We are all familiar with the little test called a ‘captcha’ (an acronym that stands for Completely Automated Public Turing test to tell Computers and Humans Apart) that we sometimes have to pass in order to access some website. You may have noticed that these tests are getting harder, in that we often fail once or twice before passing.
The video below explains why this is happening and it is not because we are becoming less human. It says that there is a race between these tests and computers, that as computers get better at doing them, the tests have to raise the level of difficulty. It also says that there is something else that is going on in the background, and that is that our responses are used to create databases that enable computers to become better at character and image recognition. For example, many of the captchas are to identify things that we see while driving, such traffic lights, pedestrian crossings, fire hydrants, and the like. The responses are used to program self-driving cars to better identify those things.
We may arrive at a point where this system runs out steam because we are not able to design simple tests that only humans can pass.
eBay had one that wasn’t a captcha (which I hate already). It was a jigsaw puzzle missing one piece, and you had to move the piece into place with a slider. It was a lot better, for a human anyway.
If computers are learning from those horrible little images, and given the general state of the software that we are universally expected to put up with, I am not looking forward to self-driving cars. I think I may know some drunks who could probably drive cars better.
Nobody believes me when I tell them I’m not, so…
http://thepunchlineismachismo.com/archives/comic/the-straw-that-broke-the-robocamels-exoskeleton
How is using these images in CAPTCHA supposed to help machine intelligences when the correct answers are already known? If there were no known answer that they would not be able to deny you entry on the basis of an incorrect response, while those incorrect responses would be messing up the machine training.
@4, that’s an interesting question. I would like to get an insiders perspective. Here is my own, ignorant proposal.
That for example, what these algorithms are actually examining is not categorization, but the timing of the selection of various images. Or that they are using consensus from a population, and the first x number of users are allowed to pass the test without image categorization.
I have had cases where I’m fairly certain the captcha was simply incorrect. Not that my gripes matter. I’ve also had occurrences where no categorization test was given and I assumed they were using the latency of my response as an identifier.
I don’t design these systems, but I used to professionally model and attempt to understand human and nonhuman primate behavior and neural signals.
@2, I believe you Abbey, your comment wasn’t first, and it was an hour 4 minutes after the first one. Seems human enough for my first order analysis.
@4 I’m not sure about the traffic images that CAPTCHA uses nowadays, but the previous ones with words did something similar:
They were used to help digitize old print books that computers had trouble reading due to smudged or worn out printing. Each CAPTCHA had two words: One was the test word that you would have to get right (usually written wobbly with lines through it). The second word was just scanned from some book they were digitizing. You actually did not have to get that one correct since the computer didn’t know the answer -- it was crowd sourcing the answer in order to get that book digitized. I wonder if I’m responsible for some odd words in digitized books since I’d sometimes type something strange for the 2nd word.
Perhaps something similar with the traffic ones? There are a few images that are test images that you need to get and a few that the computer doesn’t know and uses to learn from you.
@7 Huh, test images and unknowns. That’s sensible. A much better idea.
As a non-American I find the fixation on US road paraphernalia most frustrating. I have no idea what I fire hydrant looks like, outside of those I’ve seen in films, and many traffic lights are positioned in places you don’t find in other countries.
@9, As an American I find the fixation on US road paraphernalia most frustrating. These things vary widely across portions of the United States in irritating ways.
Couldn’t we do puppies or cats or something? For once.