On the topic of system predictability I was a bit concerned (at Sunday's meet-up) that human behavior was considered bizarre or chaotic (like the weather) when it really isn't. Chaotic systems like the weather are sensitive to initial conditions and are goalless so are only short-term predictable - the "total uncertainty time" is on the order of days for the weather. Human behavior is fairly predictable in the short and long term and humans have goals, so the overall behavior is more likely (IMO) to be convergent.

As time goes by: languages disappear, currencies disappear, cultures get absorbed, production gets rationalized and interdependency constrains behavior.

Kevin,

when *I* say "Bizarre" it has a capital B and a technical definition. It means "Reductionist Models cannot be created for it". Human behavior is not bizarre to humans; we deal with it every day, and it doesn't even "illogical" in any sense to humans. But if you try to create a robust (i.e. non-brittle) Reductionist model of human behavior you are doomed to fail. No set of equations, no ontologies, no rul-based system, and no normal computer program can be created to adequately and robustly mimic or predict human behavior, and by extension, social interactions, consumer behavior, stock markets, the behavior of corporations, or the global economy.

For building a bridge, you want *intelligent* people - engineers with degrees and slide rules. For matters of the tribe and matters of the heart - for dealing with people - you want to talk to somebody who's *wise*. We have two words in every language I know for a reason - they are two different qualities. The former is Reductionist. The latter, Wisdom, is the result of a lifelong aggregation of tested micro-intuitions expressed as context dependent patterns. In dealing with humans we want to use Intuition - the art of guessing wisely based on a lifetime of experience.

Kev, the reason human behavior is only fairly predictable is that we tend to view it extrospectively. It is why Monica seems to believe it is too bizarre for it to be reduced into individual elements. If I understand her right, she looks at the behavior of others and concludes that the range of behaviors are so great that any patterns found are so random that they are indistinguishable, perhaps not-understandable and therefore, unreduceable. The traditional “Scientific Reductionist” role is to approach problems from this perspective. We normally wouldn’t view a problem from how a gadget views us. However, if we view human behavior introspectively we can begin to understand how to carve it up into more useable pieces. Doing this, it should be possible to duplicate the capability of human behavior, not mimic or predict it. The introspection I refer to begins at the sensory input level. For humans or any other animal it begins during its period of gestation and continues for the rest of its life. By viewing it from this perspective, conceptual aggregation begins. Also, of benefit to us programmers, is that from this perspective the programming to achieve this “conceptual aggregation” is much, much more simple. How we glue together our first concepts are the same as how we do our last. The only difference is that they are concepts of concepts. This also applies to imagination, but first things first. If we can understand this than we can begin to understand how thought functions. These are the objects of our thoughts. Do you see what I mean?

I think issues are being confused. With chaotic systems accumulating more information about the system does not necessarily help you predict its behavior (e.g. the weather). Societies (systems of humans) would seem to get more predictable as more information is accumulated about the individuals - just because you can't do the simulation yet doesn't mean that it's chaotic or that reductionist techniques don't work.

Language processing is a separate issue from trying to predict behavior, and human communication may well be entirely "Bizarre"; understanding the (intended) meaning of a message and predicting the recipient's response are two separate activities.

How humans/animals/AIs work internally may not be important to predicting their behavior since they are usually constrained by various rules and limited knowledge, and being more or less intelligent may not make much difference.

Kev, I think issues are certainly being confused. It seems to me that what is confusing you about human behavior is the perspective you take observing it. It is though you are seeing it externally. As if the goal were merely to duplicate the human behaviors you see demonstrated in front of you. The supposition has been that one could reduce enough of those observable behaviors into their parts, mimic them and “wala” thoughtful AI. Is this your assertion or are we in agreement that this won’t work.

I agree what you are saying that chaotic systems only remain chaotic as long we have insufficient information to understand them. And it’s probably foolish to think that something is going to remain misunderstood forever. Of coarse this is only foolish for those of us that aren’t infallible and god like. They would have that kind of information at hand.

Language will remain “Bizarre” if you insist to view it as an external set of behaviors that are demonstrated by humans, void of any background development. Language would loose its “Bizarreness” if you would look to understand how the concepts that it is comprised of were developed. Within the creatures, which use language, there are no programmers that understand its contributing parts. Before words could describe an object to an entity, the entity needed some set of experiences that gave meaning to it. The rules you speak of only become important after the reality of previous experiences have defined the world sufficiently enough that an entity can apply those rules. Where you refer to knowledge limitations being a constraining aspect regarding human behavior and therefore an internal perspective would not apply. It seems to me that your interpretation and perspective of what internal animal conceptualizations contribute toward the development of intellect to be quite different than mine. In order to begin to understand thought it will be necessary to understand what we think about. It is my opinion that we think about concepts. These concepts are internal.

