WEBVTT - A Quick Chat About CAPTCHAs 0:00:04.480 --> 0:00:12.399 Welcome to Tech Stuff, a production from iHeartRadio. Hey there, 0:00:12.440 --> 0:00:15.800 and welcome to tech Stuff. I'm your host, Jonathan Strickland. 0:00:15.800 --> 0:00:18.799 I'm an executive producer with iHeart Podcasts. And how the 0:00:18.800 --> 0:00:21.560 tech are you. I thought today we could do a 0:00:21.760 --> 0:00:25.560 real quickie, because you know, sometimes it's nice just to 0:00:25.600 --> 0:00:29.400 do a short thing to talk about a subject in tech. 0:00:29.920 --> 0:00:32.480 And there are a whole bunch of tech Stuff episodes 0:00:32.800 --> 0:00:36.559 in which I have talked about the Turing test. So 0:00:37.200 --> 0:00:39.400 there are a lot of different variations of the Turing test. 0:00:40.040 --> 0:00:43.559 It's based off a thought experiment from Alan Turing, the 0:00:43.840 --> 0:00:48.680 computer scientist, very influential, very important in World War Two. 0:00:49.200 --> 0:00:54.160 He helped crack the Enigma code. And actually the movie 0:00:54.280 --> 0:00:59.720 that sort of depicted his efforts in cracking Enigma is 0:00:59.760 --> 0:01:03.280 called the Imitation Game. Well, the Imitation Game makes reference 0:01:03.440 --> 0:01:07.160 that is like part of the Turing test. So he 0:01:07.319 --> 0:01:09.760 kind of proposed this test when people would ask him 0:01:09.760 --> 0:01:13.240 if he thought machines would be capable of thought. Now 0:01:13.440 --> 0:01:15.520 keep in mind, like this is back in the forties 0:01:15.560 --> 0:01:18.480 and fifties, do you think that machines will be capable 0:01:18.520 --> 0:01:20.240 of thinking? And he said, I don't really think that's 0:01:20.240 --> 0:01:23.080 a very interesting question. For one thing, I don't know 0:01:23.120 --> 0:01:25.840 that there's any meaningful way to answer it. However, I 0:01:25.920 --> 0:01:29.120 do think we can be a little more precise if 0:01:29.160 --> 0:01:32.399 we think about it in terms of kind of a 0:01:32.400 --> 0:01:36.280 thought experiment, a test. So imagine this is the situation 0:01:36.920 --> 0:01:40.560 you find yourself in. You go into a room and 0:01:40.640 --> 0:01:43.760 there's a computer terminal there, and that's it. You know, 0:01:43.800 --> 0:01:45.920 you can't see into any other rooms or anything. It's 0:01:45.959 --> 0:01:48.800 just a desk with a computer terminal. And you sit 0:01:48.880 --> 0:01:52.760 down at this terminal and there's a little prompt there 0:01:52.880 --> 0:01:55.640 that lets you get into a chat session. And you 0:01:55.800 --> 0:01:59.640 enter into this chat session, and you have five minutes 0:01:59.760 --> 0:02:02.640 and you can ask the person on the other end 0:02:02.680 --> 0:02:05.560 of the chat session any questions you want within that 0:02:05.600 --> 0:02:08.760 five minute time frame. And once those five minutes are up, 0:02:08.919 --> 0:02:12.959 you're asked to determine was the person on the other 0:02:13.120 --> 0:02:16.120 end of the chat an actual human being or was 0:02:16.160 --> 0:02:20.200 it a computer program? Was it some form of artificial intelligence? 0:02:20.280 --> 0:02:23.200 A bot is what we would call it today. And 0:02:23.800 --> 0:02:28.600 if you are unable to determine whether the subject on 0:02:28.639 --> 0:02:31.120 the other end of the chat is human or a 0:02:31.160 --> 0:02:35.720 bot to any reliable degree, then you could say that, oh, 0:02:35.880 --> 0:02:40.239 that that program passed the Turing test, I could find 0:02:40.720 --> 0:02:45.399 no way of telling the difference between that computer program 0:02:45.440 --> 0:02:48.919 and an actual person. Touring suggests that due to advancements 0:02:48.919 --> 0:02:52.320 in computer science, he suspected that people would have at 0:02:52.440 --> 0:02:56.160 best a success rate of around seventy percent to be 