Banner
    A Deepness In The Mind: The Symbol Grounding Problem
    By Samuel Kenyon | September 29th 2012 04:52 AM | 26 comments | Print | E-mail | Track Comments
    About Samuel

    Software engineer, AI researcher, interaction designer (IxD), actor, writer, atheist transhumanist. My blog will attempt to synthesize concepts...

    View Samuel's Profile
    The Symbol Grounding Problem reared its ugly head in my previous post. Some commenters suggested certain systems as being symbol-grounding-problem-free because those systems learn concepts that were not chosen beforehand by the programmers.

    However, the fact that a software program learns concepts doesn't mean it is grounded. It might be, but it might not be.

    Here's a thought experiment example of why that doesn't float my boat: Let's say we have a computer program with some kind of semantic network or database (or whatever) which was generated by learning during runtime. Now lets say we have the exact same system, except a human hard-coded the semantic network. Did it really matter that one of them auto-generated the network versus the other as far as grounding goes? In other words, runtime generation doesn't guarantee grounding.

    Experience and Biology

    Now let's say symbol grounded systems require learning from experience. A first-order logic representation of that would be:

    SymbolGrounding -> LearningFromExperience

    Note that is not a biconditional relationship. But is that true? Why might learning from experience matter?



    Well our example systems are biological, and certainly animals learn while they are alive. But that is merely the ontogeny. What about the stuff that animals already know when they are born? And how do they know how to learn the right things? That's why the evolutionary knowledge via phylogeny is also important. It's not stored in the same way though. It unfolds in complex ways as the tiny zygote becomes a multicellular animal.

    "Unfolds" is of course just a rough metaphor. My point is that learning does make sense as a contributor to achieve animal-like grounding since biology had to do that, but one should consider how biological learning actually works before assuming that any software program's autonomous state changes are similar.

    I think it is a safe statement that the grounding of symbols in biological organisms is a combination of ontogenetic and phylogenetic learning [1]. To map that back to software systems--the architecture and the program and the starting knowledge (which may come from previous instances) are just as important as what's learned during a single run of the program.

    How Deep?

    To be fair, in my last post I did not say how grounded a grounded system needs to be. It would appear that the literature for symbol grounding generally means grounded to the real world. The "real world" may be up for debate, but one way to put it is: grounded to the world that humans share.



    This assumes that all of us are in fact living in the same world which is objective. I'm just saying this for the purposes of defining symbol grounding. There is, however, some potential weirdness regarding interfaces with reality that we will skip for today.

    I threw together this simple diagram to illustrate the generic layers involved with real world grounding:



    Bass Ackwards

    I'm doing things backwards and giving you a definition of symbol grounding here near the end of the essay.

    I would say that a "symbol" could be a lot of things, even a complicated structure (relevant: my post What are Symbols in AI?). But let's look at Stevan Harnad's definition of what a symbolic mental system requires [2]:
    1. a set of arbitrary "physical tokens" scratches on paper, holes on a tape, events in a digital computer, etc. that are
    2. manipulated on the basis of "explicit rules" that are
    3. likewise physical tokens and strings of tokens. The rule-governed symbol-token manipulation is based
    4. purely on the shape of the symbol tokens (not their "meaning"), i.e., it is purely syntactic, and consists of
    5. "rulefully combining" and recombining symbol tokens. There are
    6. primitive atomic symbol tokens and
    7. composite symbol-token strings. The entire system and all its parts -- the atomic tokens, the composite tokens, the syntactic manipulations both actual and possible and the rules -- are all
    8. "semantically interpretable:" The syntax can be systematically assigned a meaning e.g., as standing for objects, as describing states of affairs).

    The Problem, is, as Harnad's abstract puts it:
    How can the semantic interpretation of a formal symbol system be made intrinsic to the system, rather than just parasitic on the meanings in our heads?

    Harnad proposes a system in which all symbols eventually are based on a set of elementary symbols. And these elementary symbols are generated from and connected with non-symbolic representations. The non-symbolic representations are caused directly by analog sensing of the real world.