Kev, I think issues are certainly being confused. It seems to me that what is confusing you about human behavior is the perspective you take observing it. It is though you are seeing it externally. As if the goal were merely to duplicate the human behaviors you see demonstrated in front of you. The supposition has been that one could reduce enough of those observable behaviors into their parts, mimic them and “wala” thoughtful AI. Is this your assertion or are we in agreement that this won’t work.

I think you mean "voila", but no, my point is more that although the actors in society may be too complicated to model individually with any accuracy, that doesn't mean the joint behavior is equally unpredictable. I would contend that the larger and more structured a society is, the more predictable its behavior becomes.

I agree what you are saying that chaotic systems only remain chaotic as long we have insufficient information to understand them. And it’s probably foolish to think that something is going to remain misunderstood forever. Of coarse this is only foolish for those of us that aren’t infallible and god like. They would have that kind of information at hand.

I think we have a different understanding of chaotic - I'm working from James Gleick's book (although it's years since I read it, so I might be wrong). The weather is chaotic in that it doesn't matter how much data you collect about the current state, there is a point in the future (the "total uncertainty time") at which you have no idea what it is doing.

I can can make predictions about human society that are more likely to be true than not at some point in the future, e.g. at some point the planet will have one currency and one religion. As such it's not chaotic.

Language will remain “Bizarre” if you insist to view it as an external set of behaviors that are demonstrated by humans, void of any background development. Language would loose its “Bizarreness” if you would look to understand how the concepts that it is comprised of were developed. Within the creatures, which use language, there are no programmers that understand its contributing parts. Before words could describe an object to an entity, the entity needed some set of experiences that gave meaning to it. The rules you speak of only become important after the reality of previous experiences have defined the world sufficiently enough that an entity can apply those rules. Where you refer to knowledge limitations being a constraining aspect regarding human behavior and therefore an internal perspective would not apply. It seems to me that your interpretation and perspective of what internal animal conceptualizations contribute toward the development of intellect to be quite different than mine. In order to begin to understand thought it will be necessary to understand what we think about. It is my opinion that we think about concepts. These concepts are internal.

A boatload of assumptions there :-)

"Bizarre" is Monica's definition, and I'm happy to go along with it. Language is (usually) the product of many minds working over generations, rather than a single mind. As such you could view it as a separate entity with its own evolutionary path and internal logic. Monica's model-free methods are (as far as I know) making no assumptions about the internal workings of the human mind when analyzing language, and I don't see much wrong with that approach: there's no reason that an AI has to be modeled on human intelligence.

My original issue was just about the nature of predictability in systems with intelligent goal-driven actors vs goalless unintelligent systems, so maybe you want to start a new thread on how you would like to tackle the problems of AI.

Kev, sorry about the misunderstanding. My mistake was allowing myself to believe that this was an open-minded discussion about the nature of predictability. Does everyone else understand what you mean by “systems with intelligent goal-driven actors vs goalless unintelligent systems”. What it means to me is that you don’t approach the problem from a perspective that would allow you to understand these systems in ways that allow you to see how predictable they are. Is truth found only in arguments presented in ways that meet your expectations? Open up a little and consider what I’m saying. Without understanding the mechanisms of what gives meaning to what each of us encounter, these systems will never be predictable. When you ask of “intelligent goal-driven actors” I don’t understand why you include the term intelligent. Can there be actors, which are goal-driven, without some form of intelligence associated with the goals. And are we comparing these actors to systems random in nature (“goalless unintelligent systems”). Or are these systems simply accomplish nothing and therefore are goalless. This may be my ignorance but I don’t understand the comparisons you are asking us to make.

While language is the product of many minds working over generations, this isn’t a history lesion about it ether. I mean this is still an AI discussion isn’t it? The individual must apply meaning to these words in order for them to be useful tools. My question is how does an individual arrive at these verbal understandings. On one hand there is the formulaic process born from chaos which trickled down through evolution and the other sensory information trickled down through evolution used to form ascendingly complex concepts. I go the ascendingly complex concepts version.

I find the use of the term “Bizarre”, bizarre. The premise that we should create a term that means something that is now and will forever be incapable of definition is untrue and presumptuous. Is there among us that kind of authority? If there were I would expect much more wisdom coming from here than what I’ve been hearing. Before we can understand intelligence we must demonstrate intelligence. If from me you should hear something as unwise as this please bring it to my attention. I will be forever grateful.

