5 ms·
>>There's a big question on whether these approaches work well for onboarding the next billion users to the internet That isn't what captchas are used for. Th
by sandworm101 6y ago
>>There's a big question on whether these approaches work well for onboarding the next billion users to the internet
That isn't what captchas are used for. There are two side to every captcha, what the website is using them for and what the captcha creator or provider is using them for. In google's case, captchas are being used to hone AI. So it is no wonder that items like hydrants, store fronts, crosswalks and traffic lights dominate. Those are the objects that any AI-controlled vehicle will have to recognize. Given that the first market for AI-driven vehicles seems to be western nations, specifically the US, it is no surprise that google gears its captcha program to objects from that area.
If google were developing robots to pick fruit captchas would ask us to identify bananas and oranges. I worry about the day that captchas start asking me to differentiate people from other objects. I don't want to help train an AI to tell people from trees, soldiers from bushes.
- wolco2 6y agoIn fairness if they asked that it would be exclude the human the way streetview maps works now
- sandworm101 6y ago>> it would be exclude the human That is one way of running an captcha: set a task that is not possible for a human. Anyone who answers correctly must be a bot. "How many flowers in this field?" "Solve for X: 3567/134 = X" Anyone getting those correct in a reasonable time is probably not a human and so has failed the captcha.
- martamorena9182 6y agoThat doesn't work, because either you need to randomize which way you ask (and then all a bot needs to do is try a couple of times to get the question where it needs to answer correctly), or you always need to ask one way or the other, which also doesn't work, because then the bot will answer incorrectly on purpose. It would be different if the captcha would look at the incorrectness. I.e. the bot would now have to make the same mistakes that humans would make, which might be a workable path for captchas, however its unclear what benefit this would bring compared to answering correctly.
- iso1631 6y ago"Solve for X: 3567/134 = X" Would take me, a human, about 3 seconds to 1) highlight 3567/134 2) Press Alt-Escape 3) type "bc<enter>" 4) middleclick 5) Doubleclick answer 6) Middle click in answer box 26.61940298507462686567
- matchbok 6y agoGoogle no longer uses captcha to hone their AI.
- coherentpony 6y agoDo you have a source for this statement?
- Closi 6y agoThis seems unlikely considering all the captcha’s are still image classification problems related to driving.
- CRConrad 6y ago> So it is no wonder that items like hydrants, store fronts, crosswalks and traffic lights dominate. Those are the objects that any AI-controlled vehicle will have to recognize. The point of the GP was, it's training these AIs on specifically _American_ hydrants, store fronts, crosswalks and traffic lights, which sucks not only for those of us unpaid mechanical Turks who aren't Yanks, but also makes you wonder if nobody cares that these coming AI-controlled vehicles won't be so great at discriminating NON-American hydrants, store fronts, crosswalks and traffic lights from _non-American pedestrians._ And no, America does not equal “the West”; most of your street furniture looks slightly“ off” compared to, say, that of most European countries. Anyway, the part of the “West” that will probably be at the forefront of this wave will probably, as usual, be the most hyper-advanced: Japan, Hong Kong, Singapore, Taiwan...
- unishark 6y ago> Given that the first market for AI-driven vehicles seems to be western nations, specifically the US, it is no surprise that google gears its captcha program to objects from that area. This is a roundabout way of just saying the most obvious explanation (sites with US-centric captchas are presumably mostly interested in US users anyway) but wrapped up in an unnecessary conspiracy layer to add creepiness.
- howLongHowLong 6y agohttps://www.motorbiscuit.com/how-security-captchas-crowdsourced-self-driving-car-technology/ https://www.motorbiscuit.com/how-security-captchas-crowdsour... (Not the best source perhaps, but I just picked one among dozens) It's not a conspiracy theory, it's actually true that captchas are used as data for self-driving car algorithms, and that the data would be useless unless it was relevant to things those cars are likely to encounter. The reason they're so common is they are distributed for free, for the same reason: more data. I often feel like I ought to get paid when presented with an annoyingly long queue of them.
- dafoex 6y agoGiven Google tracks everything about me, there's no reason why they can't find-and-replace "crosswalk" with "zebra crossing" to match my locale. Its not a conspiracy, its just laziness on Google's part
- sandworm101 6y agoWhat conspiracy? Google has been rather open about this for years.