    Harnad's low level interface, which he abstractly constructs from a notion of connectionism, reminded me of something ubiquitous (whether you realize it or not) and realistic: tranducers and ADCs.

    The transducers I have in mind are devices which convert continuous analog movement into a digital numeric representation, e.g. you move a joystick forward and the computer reads that as an integer and adjusts the speed of your video game character accordingly. ADCs (Analog-to-Digital Converters) sample the real world at some frequency and force the measurement into a limited number of bits. For example, recording music.



    A digital-to-analog conversion does the opposite, e.g. playing an MP3 results in real world sound waves hitting your ear.

    Practical Applications


    If the Symbol Grounding Problem defines the ground as the real world, what about software agents and other non-situated, non-embodied entities? For practical purposes, perhaps we need to acknowledge different kinds of grounding. Of course, any arbitrary environment which is a target of grounding will not necessarily result in a system that humans can interface with.


    References

    [1] animal behaviour. (2012). In Encyclopædia Britannica. Retrieved from http://www.britannica.com/EBchecked/topic/25597/animal-behaviour/282494/Ontogeny

    [2] S. Harnad, "The Symbol Grounding Problem". Physica D 42, 1990., pp. 335-346. Available: http://users.ecs.soton.ac.uk/harnad/Papers/Harnad/harnad90.sgproblem.html


    Image Credits


    1. http://my.opera.com/Words/blog/2008/05/27/baby-herring-gulls

    2. http://www.surlalunefairytales.com/illustrations/jackbeanstalk/hassalljack6.html

    3. Samuel H. Kenyon

    4. http://www.dilettantesdictionary.org/index.php?search=1&searchtxt=analog-to-digital

    Comments

    Stevan Harnad

    There's Just One Real World: Rest Is Just Hermeneutics

    Until further notice, there is just one "world": The physical universe all around us. The virtual "worlds" we generate with our software -- even when wired to a "virtual reality" that interfaces with our (physical) senses with (physical) transducers -- are merely hermeneutics (interpretations), generated by our grounded brains (likewise physical, not symbolic).

    It's easy for our minds to get lost in the hermeneutic hall of mirrors they create, mistaking our interpretations of symbols for being entities unto themselves, rather than projections from us groundlings.

    Harnad, S. (1990) Lost in the hermeneutic hall of mirrors. Invited Commentary on: Michael Dyer: Minds, Machines, Searle and Harnad. Journal of Experimental and Theoretical Artificial Intelligence 2: 321 - 327. http://cogprints.org/1577/

    SynapticNulship
    Thanks for the comment Stevan. It's always awesome to get the author of a paper I mention to reply.
    Until further notice, there is just one "world": The physical universe all around us. The virtual "worlds" we generate with our software -- even when wired to a "virtual reality" that interfaces with our (physical) senses with (physical) transducers -- are merely hermeneutics (interpretations), generated by our grounded brains (likewise physical, not symbolic).
    And it is important to state that as part of the definition and philosophical underpinning of "symbol grounding", else people will be claiming grounding for their computer programs (e.g. expert system style programs such as NARS ) and/or single algorithm (e.g. Hierarchical Temporal Memory). 


    Re "Lost in the hermeneutic hall of mirrors":
    The real target of Searle's criticism, however, is a hypothesis I have dubbed "symbolic functionalism" (also known as "Strong AI," and the "symbolic" or "representational" theory of mind) according to which thinking is just symbol crunching (T=SC).
    T=SC has three prominent strengths, plus one fatal weakness,

    And your solution is: create the system using physical interfaces. 

    Well, this would leave no alternative but dualism and mysticism, for what else could it be but T=SC? Answer: Nonsymbolic functions such as transduction and analog transformations could play a much more important role than symbol crunching in implementing a mind in a robot, grounding the interpretations of the robot's internal symbols in its sensorimotor capacity to discriminate and identify the real objects and states of affairs its symbols stand for 
    Having never read this paper until now, it would appear that I was correct in predicting that your "Symbol Grounding Problem" paper approach might lead to special importance of the role of transducers and analog transformations, despite that paper clouding (perhaps) the situation with the supposed dichotomy between symbolism and connectionism.


