Hacker News Reader

Comment by awalton | Hacker News Reader

I'm old enough to remember when reCAPTCHA was first introduced to help with deciphering text from OCR'd books. At that time, it didn't feel so bad answering those, as we were using our intelligence to genuinely help preserve our cultural history.

It then switched to being numbers on buildings and street signs, and it immediately felt worse - we were now doing a job for Google, and an annoying one. Mechanical Turk from Amazon was invented to do this kind of chore.

It's now creating training datasets for whatever else Google wants them to - it appears to mostly be for self-driving cars now, to identify landmarks and road signs.

It's definitely not on autopilot, and it's definitely a real problem.

View on HN

There's no difference between helping google solve book scanning vs other datasets for whatever else they decide.

The fact that captcha exists on a website is the website owner's intension to cause friction for their users. It has nothing to do with google's use of reCapture directly. The only vote you have is to not use said website - sometimes harder said than done but that's the only option you have.

>we were now doing a job for Google

And in return you get an internet that is not filled with spam and bots. Or at least less filled with bots.

Bots that we've trained. And there's also an xkcd for this: https://xkcd.com/2228/.

Every time I get a captcha with traffic lights I imagine a Google self driving car stopped at an intersection waiting for me to complete the captcha so it can figure out what it sees and move along :)

Naturally, there's an xkcd for this: https://xkcd.com/1897/

I used to have fun filling in those reCAPTCHAs with incorrect answers. They would show you two words, and it was always obvious which word was the actual CAPTCHA test and which was the OCR input, because the former would be warped into a funny shape while the latter was always a rectangular block of normal text. So I'd type in the correct answer for the warped word, and something like "fuckface" or "cocksucker" for the regular text, and it would be accepted.

I like to think that somewhere out there on Google Books, my efforts have resulted in an innocuous word being replaced with something offensive.

Probably not, as statistically your answers are filtered out as noise. Remember, hundreds of people were getting the same images.

What a sad way to indulge the urge for petty vandalism. At least graffiti carries some risk and sex appeal.

A literal oldie but goodie: https://wraabe.wordpress.com/2009/03/07/an-ocr-cliche-into-h...

Granted this is mostly a kerning issue.

Without wider textual context the correct answer is hard to determine.

Before I discovered the Buster plugin, I used to try to be as unhelpful as possible in my forced Google AI training sessions.

I'd use the audio option in the reCraptcha and then see how little of what I heard I could get away with actually entering into the form. Often it sufficed to just type one word out of the entire audio clip or even enter a word which sounded similar [eg. audio clip says "tranced", I write "transit"]. The most satisfying ones of all where when, after listening to a complete sentence containing either word, I could pass the reCraptcha by simply typing "the" or "an" into the form.

One of the things that I find slightly disappointing about Buster is that it types in the entire sentence, when solving the reCraptcha. I hate to think Google are under the impression I've suddenly started trying harder!

>>I'm old enough to remember when reCAPTCHA was first introduced to help with deciphering text from OCR'd books

Am I the only one who always, on purpose, put in the wrong answer for the clearly scanned word? For me it was kind of a rebelion for being used this way, but it was always super easy to tell which word is generated and which one is scanned - and the algorithm only required the generated word to be correct, so I always put in absolute nonsense for the scanned word to break their OCR detection.

Reply on news.ycombinator.com