I agree with all those points. And I’d go further - or at least be more explicit. I’d say that an object - such as a coffee cup - consists of those local features, organized by allocentric mapping, into a model of that cup. This model is completely new, only ever made by those who have seen a coffee cup. But the model contains - is made up of - the SDRs found in other layers of cells. And those SDRs contain those cells that represent / mean the local features which constitute the cup. eg the curve of the handle, whiteness, the rough china at the bottom, etc.
I’d be v interested to know if you think I’m inventing new ways of understanding HTM.
Or just re-inventing the wheel?
Are you sure you want to say that ?
I suspect that the idea you can ‘learn’ yellow is incorrect as I can conceive of no ‘temporal associations’ that will convey ‘yellowness’. You have to start off standing on solid ground. Bootstrapping doesn’t work because you can’t lift yourself up by tugging on your shoe laces.
Also we know that some quite complex things can be genetically programmed and thus chicks, for instance, come with some knowledge hardwired. I think colour would be the sort of thing evolution would want us all to have prewired.
You have hit on exactly what Jeff Hawkins and crew is saying with this:
You are right as far as you go: what you are missing is that as the WHERE stream progresses around the inferior Parietal Lobe onwards to the Temporal Lobe we pick up semantic meanings - “it is convex so it is likely to be able to hold things, perhaps even liquids if it does not have holes.” This higher level representation/abstraction is really more of the same - parsing of properties of the real world by learning and then recognizing.
The same way you described where a small clique of neurons agree (so the system noise is reduced by the square root of the number of observations) so the facts of an object gather into a cloud of meanings we get progressively more certain about what we are observing.
This freighting of semantic meanings is a key part of the development of language. At some point language stands as its own system outside of the sensory stream, but connected to the (motor) planning system.
Of course there are temporal associations for “yellowness”. When I wake up in the morning, the light coming in my window is yellow. When I look at a lemon it is yellow. When I type “y-e-l-l-o-w” I “see” the color in my minds eye. Yellow is simply some pattern of bits coming from my eyes (doesn’t have to be the same pattern for every person) and associated via other patterns coming from past experiences that have a temporal correlation.
Sure, a lot can be prewired, but the point I was making is that it doesn’t have to be prewired to such minutia. Yellow may be a bad example (for example we are probably hard-wired with some yellow-related stuff to avoid dangerous bugs). But is that necessary for an HTM system to function, or can such things be learned over time?
For example, a baby might be hard-wired to be frightened by a burst of signals following a certain pattern or pathway from his eyes that has a high overlap with some pre-wired recognition of something yellow buzzing by. But he has no concept of what a yellow-jacket is or why he is frightened. He assigns meaning to those signals and formulates an explanation for his fear through life experiences.
I’d have thought that yellow is a primary meaning. In that one can’t ‘create’ a sensation - an experience - of yellow, by saying ‘less red than orange’. Plus the eyes are wired to react to those wavelengths we usually call ‘yellow’ so, yes, I’d say it is hardwired but that is not what I meant. What I meant is that the information that forms the basis for one’s conscious experience of seeing yellow, is a Fact, a thing in itself. What did Plato call them? An essence? An ideal?
But why (other than to pre-wire some behavior like avoiding yellow-jackets) would the yellow I see necessarily be the same as the yellow you see (or the “x1,y1”)? It seems to me that the fact humans as a whole are more intelligent than some other species such as wildebeest with a lot of pre-wiring, and yet start out not even being able to lift their head or focus their eyes, tells me that pre-wiring may not necessarily be a precondition for intelligence. But I could be wrong… just seems at the moment to be unnecessary to me.
I’ve noticed two things about the way you think. One is that you often focus on the activity associated with the higher levels of HTM - the more abstract stuff - while I am more focused on things closer to input, like perception because I am most interested in the nature of meaning. And secondly that you seem to assume that language is a basic part of our understanding. Whereas I’m not interested in the use of the words ‘red’, ‘rote’ or ‘rouge’. I’m fascinated by the thing that is common to them all - redness. I’m interested in their meaning, their referent; both the conscious, visual experience of the colour. And its role in our unconscious, or at least preconscious, processing. If you know Wittgenstein then I’m interested in the Facts of the world which he describes in the Tractatus while your interest seems more congruent with this thoughts about meaning as the product of a language game one plays with words.
Does anyone have a clue what I’m talking about? Or shall I explain? Or would you rather I dropped it ?
I believe I understand what you are saying. You are basically talking about the lowest level of output from an “encoder” (sense organ). The signals that represent “red” or “yellow” or “hot” – before any abstractions are formed. The basic signal that goes to the brain before any learning has been applied. There are also such basic signals for location (“1x,1y” in the above example).
