I googled "inductive intelligence" & hit on John W Edser' post of the last year. I share his frustration with IQ measures & came up with an explanation of why real-time tests can't capture a high-level generalization ability, caused by a longer-range induction. Basically, induction is a search for matches/patterns, & deduction is an interactive projection of these patterns. In neocortex there's a tradeoff between the number of basic units: minicolumns, & the range of connections between them. Longer range connections allow for range^2 number of possible comparisons, which would slow down the learning process but improve quality of matches & generality of the resulting patterns. Correspondingly reduced number of minicolumns will decrease represented detail/precision. So, this is a speed & detail, vs depth & generality tradeoff, & real- time tests are biased for the former & against the latter. The tradeoff is somewhat ambiguous in terms modern societal utility: - On one hand, speed & precision was far more important for survival "in the wild", which probably explains why apes likely have a photographic memory, superior to humans. - On the other hand, the functional differentiation of a modern society probably rewards specialization & precision more than higher generalization ability on the opposite end of cognitive diversity spectrum.
cogno...@gmail.com wrote: > I googled "inductive intelligence" & hit on John W Edser' post of the > last year. I share his frustration with IQ measures & came up with an > explanation of why real-time tests can't capture a high-level > generalization ability, caused by a longer-range induction. Basically, > induction is a search for matches/patterns, & deduction is an > interactive projection of these patterns.
Most work in AI is based on a HUGE assumption, i.e., the assumption that rationality can be reduced to logic. This runs counter to Godel's logical discovery that mathematical rationality can never be found in its entirety with a formal (e.g., logical) system.
Here's a line of argument that suggests that "reasoning" is a bit of a different animal from the one characterized by the Kahneman, Tversky etc. crowd who are still operating on the logic = rationality assumption (IMHO). Its from one of my papers presented at a number of conferences over that past few years in which I draw a distinction between higher cognition and lower cogntion as follows:
[begin quote]
Higher Cognition (class) (analogical “reasoning”)
Associations arising from 'the cognition of abstruse similarity and difference', e.g., electricity is like water flowing in a pipe. Highly individualized.
Lower Cognition (class) (conditioning)
Associations arising from 'the cognition of obvious similarity and difference', e.g., this A + B sequence is like one's previously observed. Inter-environ- mental individualization, intra-environmental iso- morphism. Likely progen- itor of higher cognition.
Higher Cognition:
Not uncommonly, deductive syllogisms such as ‘Socrates is a man, all men are mortal, therefore Socrates is mortal’, are offered as examples of reasoning. This is not how I am employing the term in the phylogeny, which is why it appears in quotation marks. I mean for it to refer to whatever thought process lies at the heart of ampliative inference, a process often associated with ‘Aha!’ or ‘Eureka!’ experiences, but commonly falling below the threshold of an identifiable event in which much, if not most, of the processing is not introspectively available. Even so, by applying a bit of the abstraction and generalization sanctioned by our methodology and in contrast to the Nisbett and Wilson approach to the study of “higher order, inference based responses”), I believe enough is available for us to make a reasonable guess that the cognition of similarity and difference (analogical/metaphorical “reasoning”) does most of the heavy lifting. But then I am hardly the first introspectionist to arrive at that conclusion:
‘All kinds of reasoning consist in nothing but a comparison and a discovery of those relations either constant or inconstant, which two or more objects bear to each other’ (Hume, 1739).
Lower Cognition:
My unorthodox definition of conditioning as ‘the cognition of obvious similarity and difference’ stems from my unorthodox definition of reasoning as ‘the cognition of abstruse similarity and difference’ which, when combined with the former, offers a number of explanatory advantages:
1. It allows for continuity between the two concepts and, as such, allows for an appreciation of how “reasoning” might have evolved from conditioning. In this view, the ability to understand electricity by comparing it to how water flows in a pipe is just an extension of the process that underlies an organism’s ability to understand a currently observed A + B sequence (e.g., Pavlov’s dogs) by comparing it to ones previously observed.
