|
Author
|
Topic: The Characterization of Intelligent Causation
|
aiguy
Member
Member # 3736
|
posted 06. April 2007 17:02
LE,
quote:
I have defined or identified a mathematical framework or perspective for modeling information processing behavior. That is essentially what Turing did and in fact the F(S)=R is the same framework defined by Turing. I am simply adding a classification of R or the GC-NGC property to the Turings abstract framework.
I have a very different understand of Turing's work. The Church-Turing thesis says essentially that every function can be stated algorithmically, and that every algorithm can be run on any physical system that meets a few basic requirements.
quote:
I am proposing that all behavior be modeled by a slightly modified Turing compatible model. You seem to agree that all behavior modeled by computers involves a Turing type model.
Every model that is stated with sufficient clarity is Turing computable, period. Saying that a model is "Turing compatible" is redundant.
quote:
I wonder how you reconcil this rather stange position with the obvious fact that there are all sorts of computer models in existance of all sorts of behaviors. It is quite contradictory to both claim there are lots of computer models and then suggest there are no such models.
Read what I said: I said that no single model underlies all of AI. Each system uses a different approach, and there is no single model that can be said to represent intelligent behavior.
quote:
The relevant point here is that we can start with any existing models of either human behavior and existing models of evolutionary change and if there is sufficient information available in the model we can determine if the model satisfies the proposed definition of intelligent behavior or the definition of unintelligent behavior.
Let's agree, arguendo, that both evolutionary change and human behavior can be formally modelled. Now, the question is, what is the similarity between the models? What if one model used predicate logic, and another used a connectionist framework, and another used genetic algorithms, and another used a completely ad hoc functional approach? The only thing that all these models share that they are all models (in other words, they are all Turing computable). In every other respect they are different.
IP: Logged
|
|
Melvin H. Fox
Member
Member # 1684
|
posted 06. April 2007 21:34
aiguy,
You don’t like my definition of intelligence. That is understandable since it excludes your pet AI systems. What is more, you don’t understand my definition since you seem to think I am equating inconsistency with intelligence. So, I will start over only this time I will try not to use the word intelligent.
You would like to find out if the cause of human behavior and the cause of biological complexity are the same. In order to do that you will have to understand the cause of human behavior. You want it to be the case that humans are heuristically programmed machines and I admit trial and error is a big part of the human modus operandi. You would like it to be the case that the causes of human behavior can be reduced to two: fixed law and chance. Nobody questions the intense influences on human behavior by fixed law. Also, nobody contends to understand all the causes of human behavior and so the scientific and neutral term chance can be used to express this lack of knowledge.
You have agreed that human behavior is inconsistent. Since the MO is trial and error and not just trial, then we would suspect that humans can learn from their mistakes. This we do observe. Learning implies that error would be reduced over time. Reduced error would imply convergence of behavior. We do not observe this convergence. In fact human behavior is diverging. Why I watched on the news some months ago as a woman joined in marriage with a dolphin (the dolphin’s consent was assumed not demonstrated). This sort of error should have been corrected long ago in a heuristically programmed system of humans that learn. You will be tempted to explain this as a random act. Unfortunately this sort of thing is more common than you or I would like to think and the relationships are not always as honorable.
I suppose one could try to explain the divergence as part of the whole natural selection game. I mean what kind of selection can be made from a bunch of humans that all act in the same identical way. Well then, all you would need to do is demonstrate some allele that carries the trait where the human will reject the general common wisdom stored in the human system. I await the announcement of this rebel allele.
Until then, a more reasonable explanation is free-will. Now, if you don’t like the connotation of the term, then call it what you like. I submit that the individual human has, at the center of his being, a specific self-determined set of designs for his existence in reality. I submit further that all his behaviors are in accord with these designs first, above and beyond all other constraints. The set is constructed very early in life and can indeed be modified by experience. However, all modifications are subject to the authority of self and can not be imposed on him by fixed law, chance, another’s will, or even logical coherence. Finally, I submit that the above characterization is a better predictor of human behavior than the exclusive trio of trial and error, fixed law, and chance.
-Mel
IP: Logged
|
|
aiguy
Member
Member # 3736
|
posted 07. April 2007 04:03
Mel,
quote:
You don’t like my definition of intelligence.
No, I like it just fine; it's just that we can't use it to say that the intelligence of a human being has anything to do with the purported cause of biological complexity.
quote:
That is understandable since it excludes your pet AI systems.
