IQ Tests & Artificial Life: Can Robots be More Intelligent Than Men?

[ryokan]

IQ tests are designed in basis to perception of relationships among forms, words, numbers ...
A computer could be programmed to solve all the questions of any IQ test. Two alternatives result.
a) A well programmed computer could be more intelligent than man. Robots could be more intelligent than men.
b) IQ tests are inefficient to measure intelligence.

 

[humanino]

My opinion is a). Even : that will be tough for us to accept it when computers will compose poems I hope to see this before I die.

 

[Adam]

Sure, you can programme a computer with all the answers of a specific test. You can be pretty sure it will perform many of the tests that make up IQ testing better than a human will. On the other hand, a computer will have a hard time explaining why it is right or wrong to kill babies. At the moment, the best it can do is print out a programmed response. But I'm working on that. :P

 

[amwbonfire]

b). IQ tests don't measure (fairly/accurately) all the different aspects of intelligence.


It's hard (impossible?) to program emotions into AI. And our emotions, and experiences, are what govern our thought process the most. Sure, we can program AI that thinks and acts based on past experiences, but can we make it feel?

It'd be fun though, to have a computer that you can make fun of, and it cries. Or you could be friends with it.

 

[ryokan]

b). IQ tests don't measure (fairly/accurately) all the different aspects of intelligence.

Then, do IQ tests have any value?

 

[Nereid]

Then, do IQ tests have any value?

They provide employment for psychometricians?

 

[Zantra]

Computers can be programmed to to have a higher degree of accuracy than a human, but you can't yet give a computer all the sum total of experience that a human mind holds. I believe it's been proposed that the sum of the human intellect can be contained in 15 petabytes of storage space? Then you have the learning curve. AI algorythms from what I know (I'm no AI guru) can only approximate the most simplistic learning abilities in comparison to the human mind. Old blue beat the world chess champion, because this was it's single devoted purpose. But could it generate a unique piece of art? or even serve as an email server? Not unless it was reprogrammed. I think so far we lack the hardware capabilities to to fully approximate human intelligence. We will be there someday, but we aren't yet. In order to approximate human intelligence a computer has to be able to draw on a vast and very broad store of knowledge. Only then will it have this capability.

 

[BlackVision]

IQ tests are designed in basis to perception of relationships among forms, words, numbers ...
A computer could be programmed to solve all the questions of any IQ test.

Not necessarily. And all the answers? Only if you preprogram the computer to know the answers in advance. I'm not quite sure how well any computer currently will do in solving a random IQ question but I would guess not that well.

a) A well programmed computer could be more intelligent than man.

Correct.

Robots could be more intelligent than men.

Yes robots can be more intelligent than men. Computers are nowhere near the level of human intellect currently but Stephen Hawking and other scientists have predicted that computers will eventually pass the intellectual level of humans.

b) IQ tests are inefficient to measure intelligence.

Well this statement has been disproven a billion times by countless researches so we do not have to worry about this one.

 

[BlackVision]

They provide employment for psychometricians?

They do far more than that. Research after research showed high levels of correlation between IQ and numerous factors. Including academic success. Economical success. Probability of going to prison. Probability of a happy marriage. Probability of having a illegitimate child. Probability of being a drug addict. The list goes on and on. IQ certainly isn't everything but it seems to be the single greatest factor in one's life. Every task is said to have some level of g.

 

[BlackVision]

But could it generate a unique piece of art? or even serve as an email server? Not unless it was reprogrammed.

Not necessarily. There have been current successful studies of AI allowing robots to learn on their own. Gain new knowledge through trial and error like humans and animals do.

I think so far we lack the hardware capabilities to to fully approximate human intelligence. We will be there someday

Definitely.

 

[Smurf]

But the human mind is also developing, by the time we create AI to match our current Intelligence, will we (at least the brighter among us) not be even smarter?
Will our inventions ever surpass us?


-Ruler of the Universe,
Smurf

 

[ryokan]

I'm not quite sure how well any computer currently will do in solving a random IQ question but I would guess not that well.

Why?
Is there a limit to computation?
Or
Is there a limit in IQ questions?

And... What would be a random question?. One question randomly taken from a finite sample? Or infinite?
For example, does chess play involve a series of random questions?

 

[ryokan]

For example...

They do far more than that. Research after research showed high levels of correlation between IQ and numerous factors. Including academic success. Economical success. Probability of going to prison. Probability of a happy marriage. Probability of having a illegitimate child. Probability of being a drug addict. The list goes on and on. IQ certainly isn't everything but it seems to be the single greatest factor in one's life. Every task is said to have some level of g.