2. It allows one to forego syllogistic deduction (‘Socrates is a man…”, etc.) as a paradigm for reasoning in that, based on the analogy with conditioning, concluding that Socrates is mortal can be viewed as analogous to a conditioned mouse remembering it must go left at the fourth fork in a maze. In much the manner the mouse’s recollection would be construed as more a manifestation of conditioning that has already occurred, we might also conclude that deducing Socrates is mortal is more a manifestation of reasoning which has already occurred, and perhaps closer to remembering than reasoning, at least in an ampliative sense of coming to a deeper understanding of the nature of reality, and thereby serving to produce a net increase in one’s rationality.
‘If analogy were merely a special variety of something that in itself lies way out on the peripheries, then it would be but an itty bitty blip in the broad blue sky of cognition. To me, however, analogy is anything but a bitty blip -- rather, it’s the very blue that fills the whole sky of cognition – analogy is everything…’ (Hofstadter, 2001).
3. It allows for a naturalistic indeterminism in that one can surmise that once an event sequence or feature has become cognized it is easy to appreciate how one might then have the option of following the sequence or conforming to the feature or not, and thereby becoming less determined by it, i.e., aware of more options than prior to the cognition. Another way of saying this is that it lends itself to the suspicion that there might well be an inverse correlation between ‘being cognizant’ or ‘being rational’ and ‘being determined’.
[end qutoe]
'Rehabilitating Introspection: A Procedure for a First Person Psychical Science'
> Most work in AI is based on a HUGE assumption, i.e., the > assumption that rationality can be reduced to logic. This runs counter > to Godel's logical discovery that mathematical rationality can never > be found in its entirety with a formal (e.g., logical) system.
Logic in general is consistency, there's no virtue in being inconsistent. But I find conventional formal logic irrelevant, my logic is Bayesian beyond Bayesian.
> My unorthodox definition of conditioning as ‘the cognition of > obvious similarity and difference’ stems from my unorthodox > definition of reasoning as ‘the cognition of abstruse similarity > and difference’
"Abstruse" is a matter of degree, not of a kind. "Conditioning" refers to reflexes, not cognition. The difference is that reflexes work in isolation, they shortcut the cortical generalization hierarchy & produce an immediate "value judgement".
> 3. It allows for a naturalistic indeterminism in that one can > surmise that once an event sequence or feature has become cognized > it is easy to appreciate how one might then have the option of > following the sequence or conforming to the feature or not, and > thereby becoming less determined by it, i.e., aware of more options > than prior to the cognition. Another way of saying this is that > it lends itself to the suspicion that there might well be an > inverse correlation between ‘being cognizant’ or ‘being rational’ > and ‘being determined’.
> - On the other hand, the functional differentiation of a modern > society probably rewards specialization & precision more than higher > generalization ability on the opposite end of cognitive diversity > spectrum.
I believe that in human mating rituals, higher generalization ability may be more important than specialization and division. When two people with different classes in society compete, this is an asymmetric arms race. The resource being fought for is likely to be the same in all battles. Food is food, a winning strategy should be repeated exactly. The specialization of each class gets emphasized. Therefore, generalization may kick the loser out of the competition. The next battle may be over the same resource. So specialization pays for the winner. In competition of two males for the same female, the competition is symmetric. Generally both males are in the same class, and if not they are at least competing for a variable resource. Therefore, specialization is likely to kick the suitor out of the game. Girls are different. The next girl will probably want different things than the first. Same for female on female competition. The next boy may be different. So generalization pays.
Lorentz <drosen0...@yahoo.com> wrote:- >> cogno...@gmail.com wrote: >> - On the other hand, the functional differentiation of a modern >> society probably rewards specialization & precision more than higher >> generalization ability on the opposite end of cognitive diversity >> spectrum. >I believe that in human mating rituals, higher generalization > ability may be more important than specialization and division.
JE:- I think that is true of any ritual. A ritual can contain a useful correlation such that if you faithfully repeat that ritual you may be able to repeat a correlation and be able to predict the future. This has enormous survival value and can be observed in animal systems in which cognition is not present. To be able to make up just a myth which can also contain a subconscious correlation requires inductive intelligence. Of course, turning alchemy into chemistry required falsifiable inductions which do a lot more than just contain a useful correlation.