As far as I can see, your definition includes AI systems just fine.
quote:
You would like to find out if the cause of human behavior and the cause of biological complexity are the same. In order to do that you will have to understand the cause of human behavior. You want it to be the case that humans are heuristically programmed machines and I admit trial and error is a big part of the human modus operandi. You would like it to be the case that the causes of human behavior can be reduced to two: fixed law and chance.
No, not quite. Rather, what I am saying is that we can't demonstrate that anything else is at work. And if we hypothesize that something else is at work, then we can't characterize this something in such a way that we can decide if our hypothesis is true or not.
quote:
You have agreed that human behavior is inconsistent. Since the MO is trial and error and not just trial, then we would suspect that humans can learn from their mistakes. This we do observe. Learning implies that error would be reduced over time. Reduced error would imply convergence of behavior.
No, reduced error really does not imply convergence of behavior: You are assuming that every person is attempting to reach the same goals, and this isn't the case.
quote:
We do not observe this convergence. In fact human behavior is diverging. Why I watched on the news some months ago as a woman joined in marriage with a dolphin (the dolphin’s consent was assumed not demonstrated). This sort of error should have been corrected long ago in a heuristically programmed system of humans that learn.
Aha! Now I see what you mean by "human behavior is diverging". You mean that people are doing crazy things, like taking drugs, or having abortions, or men who want to marry men, or women who want to marry dolphins, or all sorts of crazy nonsense. How, you ask, could these nutty and erroneous behaviors possibly happen, if human beings were able to learn? Wouldn't everybody simply learn that it is wrong to marry dolphins?
Is that what you are saying?
quote:
You will be tempted to explain this as a random act.
No, I would not explain this as a random act at all!!! There could be lots of explanations for a woman who wants to marry a dolphin. Maybe she wants to make a statement to society that we can extend our love to other species. Or maybe she look too much LSD in the 60's and thinks the dolphin is Jesus Christ. Or maybe she was born with a genetic trait that makes her get really hot whenever she thinks about dolphins. None of these explanations are "random".
quote:
Unfortunately this sort of thing is more common than you or I would like to think and the relationships are not always as honorable.
What? Are lots of women marrying dolphins nowadays? And their relationships with these dolphins are not honorable? Good grief, this does sound serious. What is the world coming to? (When I was a kid, when a woman married a dolphin, at least her relationship was always honorable).
quote:
I suppose one could try to explain the divergence as part of the whole natural selection game. I mean what kind of selection can be made from a bunch of humans that all act in the same identical way. Well then, all you would need to do is demonstrate some allele that carries the trait where the human will reject the general common wisdom stored in the human system. I await the announcement of this rebel allele.
Come on, Mel. Here is what you are saying: There is a single, common, obvious wisdom that tells humans what we ought to do. Clearly, this wisdom tells us that it is wrong to marry dolpins (and you probably have a very long list of other things that this wisdom decrees as wrong, correct?) And if humans were rational creatures, acting according to fixed law that enabled us to learn, we'd all eventually learn just what the right thing to do is, so we would all do the same thing.
There are more than a few errors in this argument, but I'll just point out four obvious ones. First, of course there is no reason to believe that people are acting any crazier now than they did a hundred, or a thousand, or a hundred thousand years ago. Second, you apparently imagine there is one received canon of wisdom that we know is the "correct" way for human beings to act (I'm afraid to ask where you think this wisdom might have been written down) - but there is obviously no way we can agree on what that wisdom is, and science is of no help at all here. Third, of course there are "rebel alleles" - why in the world would there not be some genetic component to any behavior that you deem "incorrect"?
Fourth, and most relevant to our discussion here, is this: I could write a program with lots of completely deterministic agents that learned and changed their behavior based on their interactions with other agents and their environment. Depending on the nature of these agents, their behavior may converge, or it may diverge, or it may oscillate, or it may vary randomly over time. If the system is sufficiently complex, we could not possibly predict what would happen - we'd just have to watch and see.
quote:
Until then, a more reasonable explanation is free-will...Finally, I submit that the above characterization is a better predictor of human behavior than the exclusive trio of trial and error, fixed law, and chance.
First, trial and error reduces to fixed law and chance, so your trio is a duo. Second, there is no "theory of free will" that has ever made any prediction at all as far as I know. Can you give an example?
IP: Logged
|
|
aiguy
Member
Member # 3736
|
posted 07. April 2007 04:23
LE,
quote:
But, of course, genetic algorithms involve intelligent or efficient searches rather than so called Darwinian chance variation and selection searches.
