Gavin’s Digital Diner

At the risk of being Pollyannaish, I find it pleasing when “good” things are born from “bad.” One of these “good” things caught my eye recently. It something called reCAPTCHA. Not only is it neat, it also turns a sad state of affairs on its head. It helps create a public good, a silk purse in the form of giant, online digital library, from the sow’s ear of having to prove we’re a human to some impersonal computer, over and over again.

We’re now confronted regularly with the requirement of proving our humanity before we’re allowed to comment on a blog posting, or sign up for a Yahoo! account, or send a happy note to Congress. I’ve written about that before — the Congress stuff — probably alienated a few friends in the process. (If you’re interested in my take on the Congressional move to institute so-called “logic puzzles,” look here and here.) Now that I think of it, it might be useful to have members of Congress prove that they’re human. I wonder how many would pass.

I admit it: there are days where I feel I might not pass such test. Moreover, in the grand scheme of things, I also admit there is a part of me that finds it altogether strange that we actually need such a test.

Nevertheless, tests to see if you’re “human” are now part and parcel of daily life on the internet — and appear as such things as CAPTCHA or other forms of logic puzzles. They’re all tricks, barriers to keep the ‘bots from flooding blogs, discussion groups, message boards, and web sites with auto-generated crap about drugs or body parts, or both. I’m always surprised, but apparently Christian Singles want to meet me! Obviously, they don’t know me (or my wife).

Metaphysical and political ponderings aside, tests that make you prove you’re human, officially called “Turing Tests,” are now downright commonplace.

The original “Turing Test” was developed by Alan Turing in the 1950’s. It was based on a parlor game in which a man and a woman would go into separate rooms and, remotely, using only the written word, try to convince other players in the game that they were the other person. Turing took that parlor game and tweaked it a bit, and came up with a test of a machine’s ability to demonstrate intelligence. A classic Turing Test asks whether: “a person could reliably tell the difference between a human in one room, and a computer in another, based solely on their written responses.” If (or when), a computer can fool that that human judge, computers will have passed the test, and have to be considered intelligent. [I'm here to say, right now, I want equal standing. If I can fool a judge into thinking I'm human, I want to be considered intelligent too.]

Today’s anti-spam and anti-bot tests are called “Reverse Turing Tests.” Unlike the classic test in which a human tries to tell the difference between another human and computer, a Reverse Turing Test has a computer judging if you’re a human or if you’re just another spam-generating, Viagra-pushing spam-bot.

Of such tests, CAPTCHA is probably the best known. It’s the one that asks you to use your uniquely human ability to recognize letters and numbers, even when they’re skewed, warped or wiggly. That’s something computers just don’t do well. CAPTCHA, by the way, stands for “Completely Automated Public Turing test to tell Computers and Humans Apart.” It’s trademarked by Carnegie Mellon. Here’s a sample of a CAPTCH – actually a reCAPTCHA. Let me tell you what the “re” in reCAPTCHA is all about, and why it’s so damn neat. reCAPTCHA takes the idea behind CAPTCHA and makes it work for a living.

An Example of a reCAPTCHA Turing Test

According to the reCAPTCHA web site, about 60 million CAPTCHAs are solved every day. In each case, some human takes about ten seconds to do what only humans can — recognize some funky text and translate it back into letters and numbers. Ten seconds. Not that much time, and with good effect. It stops the spambots and Christian Singles.

But, when you think about it, though, in the aggregate, those seconds add up. Supposedly they add up to over 150,000 hours a day — 150,000 hours spent closely interpreting a funky image of some text, and then entering that into a form. With this in mind, the reCAPTCHA people apparently had a bright idea: Why not use that 150,000 hours for some positive good.

reCAPTCHA does exactly that. It channels that human effort into helping verify and interpret things that scanners can’t figure out. reCAPTCHA uses this massively parallel human processing system — all those people using their human abilities to recognize badly formed letters — to correct and/or verify OCR mistakes as books are being scanned into the Internet Archive. Damn neat. A silk purse from a sow’s ear.

For all of you that immediately thought: Well, what good is that? If the system doesn’t know what the word is, how can you use it for a Turing Test? I can hear you looking for the comment button… Wait… Hold your horses, the answer here is as brilliant as it is simple. First, take a look at the sample graphic above. What do you see? Two words, not one, right?

reCAPTCHA uses word pairs, one known word, and one unknown word. Only one of the words is from the book scanning process. The reCAPTCHA system pairs up the words, and asks you to translate both. In this way, using known/unknown word pairs, reCAPTCHA allows you to both prove you’re human and extracts free labor, getting you to transcribe and verify some mystery word from a book being fed into the Internet Archive.

Even neater, they use the same mystery word ten or more times, with different people, checking and double-checking the word. Only then is the most common, and most likely correct, answer fed back into the Internet Archive. This, of course, overcomes that pesky thing called human error. After all, humans… well, they’re not to be trusted. Besides, how do we really know you’re really human, anyway…?

Trackback URI | Comments RSS