0:02:56.200 --> 0:02:59.160 able to tell whether or not the quote unquote person 0:02:59.200 --> 0:03:00.720 on the other end of the church chat was a 0:03:00.800 --> 0:03:04.320 human being or a computer program, And he said he 0:03:04.400 --> 0:03:06.720 expected that to be capable in just a few years time. 0:03:06.880 --> 0:03:10.040 It took a little bit longer than that, but I 0:03:10.080 --> 0:03:14.120 would say that with the sophistication we've reached with chatbots 0:03:14.200 --> 0:03:17.640 these days, I think you could fairly conclusively say that 0:03:17.680 --> 0:03:21.600 we've got programs out there that can quote unquote beat 0:03:21.639 --> 0:03:24.240 the Turing test. Part of the problem is that Turing 0:03:24.360 --> 0:03:28.000 was saying that in the future, these programs are going 0:03:28.040 --> 0:03:30.560 to be sophisticated enough that they will fool people into 0:03:30.560 --> 0:03:33.720 thinking it's another person. He wasn't saying, oh, you have 0:03:33.800 --> 0:03:38.480 to meet this specific threshold for your system to have 0:03:39.560 --> 0:03:43.960 achieved beating the Turing test. That would come afterward. Other 0:03:44.000 --> 0:03:47.720 people would kind of create the criteria. But since then 0:03:47.800 --> 0:03:51.320 people have subsequently used the phrase Turing test to reference. 0:03:51.480 --> 0:03:54.520 Essentially any kind of task designed to determine if a 0:03:54.560 --> 0:03:58.720 machine has or at least appears to have the property 0:03:59.280 --> 0:04:01.800 of intell and when I say that, I mean really 0:04:01.960 --> 0:04:07.920 general intelligence. But there's another specific use of Turing tests 0:04:08.000 --> 0:04:10.840 that I would like to bring up today, and that 0:04:11.040 --> 0:04:16.080 is the completely automated public Turing test to tell computers 0:04:16.200 --> 0:04:21.360 and humans apart that once you turn it into an acronym, 0:04:21.440 --> 0:04:25.080 becomes cap chop. So these are those little tasks that 0:04:25.160 --> 0:04:29.520 you occasionally encounter on certain websites, and they require you 0:04:29.640 --> 0:04:32.400 to do something like you type in a string of 0:04:32.480 --> 0:04:36.280 characters that are displayed on screen. They're usually deformed in 0:04:36.360 --> 0:04:40.040 some way and up against a crazy background. Or you 0:04:40.160 --> 0:04:42.840 might be given a big selection of images and told 0:04:42.880 --> 0:04:45.320 to pick out all the ones that have a cat 0:04:45.360 --> 0:04:48.120 in them or something. Or you might have to drag 0:04:48.279 --> 0:04:51.839 a little picture of a puzzle piece into an image 0:04:52.000 --> 0:04:55.279 where it fits into a very specific spot. All of 0:04:55.320 --> 0:04:58.919 these are meant to separate out actual human visitors to 0:04:59.240 --> 0:05:03.560 a website service versus all the automated programs or bots 0:05:03.640 --> 0:05:05.560 or whatever you might want to call it. So I 0:05:05.560 --> 0:05:07.640 thought it would be fun to do a quick episode 0:05:07.880 --> 0:05:11.640 on where captures came from and what purpose they serve 0:05:12.240 --> 0:05:14.560 kind of touched on that already, but also how they 0:05:14.560 --> 0:05:18.160 fit into the grand picture of artificial intelligence, because interestingly 0:05:18.480 --> 0:05:22.039 they play a pretty important part. They have helped drive 0:05:22.240 --> 0:05:26.640 the development and advancement of artificial intelligence, not necessarily in 0:05:26.680 --> 0:05:31.080 a way that is helpful to everybody out there, but 0:05:31.120 --> 0:05:34.480 it certainly has served as a way to get people 0:05:34.960 --> 0:05:39.280 thinking about how to tackle certain AI problems. So our 0:05:39.320 --> 0:05:43.599 story actually begins with the good old website Yahoo. Y'all 0:05:43.600 --> 0:05:46.360 remember Yahoo. I mean, it's still a