|
Author
|
Topic: Evolutionary Intelligence
|
Christopher M. Langan
Member
Member # 264
|
posted 08. April 2003 16:19
There is much theoretical controversy regarding the nature of intelligence, largely because prevailing scientific models – having been designed to preclude subjective contamination in the gathering and evaluation of objective data - support the existence of neither intelligence nor any other subjective attribute. Thus, it has often been asked whether intelligence (as in "Intelligent Design") is really a measurable, empirically verifiable quantity.
The science of psychology, as practiced in laboratories, hospitals, clinics and educational institutions around the world, answers in the affirmative, and offers a very great deal of supporting data in the form of IQ test scores. IQ tests consist of problems to be solved, usually within time constraints. Correct solutions possess utility in the form of higher test scores and all utile correlates thereof. Despite questions regarding the validity of IQ testing in this era of political correctness, these measures and the data they provide exhibit a very high degree of statistical integrity.
Survival is a problem for every organism and species, and biological fitness is directly analogous to utility. That is, survival has positive utility for that which survives (while optimality is unnecessary for survival and may in any case be undecidable, survival itself is objective and observable). To this extent, "survival of the fittest" (and all of its more refined variants) refers to the solution of a problem, namely to solving for fitness as a condition of survival, and evolution is thus analogous to a problem-solving procedure.
This analogy is better than one might think. When it comes to evaluating aggregate data, the lab-coated, by-the-book scientists trained in clinical and experimental psychology typically do not concern themselves with the subjective nature of what their instruments measure; when dealing with populations, they take what a logician or systems theorist would call a "black box" approach to intelligence, concerning themselves solely with the numbers thereby generated. In effect, intelligence is operationally defined as that property of test subjects which best correlates with IQ test performance.
Although psychologists are encouraged to take account of subjective factors, they are limited by the physicalist models dominant throughout the sciences, and ultimately lack the means to make a clear subjective-objective distinction. In this respect, they resemble evolutionary biologists, from whom they differ only in their tacit acknowledgment of the subjective dimension of human reasoning and emotions (which each of them can in any case "observe internally"). As far as concerns their (highly replicable) data, there is no such distinction.
If we take a black-box approach to evolution, this experimental paradigm can be applied to evolutionary biology up to isomorphism. That is, the environment presents biological entities with problems in the form of stressors, and nature responds by modifying the entities [in] ways that solve these problems:
Survival problem --> time-constrained solution process --> fitness solution --> observation and statistical evaluation
So the question is, why can we not say that nature possesses something analogous to an "IQ", and given its admirable success rate in solving for occupation and survival in various ecological niches, that it is "objectively intelligent" in the same way that a high-scoring test subject is objectively intelligent?
Let’s take a brief look at the kinds of objection that a critic would be forced to make in order to discount this experimental paradigm, which is really just the statistical paradigm at work in a certain objectivized scientific context.
For example, a critic might say: "But we can see human beings filling out the tests, and we know them to be intelligent!" However, the intelligence of test subjects is precisely what is to be determined by testing them; as far as the statistics are concerned, they might as well be AI machines.
A critic might also say: "A logical extension of this viewpoint leads directly to the realm of the inanimate, and would thus lead to the absurd conclusion that inert physical structures and processes display intelligence!" But again, from a scientific perspective, whether or not nature is "intelligent" is the point to be determined. In any case, no reference to any subjective-objective distinction is possible without a detailed model of the distinction.
ID critics like to point out that ID lacks a model for intelligence. But so does the scientific mainstream, and yet this does not stop scientists from observing and measuring what they call "intelligence" by taking a black-box approach to it. Less formally, ask any scientist whether he or she possesses some degree of intelligence, and one will almost certainly receive a "yes" for one’s trouble; ask that same scientist whether he or she possesses or even knows of a detailed model of intelligence, and one will get a different answer entirely.
Is this a scientific argument to the effect that nature is intelligent? Or should we simply put an end to terms like intelligence and fields like clinical psychology?
[ 08. April 2003, 20:51: Message edited by: Christopher M. Langan ]
IP: Logged
|
|
Rex Kerr
Member
Member # 632
|
posted 08. April 2003 17:59
I normally associate intelligence with goal-directed behavior driven by internal models of the actual and desired state of the world.
I use the word "effective" to describe the fact that system X can accomplish task Y. (X need not be trying to or designed to do Y in order to be effective; "effective" is my classification of the process.)
Is there a way in which your suggested definition of intelligence goes beyond the definition of effectiveness?
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 08. April 2003 18:28
I have two questions of clarification. First, I'm not familiar with the phrase "statistical integrity." I know something about statistical reliability and about validity, but not "statistical integrity." Could Langan explain what that means in context?
Second, Langan wrote quote:
So the question is, why can we not say that nature possesses something analogous to an "IQ", and given its admirable success rate in solving for occupation and survival in various ecological niches, that it is "objectively intelligent" in the same way that a high-scoring test subject is objectively intelligent?
Is this looking just at the successes ("admirable success rate)", i.e., considering just one tail of the distribution of attempts while ignoring failure rates both at the individual and population levels? Is the current level of "admirable success" better regarded as the result of intelligent problem-solving or the residual consequence of a long history of a few successes interspersed among a much larger number of failures? In other words, when black-boxing nature's IQ, what is the reference set for the relative intelligence measurement? (Human IQ measures are, after all, relative measures rather than absolute measures. IQ scores are pretty certainly ordinal scales, are half-way decent interval scales within some range, but are certainly not ratio scales.) Would we call a person "intelligent" who takes tens of thousands of tries to solve a problem? How do prior failures (= wrong solutions) play into the measurement operation/estimation? In a human IQ test one gets just one try, after all.
