Great to hear about the latest progress Jeff. I’ve been following your work closely since reading your book as an undergrad. Your passion for neuroscience is inspiring. Thought is at the center of all humanity and as you express so eloquently should be a primary focus of our scientific endeavors. Your talks are extremely compelling to this end and adeptly blend foundational and advanced concepts, to cater for diverse audiences. With Numenta’s role as a theoretical pioneer and evangelist for new brain concepts is there a way to capitalize on new mediums such as VR to increase bandwidth and accelerate audience acquisition of the new brain models you describe and explore. What would a Jeff Hawkins VR tour of the brain be like? Visualizing functions of columns and layers in exciting visceral new ways! Until then keep us updated on new breakthroughs!
> Thought is at the center of all humanity Self ... feelings ... funny how that is always left out. The things that an organism must possess to make use of all the storage and association machinery. Looking for meaning and then applying it to the world.
You need 4 points to represent a point in 3d space. You need a group of grid cells to represent the origin of the cup. You then need a group of grid cells to represent the origin of the cup relative to your personal origin.
You have to consider that grid cells did not evolve to represent 3D information. They connect all information in the brain: auditory, tactile, but also abstract concepts, mathematics and language. The brain doesn't 'think' in cartesian terms. Somehow it connects everything using equilateral triangles (as strange as that may sound).
Great stuff Jeff, I've been following you guys since On Intelligence came out which was wonderful. I wonder if anyone has seen something like grid cells & displacement cells EMERGING in (artificial) deep learning networks (with appropriate training)? Because we do see edge and corner & nose & eye & wheel & window neurons emerging automatically in visual deep learning networks . .
Hi Paul, good question. There's a related thread on our open source discussion forum: discourse.numenta.org/t/compositionality-numentas-new-framework-vs-actual-deep-learning-models/4864 Of course feel free to start a new thread there on this question, too.
Hum... then, somehow, displacement cells are just a 2nd tier of location cells: they represent where the "location cells firing on the Logo point x" are placed in relation to the "location cells firing on the cup point a". Nice recursive idea: at the same time that a cup has a location space, the location cells used to represent points in that space have another "location space" of their own. Then, if a 2nd layer of location cells add relative positions of objects, a 3rd layer of such structure could account for relative movement between objects and so on.
Amazing, mind blown! This can really be applicable on self driving AI and general robotics. Andrej Kaparthy is implementing something very similar at Tesla Autopilot
47:50 - Very interesting what he says about Mars colonization not happening without explorer and builder robots going before us to prepare places for us to live. Hard to disagree with that. The whole colonization of space that as a kid I used to be so gung-ho for has a whole complexity to it that was never really put out there. Assuming that we survive long enough to build these robots.
So I guess place cells can be derived by learning simultaneous firing (good old Donald!) of particular grid cells? Hmmm.... I suspect abstract reasoning is thus formed by columns responding to inputs purely from the activity of other columns and hence have an "abstract modality". Everything else follows from there... hopefully!
So I, like many others, I'm sure, am awful at estimating and comparing distances. Just abysmal. In this model, it would seem that being unable to represent spaces in a proportionate way would impair essentially all cognitive abilities. But I have 160 points and as I understand it, that's pretty damn good. I would want to look at how the firing fields of grid cells scale between object representations, perhaps. Also there's a TH-cam video about how populations of cells form manifolds that reflect their lower dimensional logical operations embedded in an n-dimensional space (for n neurons). Computational approaches like this have to have some place in this theory. They go over place cells and grid cells specifically, so I'm curious about how those manifolds fit into the column scheme. Anyway, cheers for this. I actually abandoned neuroscience for several reasons, the lack of a plausible general purpose mechanism for cognition being one of them. So this is something I've wanted to hear for a long time.
Wonderful watching as we get closer to the construction of a thought. But I’m a little tired of hearing about the cup. Please more on to language and abstract concepts. Describe how the sentence, “Do we have free will?” Would form in the brain.
Francis crick - DNA molecule. Human intelligence Thinking about the brain To understand the brain we need new ways of thinking about it. More experimental data will not be sufficient
4:58 - the neurons that are inactive represent our current thoughts and perceptions ???? Do you mean to say active? How many times has this guy given this dinner napkin speech ... does he mean that the inactive neurons are doing something?
"In humans, 90% of the cerebral cortex and 76% of the entire brain is neocortex" en.wikipedia.org/wiki/Neocortex "There are between 14 and 16 billion neurons in the cortex". 86 billion - brain total. These numbers don't make sense: if neocortex is 76% of the brain then it should have more neurons.
