WEBVTT - How AI Facial Recognition Works 0:00:00.720 --> 0:00:03.240 Hey everybody, it's me Josh, and I'm here to tell 0:00:03.240 --> 0:00:06.240 you it's official. We're going to be in Vancouver, b C. 0:00:06.720 --> 0:00:10.880 And Portland, Oregon this March. On March twenty nine, will 0:00:10.920 --> 0:00:13.200 be at the Chance Center in Vancouver and on March 0:00:13.840 --> 0:00:16.920 will be at the Arlene Schnitzer Concert Hall in Portland. 0:00:17.200 --> 0:00:20.920 So come see us. Tickets go on sale this Friday. 0:00:21.040 --> 0:00:23.919 Go to s Y s K live dot com for 0:00:24.000 --> 0:00:26.759 ticket links and info and everything you need and we'll 0:00:26.760 --> 0:00:30.600 see you guys in March. Welcome to Stuff You Should Know, 0:00:30.840 --> 0:00:39.280 a production of I Heart Radios How Stuff Works. Hey, 0:00:39.440 --> 0:00:45.080 welcome to the podcast. I'm Josh Clark. There's Charles Scooter, 0:00:45.200 --> 0:00:52.000 Computer Bryant and Jerry Um Matthew Broderick in War Games 0:00:52.440 --> 0:00:58.320 Rolling and this is Should Know. That's good. Thanks. What's 0:00:58.360 --> 0:01:04.600 your name, Josh m hmm? Okay, Josh, Ali Shedy in 0:01:04.680 --> 0:01:08.360 More Games Okay Clark, I know Aaron Cooper is doing 0:01:08.480 --> 0:01:11.760 right now? Man? How cute was she in that movie? Like? 0:01:11.800 --> 0:01:14.959 I think they designed that movie for like every thirteen 0:01:15.040 --> 0:01:16.959 year old boy in America's fall in love with Ali 0:01:16.959 --> 0:01:21.800 shed I think you're talking about um short Circuit. I 0:01:21.840 --> 0:01:24.959 never saw that, believe that what with Johnny five. I mean, 0:01:25.000 --> 0:01:27.880 I know the movie. You gotta see it. Really, it's 0:01:27.920 --> 0:01:35.160 pretty awful, especially with um oh, what's his name? Who 0:01:35.240 --> 0:01:38.039 is the sleeves ball from Fast Times at Ridgemont High? 0:01:38.040 --> 0:01:44.720 Who is the ticket scalper? Yes, he plays a Indian 0:01:45.520 --> 0:01:52.920 like Asian Indian character, like full on brown face and everything. 0:01:53.360 --> 0:01:57.200 It's really bad. The movie is bad enough, but then 0:01:57.240 --> 0:01:59.600 now when you go back and see that, to your like, 0:01:59.680 --> 0:02:02.400 I can't believe this. I can't believe it. I think 0:02:02.440 --> 0:02:06.440 he's Italian, oh easily, maybe Jewish or maybe just a 0:02:06.480 --> 0:02:09.080 straight up white guy. He's definitely not Asian Indian. No, 0:02:09.280 --> 0:02:12.720 he's not. But anyway, go see Short Circuit. Okay, see 0:02:12.720 --> 0:02:14.919 what you think Ali shed He just keeps looking at 0:02:14.919 --> 0:02:18.120 the camera going, I'm so sorry that was a big 0:02:18.160 --> 0:02:20.760 hit though she didn't have anything be sorry about. Um So. 0:02:20.880 --> 0:02:25.480 I guess War Games is kind of in your your wheelhouse. Yeah, 0:02:25.800 --> 0:02:28.680 it was a little old for me. Yeah, because I 0:02:28.720 --> 0:02:32.480 saw that when I was like twelve, bish uh, and 0:02:32.600 --> 0:02:35.280 you would have been I don't know how how much 0:02:35.320 --> 0:02:38.359 younger are you. I would have been seven seven. Yeah, 0:02:38.440 --> 0:02:40.000 that's a little young for War Games, I would think. 0:02:40.040 --> 0:02:42.560 I mean, I still watched your wherever, but I was like, yeah, 0:02:42.600 --> 0:02:45.160 that's hilarious. The computer is called Whopper. Yeah, I mean 0:02:45.200 --> 0:02:47.799 it was right in my wheelhouse, right. I remember at 0:02:47.800 --> 0:02:51.280 the end of war Games, Uh, they lock in, you know, 0:02:51.320 --> 0:02:55.360 they're decoding the the code one number and letter at 0:02:55.400 --> 0:02:58.200 a time. Very suspenseful and yeah, very suspenseful. And it 0:02:58.200 --> 0:03:01.080 finally locks in and me and my friends memerized it 0:03:01.120 --> 0:03:03.040 so we could go home and plug it into an 0:03:03.040 --> 0:03:08.840 Apple too to see what happened. Nothing, Okay, how do 0:03:08.840 --> 0:03:11.440 you plug in a number anyway? What does that even mean? Oh? Yeah, 0:03:11.520 --> 0:03:15.400 that's true, you know. Yeah. I remember a very rudimentary 0:03:16.840 --> 0:03:21.520 program you could run