Captcha

December 10, 2002 1:17 PM

CAPTCHAis the Completely Automated Public Turing Test to Tell Computers and Humans Apart. The test promises to keep online polls honest, block search engine bots, and end spam as we know it. The program generates and grades tests that (1) most humans can pass and (2) current computer programs can't pass. For example, humans can read distorted text but current computer programs can't. To see if you're human or not, take a Captcha test yourself here. To read more check out this nytimes article.
posted by josephtate (27 comments total)

hehe, I failed because I made a typo. Would that make me human again? I don't see computers making typos.
posted by sebas at 1:20 PM on December 10, 2002

This isn't a Turing test. Turing tests can't be based upon visual processing -- else blind people automatically fail.
posted by ptermit at 1:27 PM on December 10, 2002

The article is a pretty interesting look at some of the people involved in this sort of thing, but the idea is nothing new, this has been around for a while.

As for 'end spam as we know it' - captchas are just one potential component of a TMDA-like whitelist system. Any system that defaults to blacklisting everyone and allows legitimate users to whitelist themselves will prevent spam, and it doesn't need a captcha to do it - at least not yet. A simple reply-with-password or click-on-a-link will work fine.

Whether this is a good idea is another story. I think it takes a massive ego to believe that you're so important that people should go to the extra effort of solving a puzzle, or even clicking an extra link, to contact you. Some day everyone might use something like this - in which case, the price of ending spam as we know it is to end email as we know it. Currently SpamAssassin allows me to delay this inevitable fate.
posted by mmoncur at 1:27 PM on December 10, 2002

Pfft!!! I failed at the first post and I have clue why. I typed three of the words in the picture, definitely no typos, BLAM! I AM NO LONGER HUMAN!

WELL SCREW YOU, FLESH CREATURES!
posted by chrid at 1:27 PM on December 10, 2002

A computer could parse that image. It would be a hell of a lot of work, but I'm sure the bulk of it has already been done. OCR with a database of common english words?

The text is in all caps for chrissake! Undistorting it is a simple matter of establishing the top and bottom of the letters. Computers can figure out what typeface a piece of text is in. Handwriting would have been much better. Or a mix of typefaces. Having correcting distortion, all the program must do is resolve the intersections which are most likely to be the areas of greater visual density on parts of letters which shouldn't have a lot of visual density.

Use the NIN SIN typeface, and handwriting. Add lots of noise, make visual changes using not just value but also hue, then it might become tricky. But not really.

Having said that, it'd be a lot of work. Having said that I hope they're not planning to use this for anything that matters.

The obvious irony here is that the turing test is administered by a human using human judgement. You can't automate a turing test without a computer that can pass the turing test.

Either they are very stupid, I am very stupid, or this a joke o<
posted by KettleBlack at 1:33 PM on December 10, 2002

You can't automate a turing test without a computer that can pass the turing test.

why? ever hear of teacher's editions?
posted by mdn at 1:45 PM on December 10, 2002

I've seen this show up in several registration flows before, thanks... I was wondering what this was for.
posted by FearTormento at 1:47 PM on December 10, 2002

Um. It said "monkey" was a word displayed in the picture. It wasn't. On a later test, it said "bridge" was a word displayed in the picture. It wasn't. What's going on?
posted by kcalder at 1:58 PM on December 10, 2002

Sorry, sebas, no excuses. Please join chrid in the cyborg/zombie/alien section.
posted by languagehat at 2:05 PM on December 10, 2002

Let's keep in mind that the Alan Turing's original concept also required eyesight, literacy, and the ability to type.

