Human vs Computer Intelligence

DrLudicrous writes "The NYTimes is running an article regarding tests devised to differentiate from human and computer intelligence. One example are captchas, which can consists of a picture of words, angled and superimposed. A human will be able to read past the superposition, while a computer will not, and thus fails the test. It also goes a bit into some of Turing's predictions of what computers would be like by the year 2000."

Accessibility issues?

[Ryu2]

These tests seem to be all visual in nature. Could this be a point of contention on the part of blind/visually impaired users of web sites?

Or alternatively, are they perhaps working on, say, a audio version? Wonder how would that work.

And why bother

[DunbarTheInept]

And furthermore, if what you need for the task at hand is a machine that behaves and thinks in every way just like a human being, then just hire a human being to do it.

It's the differences between computers and humans that make computers so damn useful. Tell a human to add up a list of 200 numbers and he'll likely take a long time, and get the wrong answer because humans suck at repetative boring tasks beyond the limit of their attention spans.

Can't really mimic human intelligence

[TheWhaleShark]

While it would be nifty to have a very human AI, I somehow doubt that we could truly ever replicate human intelligence.

The major roadblock is that a computer can only respond in ways that it has been programmed to do so. While you can code incredibly complex AI algorithms and simulate an incredibly complex level of intelligence, the fact remains that a computer invariably operates along rigid pathways.

It can be argued that human thought is nothing more than a complex series of chemical reactions, but there is far less rigid logic involved in human thinking. Indeed, we're still not entirely sure just HOW we think.

Never say never, but I don't think we'll be seeing a truly human AI before any of us is dead.

Re:I failed!

[roseblood]

You failed? You don't want to know how poorly I did with my contacts out. I figure the makers of the test regard those with poor/no eyesight as being of less than human intelegence.

It won't work...

[Quaoar]

Computers can be specifically programmed to solve puzzles such as a this...if a test arises that supposedly tests for "human" intelligence, humans can simply modify the code so that it can solve that sort of puzzle.

That's what Gary Kasparov was complaing about when he played against Deep Blue the first time...there was a whole team of IBM programmers modifying the code during the game to specifically counter Kasparov's playing style. It wasn't a reflection of machine intelligence, it was an example of human adaptation imposed upon Deep Blue.

tech econ boost?

[Tablizer]

Amazing all this horsepower and research just to combat spam. Just might be the boost we need to get tech spending going again. A never-ending cat-and-mouse game where the cats and mouses get bigger and bigger. This racket is almost as good as the dot-com racket. I don't like spam either, but I miss real paychecks.

The first true AI machine might be spam catcher. Spamminator 2000!

Poll Stuffing on Slashdot

[Zordak]

The CAPTCHA website [captcha.net] (how do you pronounce that, anyway) has a list of possible applications of CAPTCHA. The first mention is online polls, and recalls an event in 1999, when Slashdot (they use http://www.slashdot.com [slashdot.com] for some reason) had a poll for the best graduate CS curriculum. Carnegie-Mellon and MIT wrote competing poll-bots that stuffed the poll boxes. The point was supposed to be that a CAPTCHA would have prevented this. In my opinion, however, this was probably the most accurate Slashdot poll ever. Obviously, MIT wrote the better poll bot, since it stuffed more votes, and they didn't even start until somebody noticed that CMU was stuffing. Hence, the winner of the stuffing contest turned out to be the true winner of the poll.

This ain't intelligence

[Obfuscant]

This demonstration is not one of computer versus human intelligence. It is one of computer versus human cognition.

In other words, can the computer detect the information in the same form that the human can? Can a human read a grocery store bar-code as easily as a computer? No. Can a human read one of those bit-boxes on the FedEx shipping label as easily as a computer? No. Can a human read the Tivo-data sent on the Discovery channel as easily as the computer? No. But none of those failures means the computer is more intelligent, just more capable of recognizing the information that is there.

Both the computer and the human can recognize "moon/parma", but intelligence comes into play when the human starts thinking of Drew Carey and humming the theme music. Intelligence is not just collecting information, it is doing something useful with that information.

Training a computer to fool Stumpy

[ip_vjl]

I took the 'Stumpy' test - where it shows you six pictures and asks you to choose a word that describes them.

Looks like their system is hosed right now because it showed me 4 pictures of horses, 1 of a cowboy, and one of a turtle.

When it asked:
What are these pictures of?

I answered "things"

apparently it didn't like my answer.

Funny thing though, the images are being pulled by image number from the getty images database. You could write a piece of software to lookup the images at getty, pull the keyword list (that getty assigns to all photos) and cross reference the list to get the answer.

--

Then this got me thinking about the whole thing in general. My answer WAS correct. Reminds me of the Cheers episode where Cliff is on Jeopardy and answers the final Jeopardy question:
"Who are three people who have never been in my kitchen."

Not the answer they were looking for, but is it wrong?