It’s a bit depressing that LLMs are seemingly making much faster progress but with less long-term promise than an approach with seemingly more promise based on more advanced neuroscience. I know that Numenta is having success in specific areas but apps using 1950’s level neuroscience are storming the market with powerful applications and tech based on modern neuroscience is chewing away at niche applications.
Sure, all the thinking and consciousness and symbolism is in the neocortex, but the drivers of motivation and life force must be coming from lower down ... the brainstem, limbic, etc. This is what computers do not have, a point of view, and ego, incentives, feelings, hungers, etc. I'm not sure I buy alll this neocortex stuff, except maybe for storage and associations ... driven by emotion and self.
justgivemethetruth I agree, not enough attention is paid to the subconscious, the emotions, and memories, and how these elements end up in the neocortex. Just figuring out how we recognize a cup is a small part of the picture, but it’s a start. I wish he would go further.
@@michelechaussabel732 When Hawkins came out with his book, On Intelligence, I heard him speak. It was a very interesting lecture. Mostly I got from it that the brain was a prediction machine and I can see that very clearly.
Actually it is outdated theory (about memory formation). There was a lot of experiments with long sleep deprivation, which didn't harm memory function. Also there are some people who have a disease which make them to not sleep at all for a long time periods and they have no problems with the memory. There is a theory that during the sleep we get some learning and alignment with our original DNA programs and it starts before child even born. There is a nice book on this topic: Human Givens by Joe Griffin and Ivan Tyrrell.
Really glad to get to get the chance to hear this presentation, can't wait to hear more like this in the future.
Slowly but surely things are coming together. Amazing stuff!
Great to hear about the latest progress Jeff. I’ve been following your work closely since reading your book as an undergrad. Your passion for neuroscience is inspiring. Thought is at the center of all humanity and as you express so eloquently should be a primary focus of our scientific endeavors. Your talks are extremely compelling to this end and adeptly blend foundational and advanced concepts, to cater for diverse audiences. With Numenta’s role as a theoretical pioneer and evangelist for new brain concepts is there a way to capitalize on new mediums such as VR to increase bandwidth and accelerate audience acquisition of the new brain models you describe and explore. What would a Jeff Hawkins VR tour of the brain be like? Visualizing functions of columns and layers in exciting visceral new ways! Until then keep us updated on new breakthroughs!
> Thought is at the center of all humanity
Self ... feelings ... funny how that is always left out. The things that an organism must possess to make use of all the storage and association machinery. Looking for meaning and then applying it to the world.
You need 4 points to represent a point in 3d space. You need a group of grid cells to represent the origin of the cup. You then need a group of grid cells to represent the origin of the cup relative to your personal origin.
You have to consider that grid cells did not evolve to represent 3D information. They connect all information in the brain: auditory, tactile, but also abstract concepts, mathematics and language. The brain doesn't 'think' in cartesian terms. Somehow it connects everything using equilateral triangles (as strange as that may sound).
Great stuff Jeff, I've been following you guys since On Intelligence came out which was wonderful. I wonder if anyone has seen something like grid cells & displacement cells EMERGING in (artificial) deep learning networks (with appropriate training)? Because we do see edge and corner & nose & eye & wheel & window neurons emerging automatically in visual deep learning networks . .
Hi Paul, good question. There's a related thread on our open source discussion forum: discourse.numenta.org/t/compositionality-numentas-new-framework-vs-actual-deep-learning-models/4864
Of course feel free to start a new thread there on this question, too.
Deep mind produced grid cell like architecture in one other neural nets
@@fergusmoffat8926 Produced or found they 'emerged'? Or maybe the architecture requires additional hardware . .
@@paulpallaghy4918 I think they emerged , all though I'm not sure
Hum... then, somehow, displacement cells are just a 2nd tier of location cells: they represent where the "location cells firing on the Logo point x" are placed in relation to the "location cells firing on the cup point a".
Nice recursive idea: at the same time that a cup has a location space, the location cells used to represent points in that space have another "location space" of their own.
Then, if a 2nd layer of location cells add relative positions of objects, a 3rd layer of such structure could account for relative movement between objects and so on.
How do grid cells and displacement cells handle object dilation? The description in this lecture appears to assume that the logo has a fixed size.
Amazing, mind blown! This can really be applicable on self driving AI and general robotics. Andrej Kaparthy is implementing something very similar at Tesla Autopilot
4:59 "The ones that are *active represent your current thoughts and perceptions?"
