Slashdot Mirror
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."
Anyone that has seen Star Trek:TNG knows that Data is a pretty smart fella.
Trolling is a art,
I did the gimpy test.
ResultsIt switched pictures on me! Honest!!
The difference between computer and human intelligence is the human ability to revel in his. That is, taunt others. Until a computer can get in my grill and explain to me on a colorful fashion that I am nothing more than a grab-ass-tic piece of *human* sh!t, then I won't think much of computers.
Welcome to The New York Times on the Web!
For full access to our site, please complete this simple registration form.
As a member, you'll enjoy:
In-depth coverage and analysis of news events from The New York Times FREE
Up-to-the-minute breaking news and developing stories FREE
Exclusive Web-only features, classifieds, tools, multimedia and much, much more FREE
I don't need no instructions to know how to rock!!!!
Sure, computers aren't as smart as people. Wow.
Computers are not good at complex pattern recognition. Wow.
For the record, computers can recognize words like this, just not very easily. With a big enough dictionary and a lot of patience, you'd be suprised at what they can do. While still an undergrad I was able to write a rather simple program that would recognize images of the cardinal numerals, even if they were highly mangled, and worked with a grad student in building something that could pick out certain features of a rotated image and by comaring with some sample features, rotate the image correctly.
-Space for rent
We are never going to have a machine that is truly "human". Let me explain.
That doesn't mean we won't have intelligent machines that can do just about anything intellectually that a human can do. A human being is more than just a smart computer. Our behavior is governed not only by the higher logic of our brain, but also by millions of years of bizarre -- often obsolete -- instincts. 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.
It's simply going to be too hard/impractical and, frankly, useless to make an intelligent machine that mimicked every hormonal reaction and instinctual mechanism.
We will have intelligent machines, but we will never have human machines.
Sometimes it's best to just let stupid people be stupid.
A I mentioned at the bottom of this journal entry. I think a new version of the Turing test should be whether a computer can tell the difference between a Human and a Computer.
"Not knowing when the dawn will come, I open every door." - Emily Dickinson
[Kicks first man in balls]
First Man [falls over]: "AAAAAHH!"
Me: "Human."
[Kicks second man in balls]
Second Man [falls over]: "Gffffff-!"
Me: "Human."
[Kicks third man in balls]
Third Man [falls over]: "..."
Me: "He's the robot! Get 'im!!!"
What is music when you despise all sound?
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.
There's 10 types of people in this world, those who understand binary and those who don't.
Searle's Chinese Room theory. Strong AI vs. Weak AI and human interaction/interpretation. Fun Stuff. http://www.utm.edu/research/iep/c/chineser.htm
It's very easy to do a search at news.google.com with some of the words from the story summary and come up with the story elsewhere.
Yes, it's a nytimes.com link, but it's without the registration.
You entered: noses
Possible responses: nose
Result: FAIL.
Wohoo! I'm a robot! This test proves it! Vegas here I come!
Why does this test make me feel like I just had a run-in with John Ashcroft?
Another area not discussed in the article is vicarious experience, that is, experience and knowledge you have because some cause and effect relationship existed with someone or something else.
For example, the computer's tactile interface has to touch the oven and say 110 deg C, as opposed to taking as fact "I heard a human mention that Unit 5 already did that and it was 110 deg C, so I accept it as fact that it is 110 deg C".
I know I'll get modded down for this, but I wonder what the limits of questioning the computer / human participants was? (Article said they quized participants to see if they could tell who was human and who was a machine). Like, could they ask "What number am I thinking of?" The machine would blank out and the human would stupidly blurt out "69 dude!"
I'd hate to think that computer power didn't increase between the time of Turing's death in 1954, and 2000....
:)
Pretty much any prediction that Turing could make about computers nearly 50 years after his death - and before the advent of transistors - would pure speculation. The fact that Turing's prediction that AIs would be indistinguishable from people in the "Turing test" was wrong, and that other projects based on sheer informational density (such as CYC) have been dismal failures, indicate that it is the purely scripted/explicit logical constraint strategy of solving this problem that is faulty. Unfortunately, the 30 years after that prediction have focused pretty much entirely on scripting and logical constraints, and other methods of artificial/computational intelligence didn't see the light of day until the 80s and 90s.
Be sure to watch further developments in modeling of neurological processes, as there is still hope along this avenue of research
Mitch Kapor and Ray Kurzweil have bet $20,000 on whether a computer will pass the Turing Test by 2029.
no joke, I got this...
o x/box ... my question is how'd they know?
acid/head
acid/head
acid/great
acid/angry
b
Whoever said that computers can't handle superposition has never heard of convolutional neural networks.
Really, comparing human intelligence to computer intelligence doesn't seem like a good idea unless we're going to define what kind of computer intelligence it is.
Neural computing really screws the comparison up - the kinds of computing that normal computers are good for are quite different from the kinds of computing that neural nets are well suited to. Furthermore, different neural net architectures make for different capabilities - the tasks a feedforward network are best suited to are very different from the tasks a bayesian network are best suited to.
