... as a computer scientist spending some of my life in the area of AI I can tell you this - there is a mathematical, hard science proof that IQ tests make no sense.

First of all - very few people on this plannet knows the true definition of intelligence, which is no more and no less a capacity for grouping objects that are similar - where similarity is defined as recognizing clusters of objects in a multidimentional space. Obviously we do not see it like this when we recognize to different objects as cars, our neural nets in the brain do all the math for us.

Anybody who knows and understands that definition sees an immediate problem with IQ tests - they do not measure intelligence, but merely a current setup of neural networks in a persons brain - there is a huge difference that is very hard to explain in this short post, but I will try.

If you take any neutral net that did not go through the learning process to set up its input/responce signatures (weights), its response to the input is always going to have a very low success rate. However with a sufficient training a neural net reaches its full potential, and only then we can say how good a given net is. Futhermore through providing new inputs neural nets constantly re-adjust themselves.

In short - inteligence is a capacity for quick and accurate readjustment of the neural networks to mach the expected "good" response. That is another thing - intelligence does not exist in the vaccum - the nets tune themselves to what is considered correct at the moment and that can change over time.

IQ tests merely measure a current level of response to an input, where correct answers are not known for the nets to tune themselves. Also that is why people get better over time with IQ tests.

Also, on a little mean note: no psychologist on this planet is smart enough to design even the tiniest brain consisting of 3 neurons. Why would you trust them to measure IQ of a human brain with trillions of neurons configured in billions overlapping and redundant of neural nets?

As a computer scientist spending some of my life in the area of AI I can tell you that your post does not remotely resemble a "mathematical, hard science proof" of anything. The validity of IQ tests is borne out by the fact that they predict success in school and life.

To speak in your terms, a neural net with more units will be able to learn more, and a brain with more neurons (made possible perhaps by a larger skull) will too.

Whatever the theory of intelligence behind IQ tests they have been proven to have a high correlation to various life-outcomes. For instance, if you pick ten molecular biologists and ten bulldozer operators at random, would you feel safe in predicting which group would have the higher median IQ? Also if you took ten convicts from the local jail and ten shoppers from the local mall. Or ten Slate readers and ten National Enquirer readers.

There is a basic error in your logic, that corresponds directly to what I have written before. You assume a current snapshot of the brain in question.

While in order to see who is really more intelligent, or at least better in solving logical puzzles in the IQ tests, you would need to take your bulldozer operators and scientists and train them for let's say 1 year in solving IQ puzzles. You will discover that some of the scientists score lower after that then some of the bulldozer operators.

qaz1231

Of course it does not - it would have to be 500 pages long and full of formulas.

Your expectation for my post to be anything close to "mathematical, hard science proof" makes me question, if you are a real scientist.

There is a consensus between researchers that IQ test are BS, and at best a good entertainment for geeks and if you do not know that – well, you are out of the loop.

Size of the net does not matter – its quality does – you should know that if you are an AI researcher as you claim to be.

Oh, by "scientists" I mean the real ones. Psych's are not even remotly close to being real scientists, with their "hit the black box and see what happens" methodologies.

paxterminus, I know several molecular biologists and bulldozer operators. I don't think any amount of training could get the bulldozer operators (or me for that matter) up to the IQ level of the molecular biologists. BTW, do you think Einstein just had a good mentor at the patent office?

I'd also accept a citation to a "mathematical, hard science proof", or even a brief sketch of how it works.

There are things you can change in a network that will be more important than adding units, but, all else being equal, more units allows you to represent (and thus learn) more discrete patterns. This is not even debatable.

>You will discover that some of the scientists score lower after that then some of the bulldozer operators.

Any empirical evidence to support this? If you did see improvement in bulldozer operator scores it would be in test-taking ability, not general intelligence. The scientists are scientists because they've tested high *without* being trained on test-taking tips. The real question is whether if you trained the bulldozer operators on calculus they'd be better than the scientists. I'd be happy to put money against that proposition.

E. O'Neal, what you say actually cuts against the reliability of IQ tests actually measuring anything. If IQ tests were actually accurate measures of intelligence as completely disconnected from learning and culture and experience, then you'd expect to find lots of diamonds in the rough, as it were--individuals who, due to cultural or monetary factors, ended up without much education and in careers where their superior intelligence was not used.

If your bulldozer operators always score lower on intelligence tests than your molecular biologists, then either our society is a perfect meritocracy or IQ tests don't measure what they claim to.

RobMac, your point is why I said take ten of each group and compare the medians. There are" diamonds in the rough" like the famous longshoreman philosopher Eric Hoffer. There is evidence though that our educational system has become more efficient at identifying these gifted people than in the past, and the odds of finding a blue collar worker with a 130+ IQ are less today than fifty years ago.

This is absolutely true. No one can yet understand intelligence, but at the same time, these tests must be measuring something. If we can find out what that something is exactly, the entire debate about their usefulness could be changed tremendously. This might actually be a more productive method of examining intelligence; a round-about manner may yield far more useful results.

E. O'Neal:

. BTW, do you think Einstein just had a good mentor at the patent office?

Don't you know that Einstein just had good self esteem ? That's why he was able to contribute that much to physics. He wasn't smarter than any of us here ! Society expected him to be a genius, instilled self esteem into him and there you go ! physics genius. I don't care what racis and classist science says about some people being smarter . Anyways, those smarter people can't dance well so wat are they good for anyways. Oh and they can't survive naked in the australian outback on their own so what good are they ?

Academics have developed certain skills that manual laborers haven't. But that misses the real point...

Just as I can always predict the past, I can come up with a test the predicts who is 'smart' by some vague societal standards. This then becomes a self fulfilling prophecy. Is a Harvard student smart because he met the bar for a Harvard student, if so, isn't he now the bar.

I cannot create a test to predict who will come up with novel practical thoughts, which are really the only standard for intelligence. If it's not novel then it's just rote operations in the brain that a computer can do.