I should say that subject to a satisfactory answer to Rex Kerr's question about meaning beyond "effective," this might be an interesting way of approaching design in nature, if not intelligent and purposive design. I'm wary of the underlying "evolution-as-search-for-solutions" formulation, but with suitable caution that's not an insuperable objection.
RBH
[ 08. April 2003, 18:29: Message edited by: RBH ]
IP: Logged
|
|
Christopher M. Langan
Member
Member # 264
|
posted 09. April 2003 14:53
An effective procedure is one which admits of mechanical implementation with respect to an abstract machine model, e.g. a Turing machine. Complexity and utility are irrelevant; the issue is whether there exists a series of discrete steps, defined on the model in question, leading from some input to some output. In particular, the utility of the programmer is strictly a side issue. Solving problems on an IQ test goes beyond the realm of the merely effective because there is a value attached to test performance, and so for surviving in the wild. Indeed, a real “problem” always relates in some way to the utility of that by which it is recognized as a problem. If it is in no way a matter of utility – if nothing benefits in any way from its solution - then when you come right down to it, it’s really no problem at all.
On the other hand, any problem that deserves to be called a “problem” is related to its solution by a more or less complex transformation of its initial (unsolved) state into its final (solved) state, and this transformation is associated with a net gain of utility. This means that the result of the transformation, i.e. the solution of the problem, qualifies as a "goal" for the entity with respect to which this utility is defined, and utility specifies and drives the transformation. Survival fits this description because it possesses utility for those entities which achieve it, and achieving it often requires the complex and tightly-constrained adaptive transformation of an initial state consisting of a stressful environment-organism relationship.
The number and distribution of possible transformations of such a relationship, i.e. the contingency, determines the discriminativity of the problem, i.e. the extent to which it can be used to measure problem-solving ability or “intelligence”. So where utility corresponds to specification and complexity is defined as usual, the psychology of cognitive testing involves a sort of “specified complexity”, and because it is in certain ways analogous to biological evolution, we have a logical case for admitting the concept of intelligence, and to some extent the methodology of intelligence testing, to evolutionary biology.
Integrity means "soundness", which naturally implies validity and reliability. So "statistical integrity" refers to the soundness of statistical methodology, including experimental design, data gathering and analysis, and statistical inference as applied in clinical and experimental psychology. Thus, when statistically significant relationships are detected by means of intelligence testing, they usually stand up to scrutiny with respect to methodology and prediction.
As for IQ being a relative measure, this applies to comparisons rather than to tests and test items. Whether or not a problem has been solved to a given level of accuracy within operative time constraints is an absolute (yes or no) determination, and likewise for one’s raw score on a test. One could simply answer questions at random over many trials and take the average result as a baseline score, working from the space of all possible attempted solutions within the given constraints. The result would be a determination of absolute rather than relative problem-solving ability (at least with respect to the test in question).
The point, of course, is that although one can split hairs about where to set the points of reference and define the bounds of relevance, it would be difficult if not impossible to make a scientific case that the psychometric paradigm and its associated concepts have no place in evolutionary biology. In principle, intelligence and mental causation have as much bearing here as they have in psychology; the kind of biological data required, how to gather it, and how much of it to gather are secondary questions. Regardless of any difficulties attending the answers to these questions, there are no scientific grounds for excluding intelligence as a causal factor in evolution given the potential for an operational definition analogous to that employed in psychology.
[ 09. April 2003, 15:03: Message edited by: Christopher M. Langan ]
IP: Logged
|
|
RBH
Member
Member # 380
|
posted 09. April 2003 17:34
First, note that I am not arguing against this approach, at least not yet. ![[Smile]](smile.gif) I can vaguely see that psychometrics might be useful, or at least how some of its concepts and techniques might be. Nor am I interested in "splitting hairs." I'm trying to understand the suggestion in order to see how the analogy might be applied and if it is more than just an analogy. At this moment, given my imperfect understanding, I'm perfectly willing to entertain the notion of 'intelligence' in evolution. (I won't raise the issues associated with "purpose" and/or "intention" in this context. Yet.)
Langan wrote quote:
As for IQ being a relative measure, this applies to comparisons rather than to tests and test items. Whether or not a problem has been solved to a given level of accuracy within operative time constraints is an absolute (yes or no) determination, and likewise for one?s raw score on a test. One could simply answer questions at random over many trials and take the average result as a baseline score, working from the space of all possible attempted solutions within the given constraints. The result would be a determination of absolute rather than relative problem-solving ability (at least with respect to the test in question).
Are two notions being intermixed here? IQ is a relative measure, with raw scores being normed on age-stratified populations so as to arrive at the ratio that is IQ. Raw scores are not "relative" in that sense, being a simple count of problems solved within some time limit. But as soon as those raw scores - counts - are normed, normalized, and cast into distributions with percentile ranks assigned meanings, they are relative in the sense that I have used the term. As Langan notes, both raw scores and the IQ ratios depend on correlations with external referents - success in school being the primary one for IQ - for their utility and even their interpretation.
In the situation of evolution, we have one instance - life on earth - in our sample. I'm having trouble understanding how that one instance's effectiveness/success (or lack thereof, in the case of the multitude of trials evolution has essayed, most of which failed sooner or later) can be evaluated, validated, tested for reliability,etc., as one can do in the case of IQ. Can that be explicated? I guess the question is about the unit of analysis. What is it that can be said to have an "evolutionary IQ"?
RBH
[ 09. April 2003, 17:36: Message edited by: RBH ]
IP: Logged
|
|
|
Printer-friendly view of this topic