I was being a smartass the other day while watching sesame street with my daughter. They had pictures of 4 animals and asked which one didn't belong.
kangaroo
rabbit
grasshopper
fish

they, of course, were looking for 'fish' - because the other three live on land or travel by hopping.

I popped up that the answer could be the kangaroo - because the other three are native to north america. Or it could be the grasshopper, as the only one with an exoskeleton.

My wife reminded me that it was a kid's show. :)

It comes down to the fact that if an strict mechanism is used to judge the answers (like a computer) it may not be able to handle legitimate answers from humans.

--

Seems both the questioner and questionee need to be intelligent to participate.

Edsgar Djikstra said it best...

[CommandNotFound]

"Asking if a computer can think is like asking if a submarine can swim"

Re:Test is of no real use

[sohp]

Yeh, but, what if the Bladerunner is himself a replicant [65.107.211.206]?

Re:Braille terminals

[yerricde]

I guess blind-deaf users need day-to-day-help anyway

So what about Braille terminal users who aren't also deaf? Should Section 508 compliance (required for USA government web sites) allow a web site to require all blind users to have sound cards? /p)

Re:Test is of no real use

[Anonymous Coward]

That's why Bladerunners had to manually test the androids, instead of just letting a machine do it.

No offense, but using information from a fictional novel as support for an argument does not make for good persuasion.

Re:Maybe..

[guidobot]

I think you need to take into account that any intelligent machine would make use of learning algorithms. So if man can teach a computer how to LEARN, and its got the time and resources to learn more than a human (say hundreds of processor years and a connection to the all the world's media), then the end result could be something "more intelligent" than the programmer.

Or how about the example of the AI chess players, who can play vastly better than the people who programmed them?

Re:It says I'm not human

[Asprin]

You entered: toothbrushes
Possible responses: toothbrush
Result: FAIL.

AAAAAAAAAARGH!!! I hate stupid word guessing programs that don't
consistently account for common abbreviations and plurals!

Ahh, delightful irony. That would be the point, then, wouldn't it?

In other words, you have to be smarter than the tools you use, so it's pretty stupid to put a computer that is *not* intelligent in charge of deciding the intelligence of others.

Re:Training a computer to fool Stumpy

[dsfd]

To me, this kind of questions actually test if the person who is answering has the same cultural background as the person who wrote the question.

A certain context has to be assumed to find out the "correct" answer. The same holds for the tests that ask you to complete a sequence of numbers. There are infinite sequences of numbers with the first terms equal !

Complete: 1,2,3,4,5, ?

Why 6 has to be the "correct" answer ? I don't think that there is a good reason, the ordered sequence of the natural numbers is not better than the sequence 1,2,3,4,5,1,2,3,4,5,..

But dont try that ! If your answer is not 6, you will never get a job, because that must mean that you are either totally stupid or from a different planet.

Re:Philosophy 101

[majcher]

The Chinese Room argument is a load of pants. I don't believe it is taken seriously by anyone in the field these days - it has had a large number of holes poked in it, and Searle's reply to each of these flaws is basically, "nuh-uh!"

Too lazy. Find the links yourself.

Re:Human intelligence

[efflux]

Our behavior is governed not only by the higher logic of our brain, but also by millions of years of bizarre -- often obsolete -- instincts.

Please elaborate on what you mean by instinct. How does this differ from any other algorithm? Certainly it was created by evolutionary processes, but we can also conceive of an algorithm where the algorithm itself is compartmentalized and acteded upon by a Genetic Algorithm, thus simulating evolution. We may not expect the resulting algorithm to be very usefull due to the complexity/nuances of selection, yet it should certainly do something.

If you yanked a brain out of a body and hooked it to a computer, it would no longer be truly human because of the lack of hormonal responses that come from every part of the body.
A couple of points:
1) What is human? You have not defined what it is to be human, therefore, it becomes impossible to say unequivocally what it is NOT to be human. 2) Hormonal responses can be looked at in a variety of ways: 1) Such responses, in fact, are simply another stimulus. We would expect any intelligent machine to react differently under a different set of stimuli.
2) The endocrine system also comprises the machine that is "human intelligence" and by removing a part of the machine, we, in effect, cripple it.

As a final point, we are not interested in Human machines per se. Simply machine that are human-like, primarily intelligent in a manner that we may communicate with them and share a semblance of understanding.

Re:I failed!

[The Notorious ASP]

I failed a couple of times as well

It seems there are several problems here, First in my mind is some of these would be impossible for my color blind friends to pull off. Also, with the level of distortion, some were hard for even a person with reasonable sight to get. Finally, this just sucks and is a pain in the arse.

Why not use something simplier, easier, and less resource intensive such as:
1) What is 2 + 4?
2) What is seven minus four?
3) What year is it?
4) How many days are in January?
5) How many hours are in a day?

I mean, I could come up with these all day... I'm sure bots wouldn't be able to read a question, interpret it, and post the correct answer, but hey, correct me if I'm wrong.