    You have provided only one salvation for symbol crunching. I am not sure if there are any others. For now I can't argue against it. Fortunately the space of embodied situated AI is exactly the primary space where I have been thinking about thinking since I started realizing the problems with the expectations and claims by creators of most robotics and software. I also spend a lot of time thinking about the abstract architecture and information system issues of minds, but I never claim that a part of that architecture in isolation "thinks" or really matters at all when ungrounded. This will become more apparent as I start writing more about AI in the future (I have barely begun to try to get my thoughts about thinking out there via blogging.)
    Gerhard Adam
    Very good article.  It seems that in considering something like "learning from experience", we tend to overlook the question of why that should be necessary.  Just as you mentioned the point about what animals know when they are born, the question comes back around to the survival of the organism.  In other words, gathering experiences, acquiring knowledge and learning are all focused around things that "matter" to the organism in question.

    In short we are motivated to learn these things because our biological systems respond well or poorly based on the choices we make.  If we make too significant a mistake, then we may lose the opportunity by being killed.  Again, we have an incentive to succeed.  All of these elements provide a "value" context against which the organism will operate in order to learn from experience.

    One of my arguments regarding AI, is that while it is relatively straightforward to construct software to win at chess, the much more difficult problem is to have software that "cares" that it won.  Without that latter element there is a fundamental disconnect as to why software should care to engage in a chess game in the first place.

    When it comes to such biological systems, it is useful to consider how related their experiences are to their biology.  As an interesting thought experiment, suppose that we could literally transplant a brain with all of its knowledge, etc. successfully.  Now let's imagine that we take a human brain and we transplant it into a horse.  Would we expect the horse to behave like a human, or would it still behave like a horse?  In short, would the brain begin to interpret the world based on the physical reality of being a horse?  It seems pretty clear that the first thing that would have to happen is that the brain "remaps" it's sense of the physical body, so that it can even move around.  From there the brain would have to struggle and solve the problem of what it means to be a horse.

    As a result, while we certainly can't know how such an event would play out, it seems pretty clear that such an organism would be at risk because the brain/body connection has been compromised, and consequently almost all the intellectual effort would have to focus on establishing what is important to being a horse. 

    Anyway ... apologies for just rambling.  Good article, though.
    Mundus vult decipi
    SynapticNulship
    As always, thanks for commenting Gerhard.
    It seems that in considering something like "learning from experience", we tend to overlook the question of why that should be necessary. 
    And we also tend to get all excited by the very used of the term "learning from experience." It invokes notions of software organisms forging their way in the harsh complex ever-changing world, living free or dying. But really people are just throwing around that term when all they've got is a computer program that changes its own state. But guess what, most programs do that and/or change their state via human interaction, and that state change affects the runtime behavior. Indeed, the word "learning" has been hijacked so much that I assume guilty until proven innocent for any claims of computer learning in a manner similar to biological organism learning. There is even an entire computer science field called ML (Machine Learning); and, successful as it may be, it has largely nothing to do with animal learning as far as I can tell.
    Now let's imagine that we take a human brain and we transplant it into a horse.  Would we expect the horse to behave like a human, or would it still behave like a horse?  
    As a result, while we certainly can't know how such an event would play out, it seems pretty clear that such an organism would be at risk because the brain/body connection has been compromised, and consequently almost all the intellectual effort would have to focus on establishing what is important to being a horse.  


    Right, and if you morphed into a giant insect--how would your information patterns even map correctly at start? There's no isomorphism for the information state transfer, let alone for the body parts.

    It gets worse--let's say we want to start easy and just transplant a small amount of knowledge from you to me. How would that graft itself into my semantic network? There is some small hope that we have some common, overlapping concepts and networks, being the same species of animals with overlapping cultures. But when you think about the details, it may be incredibly hard--it's like moving a section of a spider web to another web. You might actually need some kind of external grafting mechanism (perhaps nanobots) to grow the knowledge into the target, and it seems that the knowledge would never be exactly the same between the source and the target. 

