Slashdot Mirror
Looking To Spammers To Solve Hard AI Problems
An anonymous reader writes "With bots getting closer to beating text-based CAPTCHAs for good, New Scientist points out that when they do, OCR technology will at least have advanced. The article goes on to suggest that whatever kind of reverse Turing Test that comes next should be chosen to motivate spammers to solve other pressing AI problems, such as image recognition. Are there any other problems that criminal crowdsourcing could help with?"
several years ago 'neural nets' were the big thing and they were thinking that they could make them 'learn' and do useful things.
i always thought that traffic control would be an interesting application. if a computer could look at video of an intersection (and streets leading to the intersection) and figure out where cars were and weren't, you could make traffic lights a lot less annoying.
so our CAPTCHA might be a picture/video of cars and a request to count them?
eric
Spammers are unlikely to share their results with the rest of the world. They're motivated by financial rewards, and there is absolutely no incentive to publicize their methodology in any format.
Not only would the "good guys" learn from it -- and thus potentially defeat the spammers' discovery -- but other spammers would simply steal their work.
I'm not as optimistic as the New Scientist. Spammers need a really low success rate, as compared to OCR technology which needs a really high success rate.
Wherever there is greed, it can be harnessed to actually do some good. I love it!
I would agree, if general-purpose captcha-beating software were available. But that isn't so. Each captcha system was beaten by custom code, individually written for that system. So in effect, it is not much different than adding a new font to existing OCR software.
This is a reasonably accurate description of the stock market.
That brings up a good point. When AI is good enough to get past CAPTCHA it will hopefully be good enough to filter out the spam.
But has it?
Unfortunately, CAPTCHA is radically easier than actual OCR. When cracking a CAPTCHA, achieving a success rate of 5-10% is absolutely fine. Plus, when you submit your answer, you are told whether or not you got it right. With OCR, anything short of high 90's is pretty much useless, and the only feedback available is through manual human intervention, which scales poorly.
Arguably, the only significant OCR advance has been RECAPTCHA, which is just a clever way of making humans do the hard stuff in a way that actually helps, rather than just using makework problems.
It is certainly true that CAPTCHA cracking has advanced considerably, that just doesn't apply too neatly to real OCR problems.
What about people for who $50 is a year salary? Congrats, you just split the internet into the rich and the poor. No more accessing the internet from africa from an old PC powered by a donated solar cell. Good job. You probably going to get a nobel price.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
What about people like me who can't seem to get the hang of the darn things? (I personally wouldn't be surprised if they're some kind of elaborate hoax...)
Great idea to have just a button. No way that can ever be abused. Well, to be sure, perhaps still better find some way to see if you pressed that button or a computer. Perhaps a way of hard to read letters. Oh, wait. Back to the drawing board.
Don't fight for your country, if your country does not fight for you.
The cocktail party problem - our ability to hear out a target conversation amongst a barrage of others. There's still a lot of room for improvement here as a computational problem; meanwhile, it's relatively easy to get a correct human response to multiple-talker environments if you cue listeners for what to listen for.