thing, but I 0:05:46.400 --> 0:05:50.720 remember a time where Yahoo was practically synonymous with Internet 0:05:50.800 --> 0:05:53.159 for a lot of folks. You may not even remember 0:05:53.200 --> 0:05:56.000 this if you haven't been on Yahoo in ages, but 0:05:56.480 --> 0:05:59.400 once upon a time, Yahoo was sort of an ordal 0:05:59.640 --> 0:06:02.159 to the rest of the Internet. Yahoo was kind of 0:06:02.160 --> 0:06:03.919 like a landing page. A lot of people had it 0:06:04.000 --> 0:06:06.960 set as their homepage, so when they would go into 0:06:07.000 --> 0:06:10.440 a web browser, they'd go right into Yahoo and you 0:06:10.440 --> 0:06:13.919 would find articles there and all sorts of other links, 0:06:14.279 --> 0:06:18.000 as well as chat rooms and of course the search 0:06:18.040 --> 0:06:21.440 engine where you could search for other stuff online besides 0:06:21.520 --> 0:06:24.680 the stuff that just popped up on Yahoo. Well, in 0:06:24.720 --> 0:06:29.240 those chat rooms, moderators were running into a really serious problem. 0:06:29.560 --> 0:06:35.440 The chat spaces were becoming invaded by bots posing as people. 0:06:35.600 --> 0:06:37.520 Now this is in two thousand. The bots were not 0:06:37.560 --> 0:06:41.919 particularly sophisticated, but they were creating a lot of spam, Like, 0:06:41.960 --> 0:06:45.880 they were jamming up chat spaces with just spam messages 0:06:45.920 --> 0:06:48.560 while people are trying to chat. In some cases, they 0:06:48.560 --> 0:06:52.279 were gathering personal information of users in an effort to 0:06:52.839 --> 0:06:56.719 exploit those users in some way or another. So Yahoo 0:06:56.720 --> 0:06:59.320 didn't want this to keep going. It wasn't reflecting well 0:06:59.320 --> 0:07:02.440 on the companies. So they turned to the computer science 0:07:02.480 --> 0:07:06.400 department at Carnegie Mellon University in order to see, Hey, 0:07:06.640 --> 0:07:10.200 is there some way that we could, you know, kind 0:07:10.200 --> 0:07:13.400 of like have a bouncer out front, a gate keeper 0:07:13.640 --> 0:07:18.040 if you will, that would allow humans into the various 0:07:18.080 --> 0:07:19.880 systems so that they can make use of them the 0:07:19.880 --> 0:07:23.800 way they were intended, but prevent all the robots, all 0:07:23.840 --> 0:07:28.000 the AI programs, all the computer software or algorithms, however 0:07:28.000 --> 0:07:31.440 you want to define it, keep them from getting access. 0:07:31.680 --> 0:07:36.960 So a team led by Manuel Bloom and including folks 0:07:37.040 --> 0:07:40.120 or Blum, I suppose, and including folks like John Langford, 0:07:40.360 --> 0:07:45.800 Louis von On, Nicholas Hopper, and others tackled this challenge, 0:07:46.200 --> 0:07:49.160 so they needed to come up with a test. Now, 0:07:49.240 --> 0:07:53.200 in an ideal world, the test would be a cinch 0:07:53.480 --> 0:07:56.520 for a human being to complete, but it would be 0:07:56.560 --> 0:08:01.840 a real stumper for algorithmically driven by And that is 0:08:01.920 --> 0:08:06.520 the basic philosophy of capture. Make a test that humans 0:08:06.560 --> 0:08:10.800 find really easy to complete, perhaps even trivial, like it's 0:08:10.960 --> 0:08:15.360 just a mild inconvenience, as they say, but for bots 0:08:15.640 --> 0:08:19.040 it's like a turn away. You're never going to be 0:08:19.080 --> 0:08:21.720 able to get this. Now, some of y'all might be 0:08:21.760 --> 0:08:24.920 saying something along the lines of but Jonathan, whenever I 0:08:25.080 --> 0:08:29.320 run into captures these days, they're sometimes really hard, Like 0:08:29.520 --> 0:08:32.560 it's hard to see what they spell out. I'll try 0:08:32.559 --> 0:08:35.439 and type things in three four times and get kicked out. 0:08:35.760 --> 0:08:38.839 And you're right, that is a problem. It is something 0:08:38.880 --> 0:08:42.720 that actually is happening. It doesn't mean that you're not human. 