where you could type in like 0:03:21.560 --> 0:03:24.280 four lines of whatever I don't even know if you 0:03:24.320 --> 0:03:27.639 call it code with a phrase, and it would run 0:03:27.680 --> 0:03:30.560 the phrase like a thousand times all over your screen 0:03:30.560 --> 0:03:33.280 and a big scroll. And then that just thought that 0:03:33.360 --> 0:03:36.360 was the coolest thing ever. I feel like, I remember 0:03:36.440 --> 0:03:39.000 what you're talking just like five lines was like the 0:03:39.040 --> 0:03:42.160 only part I remember is twenty go to ten and 0:03:42.280 --> 0:03:47.160 ten was the phrase I think something like I don't remember. 0:03:47.440 --> 0:03:49.560 Then I was like, man, let's just play Castle Wolfenstein. 0:03:49.920 --> 0:03:52.320 That was a good one. I remember, did Oregon Trail. 0:03:52.680 --> 0:03:55.040 I never did as well. I was Castle Wolfenstein, but 0:03:55.160 --> 0:03:57.600 like wolf and signed on the PC. Oh yeah, like 0:03:57.840 --> 0:04:02.320 move left arrow, right arrow, shoot uh shoot dash, which 0:04:02.320 --> 0:04:04.560 is some sort of a bullet. Yeah. It was fun. 0:04:04.720 --> 0:04:06.480 That was fun, and I thought it was just like 0:04:06.520 --> 0:04:10.480 the height of technological gaming. It was it was at 0:04:10.480 --> 0:04:13.160 the time. But now, Chuck, we've reached the height of 0:04:13.200 --> 0:04:17.360 technology which is being tracked everywhere you go, look at 0:04:17.400 --> 0:04:20.680 you all the time by whoever wants to do that. 0:04:20.760 --> 0:04:22.960 I'm gonna change your name to Josh smooth up a 0:04:23.040 --> 0:04:27.400 Raider Clark for that transition. Very nice. Yeah, Like I 0:04:27.440 --> 0:04:31.800 like Ali Shedy Clark, Josh, Ali Shedy and Clark. Yeah, 0:04:31.880 --> 0:04:35.000 this is a good one. Uh did you put this together? 0:04:35.040 --> 0:04:37.440 It was this you and this is Dave. Dave Brews 0:04:37.880 --> 0:04:40.279 good stuff in high Dave. We finally got to meet 0:04:40.560 --> 0:04:45.080 Dave and his family, lovely family that we cursed awfully 0:04:45.120 --> 0:04:48.560 in front of in Seattle. I felt terrible, thanks for 0:04:48.600 --> 0:04:51.800 adding yourself to the max and he was like, you 0:04:51.839 --> 0:04:55.640 know whatever, he was fine. His kids were adorable, they 0:04:55.640 --> 0:04:57.680 were they were great. They couldn't look at me though 0:04:58.040 --> 0:05:00.800 that really they were probably just him did by your present, 0:05:00.839 --> 0:05:03.880 No No, it was because I cursed so badly. So 0:05:03.960 --> 0:05:08.279 this is good stuff. Though. Facial recognition technology that they've 0:05:08.320 --> 0:05:12.560 been kind of at since the nineteen fifties, which they 0:05:12.640 --> 0:05:15.360 rolled out as a test in two thousand two at 0:05:15.360 --> 0:05:18.600 the Super Bowl in New Orleans, did not go that well. No, 0:05:19.320 --> 0:05:21.640 it was a little clunky back then, but it's gotten 0:05:21.640 --> 0:05:24.080 a lot better since then. Let me explain why. For 0:05:24.080 --> 0:05:25.880 anyone who's listened to the End of the World with 0:05:25.960 --> 0:05:31.080 Josh Clark the AI episode in particular, everything associated with 0:05:31.240 --> 0:05:35.680 artificial intelligence got way better starting around two thousand seven, 0:05:36.080 --> 0:05:41.320 when neural nets became a viable form of machine learning. 0:05:41.640 --> 0:05:46.640 Because you don't have to train a computer what constitutes 0:05:46.640 --> 0:05:48.880 a human face and what to look for. You just 0:05:49.040 --> 0:05:51.400 feeded a bunch of pictures of faces and say these 0:05:51.440 --> 0:05:54.840 are human faces, learn what a human faces and they 0:05:54.880 --> 0:05:57.840 trained themselves. And so around about two thousand seven, two 0:05:57.839 --> 0:06:00.680 thousand eight, two thousand nine, everything had to do with 0:06:00.760 --> 0:06:04.719 machine learning got way smarter because we started using neural 0:06:04.800 --> 0:06:08.719 nets and facial recognition software is no exception. Yeah, and 0:06:08.760 --> 0:06:10.920 there were a few things that kind of converged all 0:06:10.960 --> 0:06:14.120 at the same time or around the same time. