3When You Use reCAPTCHA, You're Translating Old Documents
If you have signed in to basically anything on the Internet these days, then you're most likely familiar with the whole CAPTCHA program. That's the thing where you have to prove you're not a spambot by typing some nearly unreadable words into a box:
Well, this is awkward.
What you may not know is that by using it, you've most likely contributed to the translation of thousands of old documents.
In 2009, Google and a couple of other companies had a problem. They wanted to digitize years of old newspapers and books, using software that can "read" the print it's scanning and then convert it into actual text. But even the most advanced computers had problems reading some of the poor quality scans, because the text was smudged or crooked, or in a font that has been out of use for years.
We remember the Smudgy Smear font came packaged with Windows 3.1.
So, they simply placed those unreadable words in between you and your porn, and told you that you'd need to translate them before going any further. Spambots can't read them because reCAPTCHA uses only the words that the computers already said they couldn't read. It's as brilliant as it is simplistic. The program is called reCAPTCHA, and you've probably seen it this week, if not today. It's currently used by Facebook, Ticketmaster, Twitter, 4chan, CNN.com, StumbleUpon, Craigslist, the U.S. National Telecommunications and Information Administration and thousands of other smaller sites.
"We'll put lines through them, too. Because fuck nearsighted people."
And the project is a huge success, managing to digitize 20 years of The New York Times daily newspaper in just a few months, by letting Web surfers decode the hard bits. It is estimated that websites display 200 million reCAPTCHAs a day.
Over the years, we've managed to get the word "dongs" into the NYT editorial column 17 times!
2The Crowd Beats the Experts ... Over and Over Again
Now, we don't want to give the impression that all of these crowdsourcing tasks are about farming out manual labor grunt work. If you do it right, you can get better decisions on complex issues by asking a crowd than you can by asking a few experts.
However, simple questions like "Where should we eat?" will take hours to argue out.
Some of the best examples of this involve cases where the crowd didn't know they were being asked a question at all. Back in 1986, when the Space Shuttle Challenger exploded shortly after takeoff, the stock market immediately reacted by selling the stocks of the four main contractors that were involved in building the spacecraft. But experts noticed something odd in the buying and selling habits of all of those millions of traders. Three of the contractors' stock saw only minor losses, ending the day with an average loss of 3 percent. Yet the fourth company, an outfit called Morton Thiokol, had their stock end down by a whopping 12 percent.
It was clear that the investors blamed them (or their components) for the accident, even though the news coverage had in no way fingered them as the culprit. The press did have some rumors about what caused the accident, but no one was implicating Thiokol above any of the other companies. It took a Washington blue ribbon commission six months to figure out that, sure enough, Morton Thiokol was responsible. It took the faceless mass of investors one day.
Our methods of prediction involve a financial paper, a bear and wild flailing. We're all millionaires.
Experiments have been done on this from several angles, and the results are always remarkable. You remember the game show "Who Wants to be a Millionaire?" (you saw it if you were watching TV in the same room as your grandmother). It was a quiz show where if you got stuck on a question, you had three options for help. You could eliminate two of the four possible answers, raising your odds to 50/50, you could phone an expert or you could ask the audience to vote on the right answer.
This gives us a unique opportunity to study who has the better scores, the "experts" or the audience. So James Surowiecki, author of the book The Wisdom of Crowds, actually ran the numbers. The results? While the experts got the answer right about 65 percent of the time, which is a respectable score, the audience -- made up of random dumbasses who waited in line to watch a game show -- managed to get the correct answer a shocking 91 percent of the time.
If these questions were posed over the Internet, half the audience would answer "FIRST" and the other half would call them fags.
And it doesn't just work for trivia questions, either. One researcher asked a group of 56 people how many jelly beans were in a jar. And while some people guessed way too low, and some guessed way too high, they all averaged out and guessed 871, only 21 beans higher than the actual amount.
Well, we guess this proves that millions of people can't be wrong and Twilight is the best book of our generation.
This is why you can't dismiss "prediction markets" like Intrade, where the crowd basically bets money on outcomes for everything from elections to geopolitical events. Huge corporations now use such prediction markets to make decisions, including Google, which details the predictions in this blog post. They know what they're doing, right?