0:08:43.400 --> 0:08:46.280 If you're having like existential crises. I would like to 0:08:46.320 --> 0:08:49.400 set your mind at ease by saying you're probably human. 0:08:49.800 --> 0:08:52.520 I mean, I don't think I could say anything for certain, 0:08:52.559 --> 0:08:55.480 but I feel fairly confidence saying you're probably human. But 0:08:55.880 --> 0:08:59.719 the reason why captures have become really difficult in some 0:08:59.800 --> 0:09:03.240 way cases anyway with some specific types of captures is 0:09:03.360 --> 0:09:07.800 largely because other programmers figured out how to make better 0:09:07.840 --> 0:09:12.000 automated programs that can parse and respond to captures. So 0:09:12.200 --> 0:09:14.920 as one group of programmers figured out how to design 0:09:15.080 --> 0:09:19.360 tools to defeat a capture, the captured designers would go 0:09:19.440 --> 0:09:23.120 back to the drawing board to create new tests to 0:09:23.160 --> 0:09:26.320 be more challenging for those bots, to say, well, they 0:09:26.559 --> 0:09:31.280 got good at this, let's change these things and reintroduce 0:09:31.400 --> 0:09:36.560 the capture so that this will trip up those systems 0:09:36.600 --> 0:09:39.560 because while they're good at what we used to use 0:09:40.320 --> 0:09:45.920 for gatekeeping, they've never run into this before, and unfortunately 0:09:46.080 --> 0:09:49.360 that sometimes means that the tests become more challenging for 0:09:49.440 --> 0:09:52.440 human beings as well. It no longer is a case 0:09:52.480 --> 0:09:56.160 where something is trivial for a human but difficult for robots, 0:09:56.240 --> 0:09:59.679 at least for certain types of captures. And that's particularly 0:09:59.720 --> 0:10:03.040 true so if the human has some impairments like if 0:10:03.120 --> 0:10:07.000 they have color blindness for example, or some other visual 0:10:07.480 --> 0:10:11.880 or impairment like there are real issues in making captures 0:10:11.920 --> 0:10:14.720 that do what they're supposed to do, that is weed 0:10:14.760 --> 0:10:18.320 out all the non humans but also be accessible to 0:10:18.520 --> 0:10:21.800 all humans, even those who might have impairments that would 0:10:21.840 --> 0:10:25.640 otherwise make it difficult or challenging to complete a capture. 0:10:25.960 --> 0:10:29.320 It is not an easy path to walk. We're going 0:10:29.400 --> 0:10:31.079 to take a quick break. When we come back, i'll 0:10:31.200 --> 0:10:46.200 talk more about the capture story we're back. So in 0:10:46.280 --> 0:10:50.400 the early days of captures, they mostly took on the 0:10:50.440 --> 0:10:55.600 form of distorted text that was printed over a busy background. 0:10:55.880 --> 0:10:58.960 And the idea was that most automated programs would not 0:10:59.040 --> 0:11:02.360 be able to recognize distorted texts like it would be 0:11:02.400 --> 0:11:05.760 an image, not just text letters where it would be 0:11:05.800 --> 0:11:08.360 able to read like the code used to generate the 0:11:08.440 --> 0:11:10.960 letters and then say, oh, well that's these letters that 0:11:11.000 --> 0:11:13.839 can replicate that and get through no problem. You had 0:11:13.840 --> 0:11:18.160 to have something that was going to really stump them. Now, 0:11:18.240 --> 0:11:21.480 image recognition is a pretty tricky science. I've talked about 0:11:21.480 --> 0:11:25.160 it on this show before, Like, training computer systems to 0:11:25.240 --> 0:11:28.960 recognize images takes a lot of time and effort and 0:11:29.040 --> 0:11:32.200 lots and lots and lots of samples so that the 0:11:32.280 --> 0:11:37.640 computer system can quote unquote learn what those images represent. Now, 0:11:37.679 --> 0:11:40.559 it's one thing to teach a computer how to recognize 0:11:40.920 --> 0:11:44.600 standard letters