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."
--One sex question without a captcha
Sex - Find It
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
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.
Mitch Kapor and Ray Kurzweil have bet $20,000 on whether a computer will pass the Turing Test by 2029.
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.
Yes, if you look at the captcha site, it lists "Sounds" under Captchas. Here's the text:
Sounds can be thought of as a sound version of Gimpy. The program picks a word or a sequence of numbers at random, renders the word or the numbers into a sound clip and distorts the clip. It then presents the distorted sound clip to its user and asks the user to type in the contents of the sound clip.
This would probably be similar to the visual techniques, most likely employing some audio filters so its hard for a computer to decipher (our ears are pretty sensitive in deciphering noise from actual voices/useful sounds, so it shouldn't be a problem for us)
Sorry, my friend, but you're off by 20%. According to the UN, 97% of Americans can read. Not as many was we'd like, but far from 23% illiterate.
That depends on whether you count level 1 literacy (that's roughly equivalent to being able to recognize street signs) as being able to read.
"I have opinions of my own, strong opinions, but I don't always agree with them." -- George H. W. Bush
A link to the National Assessment of Adult Literacy.
States that 21-23% scored level 1.
Results are from 1992, 2002 survey is still being collated.
Does give definitions of various levels, as well as the limitations of the test.
Also gives possible reasons why scores were so low.
http://nces.ed.gov/naal/resources/execsumm.asp