Take a look at this page for a good run-though of the different kinds of nets.
Gee, I thought you were going to say something like, "Much like with RadioShack, where you have to perform a test to see if you're talking with an intelligent being."
If all this should have a reason, we would be the last to know.
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.
Don't label something "offtopic" unless you know the topic well enough to tell what's on topic.
Involves measuring pupil dilation when asked a series of personal questions...very good method.
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.
"It never got weird enough for me." - HST (RIP)
You're looking for insight on Slashdot, that's where you're wrong.
If all this should have a reason, we would be the last to know.
To be considered truely "intelligent" a computer must:
..." post.
1. Make a "first post" posting 15 minutes after the article goes up.
2. Be the fourth person to enter a "In Soviet Russia
3. Be labeled a karma whore.
4. Whine about the masiv tipe ohs in artaculs.
5. Hate M$, Sony, MPAA because thats one of the three laws right?
CAPTCHAs have several applications for practical security, including (but not limited to): Cool, eh?
What I mean is, I don't think an intelligent being would be capable of creating something that is more intelligent than himself.
The machines need to be programmed by humans, who are limited by their own inteligence.
Can God make a rock so big that he can't carry it himself?
"The majority is always sane, Louis." -- Nessus
http://slashdot.jp
I simply use my "Taco Test" (Inspired by the Invader Zim cartoon) to thwart chat bots and telemarketers. It's an amazing, powerful test that no computer or automated script can withstand.
I ask the "suspected bot" if they like tacos. If they give me an intelligent answer, they are not a bot. If they give me an answer like "Wanna see my hot pics go to http://192.168.1.112/hotbabezzzz.pl?2345" Then they are a bot.
This test also works on telemarketers in a slightly different fashion. I tell them to "STOP... I'll only buy your product if you send me a taco with it. If not, no deal." since there are big logistical problems with sending me a taco, they are thwarted every time. I'm sure this test would work equally well with any obscure food item.
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.
I'll form my OWN solar system! With blackjack! And hookers!
Once you devise a test system, someone can write non-AI software that can fake it and pretend to be human by knowing what it needs to for the test. Only a real human can tell human and machine intelligence apart, not a systematic test. That's why Bladerunners had to manually test the androids, instead of just letting a machine do it. Real-time human insight is key to testing machine intelligence.
11*43+456^2
The CAPTCHA website (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 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.
Today's Sesame Street was brought to you by the number e.
I suppose it could generate a spoken list of words in a sound file that is linked to from the image.
The CAPTCHA web site has such a test, but of the sites that use image-based bot tests, only PayPal offers an audio alternative.
Another problem is that sites often present the tests in proprietary formats with expensive implementation royalties, such as .gif and .mp3.
But even providing both the image in a free image format (.png) and the audio in a free audio format (.ogg) won't help blind users behind a Braille terminal without a speaker, such as blind-deaf users.
Will I retire or break 10K?
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.
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.
"Asking if a computer can think is like asking if a submarine can swim"
Did anyone notice that a lot of these "human" test are also the same ones used for things like hearing/eye tests, color-vision impairment, etc.
This knocks out computers, which lack the intelligence/programming (so far) to differentiate between conflicting objects to make out a letter/numbers.
It also may knock out humans with vision problems though, especially those with colour-vision issues.For those with hearing problems, the sound test isn't good either.
It seems that right now, computers trying to translate these puzzles probably perform along par with old-folks. This also might mean that quite a few seniors may have issues getting a yahoo account though.
Abort, Fail, Retry? I know how to fix it, but do you human?
A Good Troll is better than a Bad Human.
I don't believe that Turing proposed the Turing Test as the test to use, but rather as a "mathematical proof" that you could construct such a test.
Basically he said if you could not tell the difference between a computer and a person then you would have to say it was intellignet. ie. this is a way of establishing an upper-bound test - not necessarily that this is the best test.
Unfortunately, IMHO, the AI community and other latched onto this test and put effort into fulfilling the Tring Test rather than more practical and useful goals.
If you asked "Did you sleep well last night?" and the computer said "Me not sleep, me computer." (or some question on some other biological function) then you could probably determine the difference between a human and a computer. This need not, however, preclude machine intelligence.
Engineering is the art of compromise.
Thet's part of the test - the logical computer clicks on the link and gets nowhere, while the human uses their ingenuity and sticks "captcha" into Google instead.
You can say "there are physical differences between person A and computer B", but the problem is that person A and person C also differ quite a bit.
Saying "foo cannot be done" frequently results in someone being utterly wrong. Just a few decades ago, the idea of atomic power would have been laughable -- the ability to wipe an *entire city* away? How about having a person walk around on the moon? Unthinkable.
So, at the moment it seems to be an insurmountably difficult problem. But, a few years ago, the same thing would have been said about problems that we're not starting to think about being doable via quantum computers -- the face of computer science literally changed.
May we never see th
Hate me!
*note, I'm not sure now that this has anything at all to do with the topic, but it's something that bounces off my head sporadically*
/ex
Recently, I've been working with developmentally disabled people as a job coach-- Making sure they have the ability to do the job they're supposed to, and help them to understand anything that needs to happen.