that are in a recognizable font. So if 0:11:44.640 --> 0:11:48.280 the Internet only ever used one font and only used 0:11:48.320 --> 0:11:52.839 one size of that font. Then it would be relatively 0:11:53.120 --> 0:11:56.400 trivial for those who want to defeat captures, because once 0:11:56.400 --> 0:11:58.720 you train a computer vision system on what a lower 0:11:58.760 --> 0:12:02.480 case T looks like, for example, then the system would 0:12:02.480 --> 0:12:05.160 recognize a lowercase tea every time one popped up. But 0:12:05.280 --> 0:12:08.400 of course, there are lots of different fonts and typefaces 0:12:08.440 --> 0:12:11.080 on the Internet, and they come in different sizes and 0:12:11.200 --> 0:12:15.240 colors and on different backgrounds. So teaching a computer system 0:12:15.280 --> 0:12:18.520 what a times new Roman lowercase tea looks like against 0:12:18.559 --> 0:12:21.439 a blank background doesn't mean it's also going to recognize 0:12:21.440 --> 0:12:25.160 a lowercase tea and some other font on some crazy background. 0:12:25.320 --> 0:12:28.040 Plus maybe the tea is a little wavy, a little distorted, 0:12:28.440 --> 0:12:31.640 so distorting that text makes it more challenging for image 0:12:31.679 --> 0:12:36.200 recognition systems, like they're looking for defining features to be 0:12:36.360 --> 0:12:40.040 able to recognize the image of a letter with the 0:12:40.120 --> 0:12:43.280 actual letter. You see, humans, when we teach a human 0:12:43.440 --> 0:12:46.880 what something looks like, it's a lot easier for humans 0:12:46.920 --> 0:12:50.280 to associate other things that look kind of the way 0:12:50.880 --> 0:12:54.160 the first example did, but maybe not exactly the same. 0:12:54.640 --> 0:12:57.640 So in other words, like the example, I always use 0:12:57.640 --> 0:13:00.160 our coffee mugs right. If I show you a coffee mugdug, 0:13:00.200 --> 0:13:02.199 and I say this is a coffee mug, and then 0:13:02.240 --> 0:13:05.240 I show you a second kind that looks totally different, 0:13:05.280 --> 0:13:08.520 different color, different size, you know, whatever, maybe has different 0:13:08.520 --> 0:13:10.800 writing on it, whatever it might be. And I say, 0:13:10.840 --> 0:13:13.400 this is also a coffee mug. And then I show 0:13:13.400 --> 0:13:16.200 you a third example that looks unlike the first two. 0:13:16.440 --> 0:13:18.800 You could say, oh, okay, I get the idea. I 0:13:18.840 --> 0:13:23.200 get the different features that make up what a coffee 0:13:23.280 --> 0:13:26.560 mug is. I understand now. And now when I encounter 0:13:27.160 --> 0:13:29.840 different types of coffee mugs, even though they might not 0:13:29.920 --> 0:13:32.559 look anything like any of the other ones I've encountered, 0:13:32.720 --> 0:13:36.360 I know, Okay, that's probably a coffee mug. Until someone says, no, 0:13:36.440 --> 0:13:39.160 that's a teacup, and then your world is turned upside down. 0:13:39.200 --> 0:13:41.880 But you get what I'm saying. Computers don't work that way. 0:13:42.200 --> 0:13:45.840 Computers like, if you teach it an example is a thing, 0:13:46.320 --> 0:13:52.120 it doesn't necessarily understand that similar but distinctly different versions 0:13:52.160 --> 0:13:55.200 of that same thing fall into the same category. That 0:13:55.320 --> 0:13:58.920 takes lots and lots of training. So the whole idea 0:13:59.080 --> 0:14:02.760 of distortion was that this would make it very tricky 0:14:03.120 --> 0:14:06.040 for most systems to be able to parse that information 0:14:06.320 --> 0:14:09.080 and be able to put it in reliably and to 0:14:09.240 --> 0:14:13.200 fool the capture system doesn't mean that it was fool proof. 0:14:13.360 --> 0:14:16.720 Over time, those systems did get better at being able 0:14:16.760 --> 0:14:20.640 to recognize those figures that were on screen, even better 0:14:20.800 --> 0:14:25.360 than humans could in some cases, which is obviously a problem. Now, 0:14:25.400 --> 0:14:28.480 there have been lots of other capture systems, not just Capture. 