Your argument (which may be correct) is that these things are hard-coded into our genetics to have specific patterns, such that “yellow at 1x,1y” prior to any learning is exactly the same for every (non color-blind) human. My argument is that such commonality between all individuals doesn’t seem like it should be necessary.
If I can learn to form associations at higher levels of abstraction in a hierarchy, why couldn’t the same process be applied to the very lowest levels as well? My “yellow” doesn’t have to be the same as your “yellow”. It simply has to exist as some pattern that can be recognized.
Totally disagree ! The simplest elements of our world - space, colour, pitch of a musical note, sensations of touch - are, to my way of understanding the contents of the brain’s information processing system. They are intrinsic. They constitute the basic building blocks of our perceptions and thus of our experience. Yellow is a simple sensation. It cannot be confected. Whereas a jacket is as much an artificial construct as the words of a language or a painting. Its the micro-features we perceive [the edges, the colours, the change from bright to dark, the detection of movement] that are perceived by all animals [more or less] that are hardwired. And the way we assemble them, name them, think of them by making allocentric models of them that we car rotate in our mind’s eye that is not prewired. But created in the higher levels of the HTM. While the basic building blocks, the Facts such as colours, are intrinsic to the body’s perceptual systems.
Would you agree - or do i need to work harder to convince you?
evolution ? So we all share the same perceptions of yellow [tigers hiding in the bushes]. If you saw tigers as yellow and me as orange then our kids would get eaten by something they couldn’t tell from a brown tree trunk. Ergo evolution means we have to have the same phenotype - we see the same colours.
Yes, I was dealing with these issues 25 years ago. Now you raise these points I seem to recall that while I worked thru them and came up with my own solutions I don’t recall that they were resolved in the literature. Rather, along with Searle’s Chinese Room and the Symbol Grounding Problem, they just got forgotten.
Even if my yellow is exactly the same as your yellow (which I am still doubtful about – I suspect that DNA doesn’t have enough space to hard-code things to that level – these are more likely encoded as classes of input patterns which would require a lot less space), is that sameness really necessary for intelligence? Are color blind individuals less intelligent than their colored vision counterparts? Sure they might get eaten by tigers more often…
We spend much of our time here discussing the cortex and its organization.
To my mind, the higher levels are where the more interesting things are found, the experience of experience, autobiographical memory, and language.
Bob: " And secondly that you seem to assume that language is a basic part of our understanding. Whereas I’m not interested in the use of the words ‘red’, ‘rote’ or ‘rouge’. I’m fascinated by the thing that is common to them all - redness. I’m interested in their meaning, their referent; both the conscious, visual experience of the color."
If you are interested in qualia and other philosophical tail chasings I would think you would head right to the temporal lobes. The digested senses are registered here as experience. Your experience is in terms of these higher level digested WHAT and WHERE streams. And tags like the color.
The development of the cortex and the process of how experience burrows further into the maps and adds details of experience have been described in great detail from a black-box point of view:
Bob: "And its role in our unconscious, or at least preconscious, processing. "
If you want to go with the lower hardwired automatic levels you need to steer to the limbic system. Note that this level has its own taps of the senses and can direct motor function.
Here we ARE hardwired to recognize primitive shapes, (faces in particular, while spiders & snakes and looming overhead objects are on the list) and feel emotions below the level of consciousness and generate behavior. Our social bonding starts here.
The cortex shapes this behavior but it is all limbic to start out.
I think there is a confusion of ‘levels’ here. What you say is absolutely true of our models of coffee cups and jackets. But - I’d argue - the basic building blocks have to be rock solid consistent, repeatable, universal among us all.
Sure if you are writing an AI that needs to avoid being eaten by tigers . On the other hand, if I am building one which is designed to solve math problems, it probably doesn’t need to “see” at all.
Of course the OP is about how humans are “prewired”, so I am taking the conversation off topic to talk about HTM. I’ll summarize my point by saying that clearly pre-wiring for survival is the major player. But that prewiring may not have so much to do with what makes us intelligent as it has to do with purely survival mechanisms.
Yes, I’ve noticed your work, some of your thought is really elegant. Like your theory of how we can become conscious of information. Neat!
But I’m interested in the nature of the information of which we are conscious. You explain how we can consciously perceive red. I’m interested to know how the information is embodied in an SDR. What is it about the brain state that generates the ‘redness’, what is it that constitutes the mental content of a mental state?
Red is a human label for a phenomenon of light that occurs at a wavelength of 620–750 nm. The SDR for “redness” is not the same in your brain as in my brain. We all learn red, and we all represent it differently. Some people say red is their favorite color. Others associate it with communism. We all have a unique experience of even something so basic in reality as 620–750 nm wavelength of light.
. On the other hand, if I am building one which is designed to solve math problems, it probably doesn’t need to “see” at all.