Any search strategy can be described in terms of a generator and a tester. The generator picks the next node in the search space, and the tester evaluates the node.
The simplest tester is one that can answer only a single question, viz. "Is the current node the goal node?". If the tester is this simple, then there is no information available to the generator to help decide which node to pick next - what you call a random search. If the evaluation function of the tester provides other information, then the generator can try to use this information to pick the next node based on which is most likely to be the goal, or closer to the goal than the current node.
In Darwinian evolution, the tester is the interaction of phenotype with the environment. In one sense, the generator is incapable of incorporating any information from the tester at all: No matter what happens, this information does not affect what genotype will be generated next. But the generator does incorporate information from the tester when the population is considered, since each incremental gain in reproductive advantage informs the generator to pick genotypes with the highest reproductive success as the next set of nodes in a massively parallel search. And of course there is no "goal node" in Darwinian evolution at all; it is not a targeted search. Every generation has some level of reproductive success, which might vary in the next generation in either direction.
quote:
True random search is 'in theory' the basis for genetic algorithms, but in reality, genetic algorithms utilize intelligent or efficient search processes.
This is simply mistaken - genetic algorithms model Darwinian evolution, which is neither "true random search" nor an emulation of anything like human-like reasoning. Now, you might argue that this sort of search still isn't efficient enough to have created eyeballs and flagella in a few billion years; that is a different question entirely. But you can't distinguish "intelligent" search from genetic algorithms (or Darwinian evolution) by saying that the latter is "truly random", because it's not.
[ 07. April 2007, 06:38: Message edited by: aiguy ]
IP: Logged
|
|
nosivad
Member
Member # 767
|
posted 07. April 2007 11:03
There is absolutely nothing in the Darwinian model that ever had anything to do with the emergence of any new life form. All that selection, natural or atificial, can achieve, or ever has achieved, is the production of intraspecific varieties or, in some few instances, subspecies none of which are incipient species in any event. How anyone can still place any credence in the most failed hypothesis in the history of science escapes me entirely.
Ether, Selection, Phlogiston, ESP, exrasensory perception all three, nothing but figments of the human imagination.
In my carefully considred opinion all real tangible evidence indicates that the evolutionary scenario was planned from beginning to end and the end was quite some time ago, at the genus level at least two million years ago, at the natural species level within recorded history. Homo sapiens is the last mammal to appear or ever will appear. All that remains is extinction.
"A past evolution is undeniable, a present evolution undemonstrable."
John A. Davison
IP: Logged
|
|
LifeEngineer
Member
Member # 3446
|
posted 07. April 2007 12:55
Quote: This is simply mistaken - genetic algorithms model Darwinian evolution, which is neither "true random search" nor an emulation of anything like human-like reasoning.
Which Darwinian theories are you talking about? Neo-darwinism, in an effort to claim that evolutionary searches were not purposeful, proposed that evolution involved random variation and selection operating on expressed phenotypes.
If variations are random, then the expected number of trials or iterations involved in finding a solution can by calculated from the areas of the solution space and the search area.
If the actual speed or efficiency of finding a solution is materially faster than what would be predicted by a random search, then the variations must be non-random or designed. The variation generated reflect a 'knowledge' of the goal or a knowledge of how to find a solution in an efficient or intelligent manner.
In business applications of genetic algorithms, the programmers use knowledge of the solution to identify faster or more efficient methods of finding a solution. Data suggests that actual evolutionary processes, like genetic algorithms are far faster and far more efficient than could be explained by a random search process.
While non-random, efficient or intelligent or designed searches are not compatible with neo-Darwian ideology, an argument can be made that Darwin's original theories were fully compatible with concept of intelligent purposeful evolution.
The point remains that most applications of genetic algorithms involve highly efficient non-random searches. If you recognize that efficient goal directed searches are compatible with Darwinian theories, then you are recognizing that some Darwinian theories express intelligent goal directed evolution.
I have always found it interesting that the concept of 'random searches' can be so clearly and unambiguously defined in mathematics, but people seem so willing to accept the misapplication of the label random search to genetic algorithms that are clearly vastly faster and more efficient (more intelligent) than random searches.
One wonders whether the mislabeling is intentional or if the people involved in the mislabeling are truly ignorant of the mathematical concepts involved?
IP: Logged
|
|
LifeEngineer
Member
Member # 3446
|
posted 07. April 2007 13:16
Quote: I have a very different understand of Turing's work. The Church-Turing thesis says essentially that every function can be stated algorithmically, and that every algorithm can be run on any physical system that meets a few basic requirements.