Part of this is working at a local fast food restaurant. The girl I'm working with can do math fairly well, but she has problems with logic, and pattern matching.
And a few times I started thinking about her as a computer-- She can do math fine, and if I specifically tell her how to match a pattern, she can do it for a short time, but she can't do it in situations like when people order a combo. Let's say they order a #1 with onion rings and a small drink, a #6 large combo, and a kids meal, she won't be able to recognize them as "combos" (she'll read the whole thing back to them item by item.) This brings me to a whole other tangent about user interface design (why the normal methods suck, mostly), but that'll be saved until a proper time.
This has been a difficulty with her position as a cashier, but I find it interesting that I'm more or less programming her as I talk to her and re-affirm her patterns to match.
I wonder if certain disabled humans would fail any "turing" test that were given to them, because they don't have normal pattern matching ability. Furthermore, isn't it possible that instead of trying for fully developed Artificial Intelligence, we should look at perhaps emulating those with disabilities? After all, wouldn't this creation process be easier than a "fully aware", fully pattern realizing person?
AI has always interested me, but I don't know nearly enough about it. The thing that made me notice this is I keep talking to her like I would program a computer ("If this, then that, otherwise this other thing" and "While there is someone in line, take their order").
Maybe I'm off based, or this is already an accepted practice. Can anyone correct me?
the question of whether computers use intelligence the same way as humans use intelligence has long been determined through the 'chineese room'.
the point of John Searle's Chinese Room being is to see if 'understanding' is involved in the process of computation. if you can 'process' the symbols of the cards without understanding them (since you're using a wordbook and a programme to do it) - by putting yourself in the place of the computer, you yourself can ask yourself if you required understanding to do it:
Minds Brains and Programmes (The Original Chineese Room):
http://www.bbsonline.org/documents/a/00/00/04/8
the complementary question - 'is the human brain
a digital computer' is answered by the same author:
Is the Human Brain a Digital Computer (John Searle):
http://www.ecs.soton.ac.uk/~harnad/Papers/Py104
Summary of the Argument:
1. On the standard textbook definition, computation is defined syntactically in terms of symbol manipulation.
2. But syntax and symbols are not defined in terms of physics. Though symbol tokens are always physical tokens, "symbol" and "same symbol" are not defined in terms of physical features. Syntax, in short, is not intrinsic to physics.
3. This has the consequence that computation is not discovered in the physics, it is assigned to it. Certain physical phenomena are assigned or used or programmed or interpreted syntactically. Syntax and symbols are observer relative.
4. It follows that you could not discover that the brain or anything else was intrinsically a digital computer, although you could assign a computational interpretation to it as you could to anything else. The point is not that the claim "The brain is a digital computer" is false. Rather it does not get up to the level of falsehood. It does not have a clear sense. You will have misunderstood my account if you think that I am arguing that it is simply false that the brain is a digital computer. The question "Is the brain a digital computer?" is as ill defined as the questions "Is it an abacus?", "Is it a book?", or "Is it a set of symbols?", "Is it a set of mathematical formulae?"
5. Some physical systems facilitate the computational use much better than others. That is why we build, program, and use them. In such cases we are the homunculus in the system interpreting the physics in both syntactical and semantic terms.
6. But the causal explanations we then give do not cite causal properties different from the physics of the implementation and the intentionality of the homunculus.
7. The standard, though tacit, way out of this is to commit the homunculus fallacy. The humunculus fallacy is endemic to computational models of cognition and cannot be removed by the standard recursive decomposition arguments. They are addressed to a different question.
8. We cannot avoid the foregoing results by supposing that the brain is doing "information processing". The brain, as far as its intrinsic operations are concerned, does no information processing. It is a specific biological organ and its specific neurobiological processes cause specific forms of intentionality. In the brain, intrinsically, there are neurobiological processes and sometimes they cause consciousness. But that is the end of the story.
--
best regards,
john
Trouble is we may be able to model it but it doesn't mean we understand it.
At least when we are simulating/modelling weather we can start with base points and do comparisons.
Whereas with stuff like consciousness, I suspect even if the model is broken you might not be able to tell till much later.
If we really wanted intelligent entities which we didn't understand (how they work), there are always humans and other creatures.
The GM bunch may even concoct a few more.
Back during high school, I wrote dozens of .bat files called "what" or "how" or "go" and so forth, and I basically had them parse themselves so they could keep up a semi-decent conversation. Kind-of like a shell-based Alice. (Well, if you knew what to say, since if you didn't put a recongized word first on the line, it would just say "Bad command or file name.")
My favorite was when I came back from a two-year stay in Brazil, and my friend and I were at the computer. We had both totally forgotten about those little batch files, and his playing with the computer went something like this:
C:\>dri
Bad command or file name
C:\>Huh?
Bad command or file name
C:\>What was that?
Bad command or file name
C:\>Could you repeat that please?
Bad command or file name
C:\>Thank you.
You're mighty welcome, sir!!
Any sufficiently simple magic can be passed off as mere advanced technology.