47:50 - Very interesting what he says about Mars colonization not happening without explorer and builder robots going before us to prepare places for us to live. Hard to disagree with that. The whole colonization of space that as a kid I used to be so gung-ho for has a whole complexity to it that was never really put out there. Assuming that we survive long enough to build these robots.
So I guess place cells can be derived by learning simultaneous firing (good old Donald!) of particular grid cells? Hmmm.... I suspect abstract reasoning is thus formed by columns responding to inputs purely from the activity of other columns and hence have an "abstract modality". Everything else follows from there... hopefully!
At end they will say "only neurons can do brain things,and neuron can't be replaced with nothing"~by neurons
Really inspiring. Moreover your presentation is easy to understand. Thanks a lot.
"Would not expect to see a door on the ceiling, on the floor". Unless you're Dr. Suess
We could eventually control the emotional states and mental states
Human neocortex - 75% of brain
Organ of intelligence
The most important scientific problem of all time
So I, like many others, I'm sure, am awful at estimating and comparing distances. Just abysmal. In this model, it would seem that being unable to represent spaces in a proportionate way would impair essentially all cognitive abilities. But I have 160 points and as I understand it, that's pretty damn good. I would want to look at how the firing fields of grid cells scale between object representations, perhaps.
Also there's a TH-cam video about how populations of cells form manifolds that reflect their lower dimensional logical operations embedded in an n-dimensional space (for n neurons). Computational approaches like this have to have some place in this theory. They go over place cells and grid cells specifically, so I'm curious about how those manifolds fit into the column scheme.
Anyway, cheers for this. I actually abandoned neuroscience for several reasons, the lack of a plausible general purpose mechanism for cognition being one of them. So this is something I've wanted to hear for a long time.
Wonderful watching as we get closer to the construction of a thought. But I’m a little tired of hearing about the cup. Please more on to language and abstract concepts. Describe how the sentence, “Do we have free will?” Would form in the brain.
Can human intelligence be increase
Francis crick - DNA molecule.
Human intelligence
Thinking about the brain
To understand the brain we need new ways of thinking about it. More experimental data will not be sufficient
4:58 - the neurons that are inactive represent our current thoughts and perceptions ????
Do you mean to say active? How many times has this guy given this dinner napkin speech ... does he mean that the inactive neurons are doing something?
"In humans, 90% of the cerebral cortex and 76% of the entire brain is neocortex"
en.wikipedia.org/wiki/Neocortex
"There are between 14 and 16 billion neurons in the cortex". 86 billion - brain total.
These numbers don't make sense: if neocortex is 76% of the brain then it should have more neurons.
because most of the neurons (small ones) are in the cerebellum
It’s a bit depressing that LLMs are seemingly making much faster progress but with less long-term promise than an approach with seemingly more promise based on more advanced neuroscience. I know that Numenta is having success in specific areas but apps using 1950’s level neuroscience are storming the market with powerful applications and tech based on modern neuroscience is chewing away at niche applications.
Further proof of your meta-theory?
www.sciencenews.org/article/neurons-combine-place-and-taste-make-food-maps
Everything you've experienced - neurons
Sensory organ
Skin projects
Sure, all the thinking and consciousness and symbolism is in the neocortex, but the drivers of motivation and life force must be coming from lower down ... the brainstem, limbic, etc. This is what computers do not have, a point of view, and ego, incentives, feelings, hungers, etc. I'm not sure I buy alll this neocortex stuff, except maybe for storage and associations ... driven by emotion and self.
justgivemethetruth I agree, not enough attention is paid to the subconscious, the emotions, and memories, and how these elements end up in the neocortex. Just figuring out how we recognize a cup is a small part of the picture, but it’s a start. I wish he would go further.
@@michelechaussabel732
When Hawkins came out with his book, On Intelligence, I heard him speak. It was a very interesting lecture. Mostly I got from it that the brain was a prediction machine and I can see that very clearly.
Sleep seems to be universal for mammals (many other classes too), with no seep will be no memory formed. How could that fit the theory ?
Actually it is outdated theory (about memory formation). There was a lot of experiments with long sleep deprivation, which didn't harm memory function. Also there are some people who have a disease which make them to not sleep at all for a long time periods and they have no problems with the memory. There is a theory that during the sleep we get some learning and alignment with our original DNA programs and it starts before child even born. There is a nice book on this topic: Human Givens by Joe Griffin and Ivan Tyrrell.