I am not sure that it is terribly relevant to the discussion here, but clearly computers can not precisely process or express continuous functions or algorithms. My understanding that at least part of what Turing addressed was the ability to approximate continuous functions with discrete or discontinuous computer algorithms.
The relavant point here is the assertion that both intelligent and non-intelligent behaviors can be modeled with functions or algorithms of the type "F(S)=R where R is GC or NGS" that can be expressed on computers. The point here is that 1) we are assuming or asserting that the behaviors being analyzed here are all modeled with the type of model defined and 2) a definition of intelligent causation will be based on properties of these models.
This approach or framework is only controversial to the extent that it does not consider or recognize intelligent causation or intelligent behavior that can't be modeled by standard computer or Turing algorithms or functions.
If someone wishes to develop a definition of intelligence and corresponding predictive theories involving processing that can not be simulated by computers or turing machines, they are welcome to attempt to do so. I am simply stating that my approach does not recognize any such processes.
IP: Logged
|
|
LifeEngineer
Member
Member # 3446
|
posted 07. April 2007 13:25
Moving forward, my approach defines intelligent causation in terms of changes in behavior or in terms of dynamic behavior. A relavant form of dynamic behavior might be modeled at time t=a by fa(sa)=ra where ra is NGC and at time t=b by fb(sb)=rb where rb is GC.
In less mathematical terms, we might be saying that at t=a the system being studied can not produce a solution to the problem being analyzed but at t=b it has a solution. We might say that at time t=a the species does not have a solution to particular survival problem or it can not perform some function, but at t=b it does have a solution or it can perform the function.
We might say that between t=a and t=b the system has found a solution or has learned how to solve the problem or has adapted or has evolved a solution. Although the approach being used involves somewhat unusual terminology, it seems reasonably close to the intuitive concept of intelligent causation.
IP: Logged
|
|
aiguy
Member
Member # 3736
|
posted 07. April 2007 14:54
nosivad,
quote:
In my carefully considred opinion all real tangible evidence indicates that the evolutionary scenario was planned from beginning to end and the end was quite some time ago, at the genus level at least two million years ago, at the natural species level within recorded history. Homo sapiens is the last mammal to appear or ever will appear. All that remains is extinction.
And how do you think this opinion relates to the topic of this thread?
IP: Logged
|
|
Daniel Smith
Member
Member # 3004
|
posted 07. April 2007 14:58
aiguy:
quote:
I think you missed the point here. Proteins fold up inside of cells, and they find the correct conformation out of countless possibilities in a short time. Since we can't explain how this happens by any known combination of fixed law and chance, the filter says that the result of this process - the folded protein itself - is the product of intelligence. Therefore the thing inside the cell that is folding this protein must be intelligent!
Now, do you think that the thing inside the cell that makes proteins fold up right is the same thing that enables us to design watches?
If I design a machine and program it to "make decisions" based on inputs from the field and perform actions based on those inputs, is that machine "intelligent"? Or is it merely doing what it was programmed to do?
I'd argue that we do know what causes proteins to fold the way they do - they are doing what they are programmed to do.
IP: Logged
|
|
aiguy
Member
Member # 3736
|
posted 07. April 2007 15:03
LE,
quote:
Which Darwinian theories are you talking about? Neo-darwinism, in an effort to claim that evolutionary searches were not purposeful, proposed that evolution involved random variation and selection operating on expressed phenotypes. If variations are random, then the expected number of trials or iterations involved in finding a solution can by calculated from the areas of the solution space and the search area.
Please read this again:
In Darwinian evolution, the tester is the interaction of phenotype with the environment. In one sense, the generator is incapable of incorporating any information from the tester at all: No matter what happens, this information does not affect what genotype will be generated next. But the generator does incorporate information from the tester when the population is considered, since each incremental gain in reproductive advantage informs the generator to pick genotypes with the highest reproductive success as the next set of nodes in a massively parallel search. And of course there is no "goal node" in Darwinian evolution at all; it is not a targeted search. Every generation has some level of reproductive success, which might vary in the next generation in either direction.
quote:
If the actual speed or efficiency of finding a solution is materially faster than what would be predicted by a random search, then the variations must be non-random or designed. The variation generated reflect a 'knowledge' of the goal or a knowledge of how to find a solution in an efficient or intelligent manner.
Darwinian Evolution is not a random search. In a random search, each phenotype would be explored equally without regard to fitness. This is not what happens, obviously, since "fitness" means that more offspring survive to reproduce, so these phenotypes are searched preferrentially in the next generation.
quote:
Neo-darwinism, in an effort to claim that evolutionary searches were not purposeful, proposed that evolution involved random variation and selection operating on expressed phenotypes.