0:14:28.760 --> 0:14:34.400 For example, there's one called Asira Asira. Asira did something 0:14:34.480 --> 0:14:37.200 I mentioned earlier in the episode. It would present the 0:14:37.360 --> 0:14:40.680 visitor with a collection of photographs and they would include 0:14:40.760 --> 0:14:43.960 cats and dogs, and it would ask you, okay, identify 0:14:44.200 --> 0:14:47.240 the pictures that have cats in them. So that was 0:14:47.280 --> 0:14:50.920 one way to get around this was that it wasn't 0:14:51.040 --> 0:14:55.240 just figuring out text. It was differentiating between cats and dogs, 0:14:55.560 --> 0:14:59.080 something that again computer systems couldn't do just natively. They 0:14:59.120 --> 0:15:02.600 had to be taught how to recognize the features that 0:15:02.640 --> 0:15:05.200 belonged to a cat versus those that belonged to a dog, 0:15:05.720 --> 0:15:08.720 just the same as all other image recognition software. The 0:15:08.760 --> 0:15:13.280 folks over at Google developed Recapture, and that actually served 0:15:13.280 --> 0:15:17.640 a dual purpose. It was kind of sneaky. So with Recapture, 0:15:17.720 --> 0:15:19.400 you would go to a website and you would be 0:15:19.440 --> 0:15:22.400 greeted by some you know, kind of grainy text, and 0:15:22.440 --> 0:15:24.320 you'd be asked to type it out. You'd actually get 0:15:24.360 --> 0:15:28.120 a couple of different ones, not just one. And this 0:15:28.320 --> 0:15:32.920 text was from scans made of physical digitized books, so 0:15:32.960 --> 0:15:36.200 in other words, books where they had put the page 0:15:36.320 --> 0:15:39.200 down on a scanner and created a scan. So some 0:15:39.240 --> 0:15:41.800 of these books were in you know, pretty bad shape. 0:15:41.840 --> 0:15:46.200 They were at all crisp, clear images. So your first 0:15:46.480 --> 0:15:50.720 capture you'd be presented with, Google actually knew the answer 0:15:50.880 --> 0:15:53.640 to whatever the word was. So let's say the word 0:15:54.000 --> 0:15:58.800 is salamander and you type in salamander, and so Google says, 0:15:58.800 --> 0:16:04.200 all right, I already knew that this scanned word is salamander. 0:16:04.640 --> 0:16:06.960 This is obviously a person who has typed this in. 0:16:07.280 --> 0:16:11.160 But the second image would be a scan from a book. 0:16:11.200 --> 0:16:13.720 Maybe it'd be a really smudged one, like one that's 0:16:13.760 --> 0:16:16.840 harder to read, and it would ask you, okay, was 0:16:16.880 --> 0:16:20.320 this word. Let's say it's surgeon and you type insurgeon. Well, 0:16:20.360 --> 0:16:25.200 the secret sauce here is that Google didn't know that 0:16:25.200 --> 0:16:28.280 that scanned word was surgeon. What Google was doing was 0:16:28.320 --> 0:16:33.560 crowdsourcing crowdsourcing the effort to figure out what the text 0:16:33.840 --> 0:16:39.400 in this scanned image actually said. So if you and 0:16:40.000 --> 0:16:43.280 thousands of other people all put the same word in 0:16:43.760 --> 0:16:47.360 when you were encountering this particular scan, Google would say, 0:16:47.400 --> 0:16:51.840 all right, that word is very likely surgeon. Because you know, 0:16:52.040 --> 0:16:55.280 ninety eight percent of the people who were shown this 0:16:55.680 --> 0:16:59.480 recapture typed surgeon in. So now we know that that 0:16:59.520 --> 0:17:04.880 word is which meant that they could essentially transcribe these 0:17:04.960 --> 0:17:09.840 digitized texts by using the crowd to do the work 0:17:09.880 --> 0:17:13.159 for them. And that is kind of the heart of 0:17:13.280 --> 0:17:17.840 where capture and AI meet. That captures have been used 0:17:18.119 --> 0:17:21.880 one to help train AI so that it's more effective. 