    Gerhard Adam
    Yes.  One of the things that sheds a bit of light on this, are people that suffer from brain defects.  In particular a story of a woman that lost all of her past memories due to an infection.  It illustrates how much a product we are of our own history, real or imagined.  In this case, she no longer knew what kind of person she was.  Certainly she behaved in a reasonable manner and interacted with people and had no difficulties.  However, the interesting bit was when long-term friends would indicate how much different her personality was, where before she was more aggressive, more inclined to profanity, etc. while now she was behaving without that "historical" knowledge of herself so she could only react based on her own sense of how to behave.  There was no precedence.

    From this, it would seem that one could argue that even those elements that we consider fundamental to ourselves are readily subject to radical change and that the consistency of what we consider to be our brain operation isn't well defined, but rather reinforced by our own perceptions.

    It also seems like you're using the phrase "semantic network" as somewhat analogous to the idea of a "belief system".  As you stated, if two individuals held significantly different beliefs, then how would such knowledge transfer map between the two.

    BTW, in a totally unrelated irrelevant way, your point about transferring thoughts illustrates very clearly the argument against "mind reading".  It also raises the problem of whether preserving one's brain as a kind of backup, makes any sense if the "restore" process doesn't precisely replicate the original environment.
    Mundus vult decipi
    Gerhard Adam
    ...to grow the knowledge into the target...
    Just wanted to get your thoughts on this particular point.  What do we actually mean by knowledge?  It seems to me that we might be able to acquire abstract kinds of knowledge in this fashion, although it seems hard to imagine how we would define "relevance".

    Other types of knowledge become significantly more problematic though.  After all, what does it mean to possess the knowledge of playing the guitar?  or of martial arts?  Would one find themselves in the awkward position of having the abstract knowledge of something, but their body would be completely out of sync with the task.  In that case, how that does differ from merely having a fantasy about that skill?  In other words, even transferring the knowledge to play guitar into the brain ... does that produce anything beyond a more sophisticated "air guitarist"?

    Essentially the question becomes how do we know that we know something versus merely thinking that we do.
    Mundus vult decipi
    SynapticNulship
    Of course there are different types of knowledge and perhaps different memory structures / mechanisms in our minds. 
    In other words, even transferring the knowledge to play guitar into the brain ... does that produce anything beyond a more sophisticated "air guitarist"?
    I suppose that since our brains are plastic, that a properly planned knowledge transfer would include a portion of adaption to the target body / environment. Especially for skills. And really the source person--the great guitar player--is actually adapting anyway, slowly. But how would we accelerate that extreme adapting phase?

    It's interesting to think about the potential analogous issues there will be with cloud robotics. A robot anywhere of any kind with the right interface can download information and skills and access more processing capacity. But each machine will have to have a lot of overlapping compatible interfaces in order to convert knowledge and programs...one can imagine the trouble an autonomous trash vehicle will have trying to use the guitar playing robot skills.
    I, too, enjoyed this article. Very much. I didn't comment on your last one, but please continue. You see, I have gained some 'knowledge'. Now I need to understand it.

    vongehr
    You start out with an important aspect, namely that also "necessities", like experience or evolution that made the system, can always be thought of as being simply wired up by somebody on a different planet, perhaps by accident.
    After that however, I am either too thick to see it or you say nothing of further insight, especially with the ADC stuff. It seems to me that the main problem here is the missing question. What do you want "grounding" to achieve? Your article suggests some "grounding" in the "real world", but there can be no such grounding. There is no necessity for such grounding.
    SynapticNulship
    After that however, I am either too thick to see it or you say nothing of further insight, especially with the ADC stuff
    Insight is not guaranteed here. I was talking about the ADC/transducer stuff because Harnad's paper describes a potential grounded AI system in which the transformation of sensory information into symbols is key.
     It seems to me that the main problem here is the missing question. What do you want "grounding" to achieve? 
    The goal is to figure out how a computer program and/or artificial organism can have internal meaning without an external human injecting meaning. The second goal is to figure if it's worthwhile or makes sense to describe the grounding within contexts of computer programs, i.e. a semantic network may not be self-grounded in the physical universe, but you could say here is where it's grounded to another semantic network, or to the human input, etc.
    vongehr
    The goal is to figure out how a ... organism can have internal meaning without an external human injecting meaning.
    This is either the usual question independent from artificial stuff, or it is perhaps satisfactorily solved if child-like systems are achieved, i.e. we can teach them inside a simulation (or do parents count as “external human injecting”?). In both cases, I have difficulties with “grounding of internal meaning”. Just to make sure I understand you: My “internal meaning being grounded” is no different, and this all relates to “neural correlates” of stuff I somehow "report" (behave visibly) on, and how these correlates emerge physically in detail? “Internal” just means related to stuff inside the apparatus, correct?
    Thor Russell