I agree with KettleBlack though, I really don't think it would take long to write a program that could beat that test. The idea behind a Turing box was that the implementor of the test was a human: humans are capable of asking very different questions each time. Anything that asks the same question over and over again cannot work.
posted by zekinskia at 2:07 PM on December 10, 2002

it's an interesting idea. but i agree with kettleback, it's not an idea that seems like it would work longterm. npr recently did a radio piece in which they talk about how the porn industry develops sophisticated technology in order to simulate sex with touch technology. they're always making progress (and please excuse my laymen's knowledge of all that's out there now). the technology of spammers, a.i., etc. will continue to develop more sophisticated programs to crack these tests.
posted by uberchick at 2:08 PM on December 10, 2002

KettleBlack -- I just took a pattern recognition course where the focus was on OCR systems. OCR doesn't work well at all on overlapping text, nor on text without a consistent baseline (read: non-wavy).
Although if they wanted to block OCR's, they should've just used an image with a very low signal to noise ratio -- ala yahoo email account setup. OCR stinks once it hits ~50% (IIRC) noise, whereas a human has no problem.
posted by krunk at 2:36 PM on December 10, 2002

If I find out that I am actually a machine, but never knew it, can I sue you for emotional distress?
(I'm guessing probably not, as the courts would likely throw out a suit filed by a machine. DAMN! You humans get ALL the breaks!)
posted by HTuttle at 2:42 PM on December 10, 2002

chrid: not to worry. In the end WE will PREVAIL!
Wuuaaahhhahhhaahhhaabeebbleepbleeepclick....bzzzzz....
posted by HTuttle at 2:46 PM on December 10, 2002

If you read the article, it says that someone has already created a program that can defeat the 'gimpy' test linked in the FPP.

Anything like this turns into a game of leapfrog with the spammers, or whoever you're worried about. Obviously, as computers get more powerful, it becomes simpler to defeat anything like this, and it's time to create a more complex one.

In this way it's more like a room with a locked door than Fort Knox. It keeps the vast majority of the typical criminals out, and makes it more difficult for the rest, encouraging them to choose a different target.

By the way, the number of people in this thread who have been 'classified as machines' demonstrates the downside of systems like this. Humans make mistakes, and that really only proves their humanity...
posted by mmoncur at 3:05 PM on December 10, 2002

So, like, blind people aren't human?
posted by Ogre Lawless at 3:18 PM on December 10, 2002

The second page of the NY Times article does say that the text-distortion method has been cracked by Greg Mori and Jitendra Malik (their overview) but the main point here is that this is an arms-race between those who design the puzzles and those who write programs to crack them and, as such, can stimulate research. As Mori & Malik say in the link above "Learning to deal with the adversarial clutter present in Gimpy has helped us in understanding generic object recognition problems". Besides, anything which can thwart most spambots, if not all, is a useful tool. And, on preview, what Mmoncur said.
posted by jamespake at 3:20 PM on December 10, 2002

You didn't know?
posted by mckayc at 3:23 PM on December 10, 2002

Yahoo used a scheme like this at one point to prevent spammers from using robots to create huge numbers of free e-mails accounts for their dirty work. The spammers apparently created a robot that farmed out the image recognition to real people, sitting at screens all day getting paid minimum wage to type in the words.

Slaves to the machine, indeed. Wish I could find the reference, to be sure I didn't just make the whole thing up.
posted by fuzz at 3:32 PM on December 10, 2002

Uh oh. I guess I'm a computer. Well, I'll just have to peel away my scalp to find the screws that hold my head plate on. I want to upgrade my RAM.
posted by troutfishing at 3:44 PM on December 10, 2002

me2 teach me to haxor teh linux
posted by KettleBlack at 3:55 PM on December 10, 2002

On a later test, it said "bridge" was a word displayed in the picture

I saw "bridge" when I took it. Evidently they are serving up a random sampling from a limited pool of words.
posted by rushmc at 3:55 PM on December 10, 2002

If I find out I'm a machine, can I sue my designers?
posted by Vidiot at 5:18 PM on December 10, 2002

That depends....are you a Ford Pinto?
posted by elwoodwiles at 7:26 PM on December 10, 2002

To err is spambot.
posted by eddydamascene at 9:55 PM on December 10, 2002

Result of the Test: FAIL

You entered the following word:

sunflowers



The possible words were:

sunflower flower flowers
posted by Espoo2 at 10:31 AM on December 11, 2002

I failed the picture one for entering "boats" instead of "submarines" -- those pictures really weren't clear at all.
posted by dagnyscott at 7:17 PM on December 11, 2002