0:17:21.920 --> 0:17:24.119 Like if you've encountered other Google ones where it's like 0:17:24.320 --> 0:17:26.840 pick all the images here that have motorcycles in them 0:17:27.119 --> 0:17:31.600 or stairs. Well, part of that is training Google's image 0:17:31.640 --> 0:17:37.120 recognition systems so that they're more accurate. Right, Like an 0:17:37.160 --> 0:17:41.359 image recognition system might have trouble differentiating an actual like 0:17:41.440 --> 0:17:44.920 stone staircase out in front of a building with a 0:17:44.960 --> 0:17:48.800 pedestrian crosswalk, because you know you've got those those broken 0:17:48.920 --> 0:17:52.760 lines on a crosswalk, those could look like stairs to 0:17:53.000 --> 0:17:57.440 a computer image recognition system. So by giving users the 0:17:57.480 --> 0:18:01.080 task of hey, identify all the excit samples in this 0:18:01.359 --> 0:18:05.080 list that have stairs in them, Google starts to train 0:18:05.200 --> 0:18:09.119 its own image recognition algorithms to be more effective and 0:18:09.200 --> 0:18:13.000 more accurate. So in a way, we were essentially being 0:18:13.160 --> 0:18:18.640 used as free labor to make these AI systems more accurate, 0:18:19.000 --> 0:18:20.800 just so that we could get access to whatever it 0:18:20.920 --> 0:18:23.719 was we were trying to visit, whether that was an 0:18:23.800 --> 0:18:27.159 online shop or a chat room, or you know, whatever 0:18:27.200 --> 0:18:31.719 it might be. So, yeah, we we've been working for free, y'all. 0:18:32.080 --> 0:18:35.240 Actually it's in some cases we've been working for free 0:18:35.359 --> 0:18:38.880 and denied access to tools that we wanted to use 0:18:39.040 --> 0:18:41.240 because the captures were too hard for us to be 0:18:41.280 --> 0:18:44.480 able to solve. But yeah, that's that's the quick story 0:18:44.720 --> 0:18:48.400 about the history and evolution of captures. Clearly they're still 0:18:48.480 --> 0:18:51.600 used today. Sometimes it's something simple like click this box 0:18:51.640 --> 0:18:54.040 to prove your human that kind of thing where it 0:18:54.080 --> 0:18:57.400 requires you to take an action. Those obviously are much 0:18:57.440 --> 0:19:02.159 more simple for humans to comp than for robots, so 0:19:02.240 --> 0:19:06.080 those still follow the philosophy of the original captions. A 0:19:06.119 --> 0:19:08.840 lot of other ones, though they get pretty tricky, to 0:19:08.880 --> 0:19:11.720 the point where sometimes I'm discouraged from even going further 0:19:12.080 --> 0:19:15.080 and visiting the website in particular, just like, you know what, 0:19:15.400 --> 0:19:18.520 I don't need to feel stupid because I couldn't find 0:19:18.560 --> 0:19:22.440 all the fire hydrants in these photographs, so I'm just out. 0:19:22.720 --> 0:19:25.080 But yeah, that's it. And like I said, it plays 0:19:25.119 --> 0:19:27.320 a really important part with AI. It's kind of a 0:19:27.359 --> 0:19:33.159 seesaw effect, right, Like you create a barrier that AI 0:19:33.359 --> 0:19:36.080 can't get over until it can, and then you have 0:19:36.119 --> 0:19:39.520 to go back and create a harder barrier. And meanwhile, 0:19:39.680 --> 0:19:42.840 the folks developing the AI keep making advancements that the 0:19:42.880 --> 0:19:46.600 AI gets more sophisticated and powerful over time. So yeah, 0:19:46.680 --> 0:19:50.960 delicate balance and not everybody benefits. As I said, hope 0:19:51.000 --> 0:19:54.200 that that was interesting and informative to y'all. I hope 0:19:54.240 --> 0:19:56.919 you're all doing well, and I'll talk to you again 0:19:57.480 --> 0:20:07.640 really soon. Tech Stuff is an iHeartRadio production. For more 0:20:07.720 --> 0:20:12.480 podcasts from iHeartRadio, visit the iHeartRadio app, Apple Podcasts, or 0:20:12.480 --> 0:20:14.439 wherever you listen to your favorite shows.