Related. Google has algorithms to solve 99% of some kind of graphical CAPCHA puzzles. They use them for number plate detection.

http://gadgets.ndtv.com/internet/news/google-street-view-tech-can-solve-captchas-with-99-percent-efficiency-509797

"Artificial Artificial Intelligence". ;)

There's some interesting tidbits in the Wikipedia article about it (uses, origin of the name, etc).

Maybe they just stay logged in..

I haven't seen any captcha's when logging in and out of GOG on any computer I have?

Probably this. A bot doesn't need to log out of one account to log into another, but can be permanently logged in with several sessions concurrently, each with a different account. The initial logging in is such a small amount of once-off work that it can be done by the human operator, who might not even be given a captcha to do.

There are people that have a physiological need for vengeance that can span over significant time and effort. And it's not unusual for what started it to be as simple as something like you having the audacity to not let them take advantage of you. Forgetting and forgiving aren't options, and gaming isn't near as important as payback.

Personally, I'm too lazy for vengeance. I'll stick with the gaming. Speaking of which....

It only pops up under certain circumstances, like when multiple accounts login from the same IP/machine.

Huh. Yeah, I guess I should have realized that: A single IP address might be a network of several machines, so GOG would need to allow multiple sessions from the same IP. Without checking the MAC address (or whatever they call the physical address these days; it's been a while since I brushed up on this stuff), I suppose there wouldn't be any way of knowing the sessions were originating from a single machine.

And I didn't consider that bots wouldn't ever need to log out, so a one-time human-assisted login wouldn't be a big deal. I blame GalCiv3 for eating up too much of my attention lately. ;)

I guess bots might be in use here, after all, since it seems much more trivial than assumed. Anyway, thanks for the info!

In any case, for some pocket change one can easily purchase captcha solving services online, where solving is performed by humans. Think mechanical turk, but instead the solvers are most likely slaving away in a sweatshop in some underdeveloped country.

Cheaper, faster, and more accurate than any state-of-the-art algorithm.

NO!!! It would be a major annoyance. I think it would make downloader scripts like gogrepo.py and lgogdownloader fail.

Or at least, it should be optional, ie. if you want to force captcha for every login for yourself, be my quest.

Already an irrelevant idea anyway. Read more of the thread and you'll see that forcing captcha on each login wouldn't likely solve the issue.

They can, but most of the time I can't :(

In theory yes. However computers aren't very good at object identification unless it's 1:1.

Back in the 80-90's there was a technology called OCR which builds a library of characters from fonts, and then does it's best to match everything. Unfortunately you'd get 80% good and static or glitches in areas would get things wrong. Mind you it was probably good enough for people who are blind since redundancy in language is high enough you can usually figure out what's wrong.

Anyways, a lot of Captcha today is warping the text or adding static or doing things that makes it harder if not impossible to do identification. Unfortunately a large portion of the time the images are so warped i can't figure it out either, making the natural organ we have in our heads for visually figuring things out pointless if things are swirled and warped and inverted and strikes through it and you can't tell if the random letters are an i, l, 1, or any other number of odd combinations that you effectively just blitz through to find a good easy captcha because they are crap.

Have you noticed, that a large number of Captchas consist of two separate parts? One that is clearly generated to cause problems and the other one looks like a bad scan or some (slightly) blurry image. Google does this (or at least used to do this). The use the first part to check if the answers are valid and the second part to get humans to do their ocr work for them. You could answer some garbage for the second part and it would still pass. I assume it works the same way the image categorization Captchas. They get free learning data.

It's a clever idea. You provide a service that benefits other people and in turn you get a large number of free workers.