As I've already said, the variation is random, but this is not what determines the search strategy. Selection (the tester) provides information to the generator (reproduction) so that the search follows paths that lead to highest reproduction rates. Is this not clear?
quote:
I am not sure that it is terribly relevant to the discussion here, but clearly computers can not precisely process or express continuous functions or algorithms. My understanding that at least part of what Turing addressed was the ability to approximate continuous functions with discrete or discontinuous computer algorithms.
No, not relevant to the thread, but no, this is not what Turing was talking about. Turing machines compute continuous functions to any arbitrary degree of precision, and this issue is not relevant to the Church-Turing thesis.
IP: Logged
|
|
aiguy
Member
Member # 3736
|
posted 07. April 2007 15:07
Daniel,
quote:
If I design a machine and program it to "make decisions" based on inputs from the field and perform actions based on those inputs, is that machine "intelligent"? Or is it merely doing what it was programmed to do?
I would say it is both - how about you?
quote:
I'd argue that we do know what causes proteins to fold the way they do - they are doing what they are programmed to do.
In that case, we might as well say that we know what causes biological complexity. Nature was programmed to create it. And how was Nature programmed to create it? By another program. And how about that program? Well, if you'd like to speculate about ultimate causes, we've left the realm of science of course. But it's simpler to posit that the first program simply popped into existence all by itself than it is to posit that some other thing, about which we know nothing at all, popped into existence all by itself, and then this thing in turn created the first program.
IP: Logged
|
|
nosivad
Member
Member # 767
|
posted 07. April 2007 15:28
aiguy
Neither selection nor reproductive success ever had anything to do with evolution which was always emergent, scheduled and took place independent of the environment. For the vast majority of organisms their extinction was also preprogrammed. THAT is what has to do with the topic of this thread. Get it? Don't take my word for it. Consider the words of the greatest paleontologist since Cuvier.
"...the main features of the evolutionary trend were laid out right from the start with the abrupt, discontinuous production of the type, and with evolutionary potential being restricted right from the start to certain paths."
Otto Schindewolf, Basic Questions in Paleontology, page 360. The entire statement is in his italics for emphasis.
There has never been a role for chance in any aspect of either ontogeny or phylogeny.
Now consider the conclusions of the greatest Russian biologist of his day.
"Neither in the one nor in the other is there room for chance."
Leo Berg, Nomogenesis, page 134
You have your sources which I notice you do not mention and I have mine which I just did. Who are your sources or are your ideas entirely original?
"A past evolution is undeniable, a present evolution undemonstrable."
John A. Davison
[ 07. April 2007, 15:39: Message edited by: nosivad ]
IP: Logged
|
|
aiguy
Member
Member # 3736
|
posted 07. April 2007 16:02
nosivad,
quote:
Neither selection nor reproductive success ever had anything to do with evolution which was always emergent, scheduled and took place independent of the environment. For the vast majority of organisms their extinction was also preprogrammed. THAT is what has to do with the topic of this thread. Get it? Don't take my word for it. Consider the words of the greatest paleontologist since Cuvier.
Sorry, nosivad, but I don't see any connection at all between the topic and your point. Please read the OP, and tell us how your ideas help to characterize the concept of "intelligence".
quote:
You have your sources which I notice you do not mention and I have mine which I just did. Who are your sources or are your ideas entirely original?
Which ideas are you referring to? I will provide sources for whatever I've claimed as factual that you might find controversial. As for the rest of what I've said, I'm making arguments rather than stating matters of fact, and arguments need not be attributed to others in order to be assessed; they must be judged on their own merits.
IP: Logged
|
|
Melvin H. Fox
Member
Member # 1684
|
posted 07. April 2007 16:28
Hi aiguy,
Now we are getting somewhere.
We agree that fixed laws are at work and that the human MO is trial and error. We also agree that nothing else has yet been demonstrated to be at work. You claim that chance is also at work as a cause of human behavior. But, that is non-sense. Can we agree that since fixed law and trial and error can’t explain the whole of human behavior, then there must exist other causes, we have yet to demonstrate specifically, at work? You seem to acknowledge that human behavior is goal driven and that these goals are not consistent, to which I agree.
I suspect the issue of goals will be problematic in our discussion but is the above a good description of where we stand?
-Mel
IP: Logged
|
|
|
Printer-friendly view of this topic