What is the greatest correlation factor observed?

 

[hypnagogue]

IQ tests are designed in basis to perception of relationships among forms, words, numbers ...
A computer could be programmed to solve all the questions of any IQ test. Two alternatives result.
a) A well programmed computer could be more intelligent than man. Robots could be more intelligent than men.
b) IQ tests are inefficient to measure intelligence.

It would be inordinately difficult to create a program that could perform well on an IQ test. This is not an issue of hardware; it's an issue of software. We have the wherewithal to create a computer that rivals the brain in memory and processing power-- actually harnessing all that power to make something as intelligent as the brain is another story. Probably the biggest difficulty in this regard would be simulating human common sense. IQ tests seem to ask a lot of questions that require common sense knowledge and judgment (eg Which of these animals is least like the other: horse, dear, etc...), both of which are notoriously difficult to capture with AI.

Intelligence is a relatively nebulous concept. If we don't have a good grasp on what we mean by the word, it's even more difficult to judge to what extent a given test is really measuring it. We can be confident that IQ tests measure at least one thing: how well a given person can perform on an IQ test. Generalizing that result is a tricky business.

One thing that seems clear is that comparisons of IQ scores across human subjects would be much more telling than comparisons of IQ scores across human and artificial subjects. At least when comparing humans, we have the basis of a common biology and environment to work on. Of course there are variances, but these are negligible compared to the kinds of differences that would obtain between the biology and environment of a human and the makeup and environment of a computer program. Basically, I don't think we could create a meaningful comparison between a human and computer program on the basis of an IQ score unless the computer program was essentially a reverse-engineered human brain in the same kind of environment (ie same kind of sensory inputs over a long period of time).

 

[selfAdjoint]

But I'll bet you could write software that could do Raven's Progressive Matrices. They are as far removed from what we all learned as todddlers as AI is.

 

[hypnagogue]

But I'll bet you could write software that could do Raven's Progressive Matrices. They are as far removed from what we all learned as todddlers as AI is.

I don't know much about Raven's Progressive Matrices, but I looked it up and it appears to be a test of visual pattern recognition. As such, it falls under the general category of cognitive tasks that humans do well but AI classically hasn't.

I'm not arguing that a program couldn't perform well on such a test in principle-- just that, in practice, we humans have had a difficult time at creating programs with the proper software to perform such tasks. Much of our fuzzy, common sense reasoning (if not our actual common sense knowledge bases) appears to be based on the same kind of parallel, associative computing networks that are responsible for general pattern recognition.

 

[ryokan]

IQ tests seem to ask a lot of questions that require common sense knowledge and judgment (eg Which of these animals is least like the other: horse, dear, etc...), both of which are notoriously difficult to capture with AI.

And some background of knowledge. IQ tests wouldn't be useful with a very intelligent and illiterate person.

 

[ryokan]

What's about Deep Blue?
Could be intelligence of chess players be measured in basis to their success when playing chess?

 

[Nereid]

But I'll bet you could write software that could do Raven's Progressive Matrices. They are as far removed from what we all learned as todddlers as AI is.

I second that. Better yet, the software would find all the instances where the 'correct answer' isn't (or that there are more than one), and that software could do all the tests in, effectively, a few seconds, to the level of the most difficult exercises. Further, except (maybe) for the image processing input stage (which corresponds to the non-g part of the 'human processing' system, i.e. the eyes+the low level visual system in the brain, so is irrelevant for the 'intelligence' part of this discussion), I'd bet at least a dozen PF members could come up with the core algorithmic descriptions, given no more than a month or two of effort (the conversion to a robust program that actually works would surely take more than a few months' of effort).

 

[selfAdjoint]

Software to do Raven Matrices

Yes indeedy, from a wouldacoulda standpoint, what a neat app! Paging all software mavens!

For starters you could have it construct Where's Waldo(c) panels.

 

[ryokan]

What would be a possible test for IQ lacking of a possible algorithmic approach?

 

[Tigers2B1]

While you're at this 'program writing' you'd better add to that machine the ability to understand word analogies, picture analogies, metaphors, and how word order changes the meaning of sentences or words change in context (e.g. "ring the city" verse "ring the bell") and a whole, whole lot more. Teaching a computer to play chess or a camera to take a picture does make one similar to a brain and the other similar to an eye. So who volunteers?