It makes sense to me to suppose that within our species inductive belief systems evolved long before theories did because early tribes required diverse belief systems to bond larger and alrger groups. The fact that these belief systems may also contain and can therefore pass on a useful correlation _as a tribal based meme_, remained a bonus. From this our theory building ability evolved as belief induction became more and more subject to deduction as a critical test to refutation.
cogno...@gmail.com wrote: >>Most work in AI is based on a HUGE assumption, i.e., the >>assumption that rationality can be reduced to logic. This runs counter >>to Godel's logical discovery that mathematical rationality can never >>be found in its entirety with a formal (e.g., logical) system.
> Logic in general is consistency, > there's no virtue in being > inconsistent.
I think of logic as simply the order we have cognized in the manner in which we cognize order. But for me, this process is itself largely ANA-logical in nature, which is why it's so hard to computationalize.
> But I find conventional formal logic irrelevant, my logic is Bayesian > beyond Bayesian.
Bayesian logic isn't formal?
>>My unorthodox definition of conditioning as ‘the cognition of >>obvious similarity and difference’ stems from my unorthodox >>definition of reasoning as ‘the cognition of abstruse similarity >>and difference’
> "Abstruse" is a matter of degree, not of a kind. > "Conditioning" refers to reflexes, not cognition. > The difference is that reflexes work in isolation, > they shortcut the cortical generalization hierarchy > & produce an immediate "value judgement".
Yes. Like similarity and difference themselves, abstruse and obvious are pretty much in the eye of the beholder. And yet, this something that is pretty much in the eye of the beholder is at the very heart of most ampliative inference, don't you agree?
"All forms of reasoning are nothing but comparing". (Hume)
>>3. It allows for a naturalistic indeterminism in that one can >>surmise that once an event sequence or feature has become cognized >>it is easy to appreciate how one might then have the option of >>following the sequence or conforming to the feature or not, and >>thereby becoming less determined by it, i.e., aware of more options >>than prior to the cognition. Another way of saying this is that >>it lends itself to the suspicion that there might well be an >>inverse correlation between ‘being cognizant’ or ‘being rational’ >>and ‘being determined’.
> If you don't like determinism, tough.
Perhaps. But that hardly constitutes a counter to the rationale I have offered for why I suspect there might be a chink in your metaphysical armor.
> I think of logic as simply the order we have > cognized in the manner in which we cognize order. > But for me, this process is itself largely > ANA-logical in nature, which is why it's so > hard to computationalize.
Yes, the Induction process, formalized by Bayesian (probabilistic) logic, although I add a whole new dimension to it by considering "partial" occurence, or the degree of match (similarity & difference): http://scalable-intelligence.blogspot.com/2008/04/intelligence-patter... It's only hard till you know how to do it.
> Yes. Like similarity and difference themselves, abstruse > and obvious are pretty much in the eye of the beholder. > And yet, this something that is pretty much in the eye > of the beholder is at the very heart of most ampliative > inference, don't you agree?
Not the way I treat it, my definitions of both similarity & complexity (order of generalization) are perfectly formal, see the link above. You do need intelligence to recognize intelligence, but this doesn't make the process of recognition any less formal.
> "All forms of reasoning are nothing but comparing". (Hume)
I'd add projection :).
> >>3. It allows for a naturalistic indeterminism in that one can > >>surmise that once an event sequence or feature has become cognized > >>it is easy to appreciate how one might then have the option of > >>following the sequence or conforming to the feature or not, and > >>thereby becoming less determined by it, i.e., aware of more options > >>than prior to the cognition. Another way of saying this is that > >>it lends itself to the suspicion that there might well be an > >>inverse correlation between ‘being cognizant’ or ‘being rational’ > >>and ‘being determined’.
> > If you don't like determinism, tough.
> Perhaps. But that hardly constitutes a counter to the rationale > I have offered for why I suspect there might be a chink in your > metaphysical armor.
Just because Reaction is separated from Action by many levels of generalization doesn't make the process any less deterministic. The only "metaphysical armor" I have is my concept of Meta-Evolution: http://scalable-intelligence.blogspot.com/2008/04/entropy-evolution.html I consider Cognition itself a higher form of evolution in this framework. Boris.