    You can put me more in the "bottom-up" camp. (from the top down, vs bottom up in “What are Symbols in AI?”) 

    A problem I think is that many people have made assumptions about symbols that aren’t justified. Symbols seem pretty basic, elemental and timeless to us, so therefore a good place to start, but I am not sure intelligence works that way. Symbolic thought has only been demonstrated by organisms with brains with billions of neurons and many trillions of connections and then it is unnatural, we try to avoid thinking rationally whenever possible.  However this is almost always the starting point of AI. It may be that symbolic thought only works the way we want and find useful when backed up and built on top of an already incredibly complex and capable system. I don't think you can use it as the starting point. It seems more like the tip of the iceberg.

    Regarding Harnad’s system, what is a "non-symbolic representation". It seems a bit binary to me to have non-symbolic and symbolic representations, after all what is there to stop you attaching a symbol to any representation, hence making it impossible to point to a representation that is definitely non-symbolic. Representations becoming ever more abstract/symbolic makes more sense, with no clear point at which you should attach your own symbol.

    The comment “The non-symbolic representations are caused directly by analog sensing of the real world.”  So how does this happen? I think this relates to your next quote:

    "I think it is a safe statement that the grounding of symbols in biological organisms is a combination of ontogenetic and phylogenetic learning [1]."

    Yes a lot of stuff definitely is inherited/developed before birth, but you need to explain the very significant things that aren't. I don't t think you appreciate neural plasticity and what it is capable of. 

    For example:
    http://www.nytimes.com/2000/04/25/science/rewired-ferrets-overturn-theories-of-brain-growth.html?pagewanted=all&src=pm

    Now there is no way that putting a visual signal into the audio part of one of our current AI systems would do anything useful, yet that is exactly what happens in biological organisms. The ability to detect vision in the auditory section of the brain can't be inherited or genetic because it obviously has the wrong genes expressed there. It is a result of the intrinsic ability of neurons to detect patterns that enables this to happen. Such ability does not exist in our AI and has been given too little attention as far as I am concerned. There other similar examples too, for example if you have an arm re-attached, then the nerves are often connected up all wrong, yet the brain re-wires itself to adapt.

    Perhaps also relevant it is possible to remove most of one hemi-sphere of someone’s brain and have them function surprisingly well if they are young enough when it happens. (This can be done because of severe epilepsy I seem to remember). Someone with Alzheimer’s or mad cow disease can have large percentages of their brain not work with little to no apparent symptoms. Once again very different to our computer systems but not regarded as an essential feature to be built in from the start by many people who attempt AI.

    Now back to Harnads model, the part where raw analog data makes the "non-symbolic representation" (should be at least partly symbolic representation to me) is very important. Neural plasticity includes examples of this happening. So this initial stage could be considered to be a “building block” of intelligence. We don’t use such a building block (where such representation happens automatically) to attempt to make our AI because we don’t know how it works properly.

    I consider the building blocks we do have (OOP etc) are not anything like as capable as they could be. Comparing our programming techniques to the neural building blocks of biological systems could be like comparing sand held together with weak glue to metal, then wondering why we can't build anything tall.

    To make an analogy regarding starting with symbols, and weak building blocks:

    Building a high level symbolic intelligence in this way is like trying to build the Eiffel tower with sand and weak glue instead of metal, starting from the top down, and then wondering why it collapses or isn’t “grounded”. (Also our computers are less powerful then a human brain, this could be considered like a lack of building material)

    In my view we need to look at the basics of how neurons wire/re-wire themselves etc, with automatic learning/plasticity being a goal. We haven’t cracked the basic “neural code”. If nothing else, mastering these will make it possible to build complex AI systems faster. Focusing on “grounding” issues etc could just be a distraction from this and end up holding us back.