Kev, What boatload of assumptions are those? I make some generalizations but I don’t understand the assumptions you refer to. I put these thoughts out here to be understood, please point out my misdirection, I have yours. I view this as healthy exchange. And I see you as being very tolerant in that regard. Thanks!

Kev, sorry about the misunderstanding. My mistake was allowing myself to believe that this was an open-minded discussion about the nature of predictability. Does everyone else understand what you mean by “systems with intelligent goal-driven actors vs goalless unintelligent systems”. What it means to me is that you don’t approach the problem from a perspective that would allow you to understand these systems in ways that allow you to see how predictable they are. Is truth found only in arguments presented in ways that meet your expectations? Open up a little and consider what I’m saying. Without understanding the mechanisms of what gives meaning to what each of us encounter, these systems will never be predictable. When you ask of “intelligent goal-driven actors” I don’t understand why you include the term intelligent. Can there be actors, which are goal-driven, without some form of intelligence associated with the goals. And are we comparing these actors to systems random in nature (“goalless unintelligent systems”). Or are these systems simply accomplish nothing and therefore are goalless. This may be my ignorance but I don’t understand the comparisons you are asking us to make.

An ant colony is pretty much a system of non-intelligent goal driven actors: the individuals make decisions mostly based on preprogrammed behavior.

Evolution and intelligence are about choice and reaching goals: evolution makes random choices, intelligence makes deliberate choices. Obviously (given that) an intelligent system is more predictable because it will avoid "bad" choices. Increasing intelligence will reduce the number of acceptable choices.

While language is the product of many minds working over generations, this isn’t a history lesion about it ether. I mean this is still an AI discussion isn’t it? The individual must apply meaning to these words in order for them to be useful tools. My question is how does an individual arrive at these verbal understandings. On one hand there is the formulaic process born from chaos which trickled down through evolution and the other sensory information trickled down through evolution used to form ascendingly complex concepts. I go the ascendingly complex concepts version.

I'm not saying you are wrong, I just tend to agree with Monica that the model-free methods are a reasonable alternative approach. While humans could spend a lifetime reading and still have most of what has been written still to read, a machine could absorb it all, so the model-free approach seems to be better suited to machine learning for this type of problem (learning language).

I find the use of the term “Bizarre”, bizarre. The premise that we should create a term that means something that is now and will forever be incapable of definition is untrue and presumptuous. Is there among us that kind of authority? If there were I would expect much more wisdom coming from here than what I’ve been hearing. Before we can understand intelligence we must demonstrate intelligence. If from me you should hear something as unwise as this please bring it to my attention. I will be forever grateful.

Within the context of this group/message-board Bizarre is fairly well defined. The question in my mind is: given your assumptions about the meaning of what I said, would Monica's AI do better ? :-)

If our only choices were Monica’s perspective on AI or those of the past than, hands down, I’d go with Monica’s. She brings a wealth of research and resourceful ideas to the table. But to make statements that AI can never be understood clearly enough to be duplicated would suggest a clairvoyance that I don’t believe she possesses. Predictability is a symptom of the organizational capability of intelligence and not its only product. As far as I can tell, Monica refers to predictability and a vague association with intuition as key to support her holistic model free approach. She states emphatically that the basis of intuition can’t be defined “no ontologies, no rul-based system, and no normal computer program can be created to adequately and robustly mimic or predict human behavior”. She has told me several times that I am a reductionist and than summarily dismisses my Ideas. There is no demonstration that she understands the direction I have taken at all. I find this particularly interesting sense a good deal of the direction she has taken (e.g. AI is developed from a learned database, is open-ended, intuition is a valid process, albeit definable) agrees with my ideas. I’ve read her articles and seen her videos; and, she talks about what doesn’t work quite a bit. But where are the thinking definitions, how does creativity work, how do emotions fit, what is the subconscious, what are dreams, are these traits visible in lower species and finally how can this be digitized? From my perspective I can do this! It might only be my perspective, but for me it opens a panorama into the human psyche. And, I can explain it. It offers an understanding of how animal evolution has become complex. It is all based on the physical world. It begins with input and output structures. Not the simple ones currently used in robotics but those that can aid in concept formation. We need to discuss, in considerable more detail, what motivates us. We need to understand how we take no action without a need to do so. How these needs migrate through us from when we were fetuses to become what motivates our most complex behaviors. I don’t see discussions of this nature. If we cannot answer these questions, than how are we going to make hardware behave in this manor? Keep in mind that animal’s do this from the constraints of a learning process and so will/can a machine!