier(which flows from the south and south-west side of Everest). As we climbed into the valley we saw at its head the line of the main watershed. I recognized immediately the peaks and saddles so familiar to us from the Rongbuk(the north side): Pumori, Lingtren, the Lho La, the North Peak and the west shoulder of Everest. It is curious that Angtarkey, who knew these features as well as I did from the other side and had spent many years of his boyhood grazing yaks in this valley, had never recognized them as the same; nor did he do so now until I pointed them out to him." - Eric Shipton(through Jacob Bronowski's "Science and Human Values", page 30) Perception is like two complementary surfaces. Either the surfaces match, or they don't. This is like the constituent relations stage of creative concept formation, as I mention in my Origins of mathematical knowledge(third post of this blog). Jacob Bronowski goes further . . . Jacob Bronowski points out that true and false exists at the proposition level and not the formation of abstractions/concepts level(constituent relations level). Once the sense data is put into a concept form, then we ask what is true. "If this is really a mountain, we say, then the bearing of that landmark should be due east;" I'd like to quote the Shakespear he also quotes as well, "Is this a dagger, which I see before me, the Handle toward my Hand? Come, let me clutch thee- Come, let me clutch thee: I haue(have?) thee not, and yet I see thee still. Art thou not fatall Vision, sensible To feeling, as to sight? or art thou but A dagger of the Minde, a false Creation?"

[@FranckDernoncourt]

Lecture starts at 3:41

[@Stan_144]

Jeff is on something big .. I expect he will start getting big recognition .. The key to solve the problem is to have big expertise both in neuroscience and in IT. You cannot solve it if you are just neuroscience expert or if you are just IT expert.

[@paultulloch3483]

amazing video- I can see the future of machine learning coming on loud and clear- selective analysis- and therefore the future of machine learning will be much more predictive oriented- find out the differences and focus the processing power. Layers of the brain focused on predicting the unusual- but a lot of hardware dedicated to keeping the human in the usual- its no wonder we dislike change!

[@thorkrynu4551]

I couldn't stop thinking about the touch my body challenges on KZbin. Sad but true.

[@tobiaszb]

50:00 can this help with modern representation democracy crisis?

[@yejoonlee243]

48:08 Rethinking Hierarchy

[@tobiaszb]

54:00 just read some phenomenology stuff about social (features, behaviors) and sensible dimensions of non conscious field.

[@tobiaszb]

So - a little represetation of a cup lives in me (on some layers) and a represetation of myself lives in me. If I treat self as all the living representations - the World that is for me indistiguishable from representations, I accept partial predictive features of it, all people lives in me. Than the idea of "outside" and "inside" can flip. Just like in a joke: How does a mathematician catch a lion into a cage? . . . . . -He goes inside of it, lock it and make an inversion!

[@CaptchaSamurai]

And then again, you can do invert on inverted. Question: what has changed after first flip and what is a difference between first situation and situation after two inverts (comment to both parts of question: after all we can say that you do "invert operation" on just another different layer).

[@tobiaszb]

I would make a hypothesis that self identity changes and whole structure of computing (constitution) gets smoother - increased computing power, more conections, faster pace? Or overloading..

[@tobiaszb]

Maybe, by doing so constantly You are strathening conections between layers of object and orientation. I need to think more with structure from this lecture.

[@tobiaszb]

I have concerns if a second inversion is possible at all, if everything is a part of You.

[@RobertsMrtn]

I think that Jeff's idea about the predictive nature of the brain is spot on. My only criticism is that he seems to be going in at a too high a level when trying to understand the brain. We need to understand the underlying simplicity and from the simplicity comes the complexity.

[@collinwalker550]

How do we know if there's an underlying simplicity though? The underlying simplicity you're talking about could very well just be interactions between structures of neurons, which we already do understand in the form of neural networks. The brain is the most complex structure we have ever encountered to date, so who's to say even that the complexity Jeff is talking about isn't the underlying simplicity?

[@RobertsMrtn]

Hi Colin, thank you for your reply. I think that there has to be an underling simplicity because the brain structure is defined in our DNA which is stored in every cell in our body. This is a lot of information, but not enough to encode the complexity of the brain. The important thing to understand about the brain is that it is a self organising organ. Practically all our abilities are learned. That is why I think that trying to understand cortical columns is going in at a too high a level. If we can create software to mimic these abilities, I do not want the functions of each layer of a column hard coded into the software. I want the software to try out different configurations until it recovers for its self which one works best.

[@bradleyfrank7933]

Fractal brain theory sounds way more promising... kzbin.info/www/bejne/p2SahJ-vq8pkfcU

[@RobertsMrtn]

Interesting link. Thank you for that.

[@DSBrekus]

In what way is he looking at things from too high a level?

[@j.hanleysmith8333]

This guy talks fast

[@userou-ig1ze]

J. Hanley Smith not so fast that you wouldn't understand it at 1.5x speed- maybe the density of effectivr informationflow is low.

[@AllanBrunoPetersen]

Then turn the speed down to about 2.2-2.4x

[@userou-ig1ze]

Allan Bruno Petersen *up?!

[@Eric.Morrison]

He didn't say anything about AI.

[@Anon-xd3cf]

Computer science geniuses can't get their heads around using a microphone consistently.