    So I expect that the groups such as Google with their neural net that could finds faces and cats from YouTube images and companies such as Numenta will start to pull ahead of the symbols first (and attempt to ground symbols) approach. As the computing power they have available increases and they get a better understanding of how the brain works at a fundamental level they will make ever more capable systems with less human effort and other people will start to follow because their results will be superior.
    (There is a bug where heaps of blank space keeps appearing at the end of the comment. Even if I edit it, remove it, repost without it, it reappers. I am using Chrome.)
    Thor Russell
    It must be an issue with Chrome. Firefox does fine.

    SynapticNulship
    Regarding Harnad’s system, what is a "non-symbolic representation". It seems a bit binary to me to have non-symbolic and symbolic representations, after all what is there to stop you attaching a symbol to any representation, hence making it impossible to point to a representation that is definitely non-symbolic. Representations becoming ever more abstract/symbolic makes more sense, with no clear point at which you should attach your own symbol.
    There are possibly different ways to implement the zone where we touch down with the physical universe. Harnad gives a philosopher's concept which is far from a design and doesn't go into all the details we might deal with while actually implementing such a system. Hopefully I will be able to give some more detailed and implementable descriptions in the future. Also there are other ways of thinking about animal mind grounding which I haven't mentioned yet. More to come...
     I don't t think you appreciate neural plasticity and what it is capable of. 
    Such ability does not exist in our AI and has been given too little attention as far as I am concerned. There other similar examples too, for example if you have an arm re-attached, then the nerves are often connected up all wrong, yet the brain re-wires itself to adapt.
    Actually I am well aware, and jealous, of biological plasticity and flexibility. I am all for making AIs and even run of the mill computer interfaces more flexible and automatically adaptable. See my article "Softer, Better, Faster, Stronger" .

    Now back to Harnads model, the part where raw analog data makes the "non-symbolic representation" (should be at least partly symbolic representation to me) is very important. Neural plasticity includes examples of this happening. So this initial stage could be considered to be a “building block” of intelligence. We don’t use such a building block (where such representation happens automatically) to attempt to make our AI because we don’t know how it works properly.
    Well neural plasticity might be involved in the jump from raw analog data to representation--or maybe it's orthogonal. Certainly the ability to change representations is used for one's self model, and it's hard to imagine how our self models would work if they didn't change as we grow larger (or smaller). I don't think you have proven that the plasticity itself IS a representation generator. Plasticity may just be an attribute of a representation generator and thus does not actually specify how raw data is turned into icons or other analog data. 
    So I expect that the groups such as Google with their neural net that could finds faces and cats from YouTube images and companies such as Numenta will start to pull ahead of the symbols first (and attempt to ground symbols) approach. 
    I think people have been saying similar things since the 1950s. ANNs are just nonlinear functions configured with a training phase of input data. Architectures and systems are what matters, not any single algorithm. Likewise you could argue that the groups doing parallel processing are the ones who will "pull ahead", yet parallel processing itself doesn't give you anything without a good design of the system including a design for how all the processes communicate with each other.
    Thor Russell
    OK yes neural plasticity may not be the representation generator, and I am also jealous of it! Without our  computers lack of plasticity programmers could have a lot more competition for their jobs. My point about neural nets is that Google used a more sophisticated one and Numenta is very much into making systems and architectures match biology. The old ANN's definitely aren't up to the task. 


    Glad the comment issues is sorted. Could it be to do with me entering the comment not in "Plain editor" mode? (I have a button that says "Switch to Plain Editor" ) Perhaps I will try that later with a longer comment.
    Thor Russell
    SynapticNulship
    Thor Russell, I deleted your extra whitespace, but now the comment has my name instead of yours. Sorry about that. You should probably edit it to get your name associated with it again.
    Gerhard Adam
    I didn't realize I could do this, but I went into Thor's comment and reassigned the authorship of it back to him.  So, it should be good now.

    Just FYI ... I don't know if you see all the same options when you edit a comment, but there's a bullet at the top called "Administration".  If you expand that, you'll see the author block.  If you put the author's identity in there, it should retain the authorship as you want it to.

    Mundus vult decipi
    SynapticNulship
    Thanks. Yeah, I don't have access to the Administration thing.
    vongehr
    A common problem, especially with Russell's white space. Maybe you could once more wield your awesome administrative powers? ;-)
    Gerhard Adam
    Done ... I had no idea I had such powers :)
    Mundus vult decipi
    blue-green

    This notion that actual physical world conditions are essential when it comes to learning and behaving has been important to me for decades. My mnemonic device for it is “matter matters”.

    My reasons for understanding WHY “matter matters” are far flung, and yet, here are two. The first one comes from my Bohrish way of understanding how the entire experiential framework is responsible for the phenomena we see and record. This is a holistic or top down approach, if you prefer those terms.

    My second reason for taking “matter matters” seriously has to do with the degree of resolution and depth in the real world. A 100 mega pixel camera with built in 16-core 100 mega hertz processors would be an engineering feat, and yet, just what do think is the pixel depth and processing rate of the world at nano scales, or nuclear scales or …. Planck scales? No computer can reach them. An artificial cut-off isn't going to work or do justice to what is available. As Stephen Wolfram insists, real processes outrun any simulation that you can make of them. It is only a sliver of “set-measure zero” that can be encapsulated in full with equations.

    SynapticNulship
    Here's a scary thought: Imagine that Motorola, MOS, Intel, never existed, and never came out with cheap 8-bit processor chips in the 1970s. Instead companies start releasing relatively cheap analog computers which become super popular and increasingly complex. All work on artificial intelligence from then on is composed of analog components. Analog to digital conversion and digital computation is almost unheard of and limited to a few crazy professors...
    blue-green

    Here's a scarier thought. Children no longer take an interest in the material world. Everything is done using screens and artificial vibrations. They have no grounding. Physical travel is deemed to be pointless. Relationships are entirely mental. Carnal knowledge advances from being taboo to being meaningless. Surfing is done without oceans. They call it surfing, but what is it really? 

    A groundless afterlife?

    Perhaps all of the interest in grounding and symbols is good, but this interest may be masking the more important question. What good is all of the knowledge in the world without an understanding of any of it?

    I'm fairly certain that smart humans will develop a way for a given computer system to gather more knowledge on its own. I'm less certain that this will ever amount to intelligence. Of course that begs the question 'what is intelligence'.

    As you know, a computer with a sufficiently large storage system could contain all of the accumulated knowledge of all of the humans that ever existed. This computer would be able to dredge up the answer to most any question that could be reasonably asked of it. However large the storage system may be, though, the computer itself cannot think. It cannot spontaneously initiate a conversation.

    My belief systems tell me that intelligence is more than knowledge. If a system cannot internalize the knowledge, turn it on its head, lift up a dark corner to see what lurks there, think about thinking about the knowledge, then it can't ever be expected to understand the knowledge it has.

    Educators have realized that 'rote' learning is not learning at all. Memorizing tables of data without an understanding of where the data comes from and how to use it is worthless until, and unless, the student 'gets' it. I believe that is where we are now in our attempts to create AI.

    Create a system that can understand and you will have your intelligence.

    SynapticNulship
    You had to go all epistemological on me, didn't you?

    Symbol grounding is how an organism's information content can be used. So it is necessary for there to be any knowledge at all, at least in evolutionary epistemology.

    As you know, a computer with a sufficiently large storage system could contain all of the accumulated knowledge of all of the humans that ever existed. This computer would be able to dredge up the answer to most any question that could be reasonably asked of it. However large the storage system may be, though, the computer itself cannot think.
    That's basically Searle's Chinese Room Argument. And that line of thinking leads back to...the Symbol Grounding Problem.
    "You had to go all epistemological on me, didn't you?" Heh. Yes, because I intend to follow your articles closely. You now know where my thought processes lead and will be better able to straighten me out when I question you later.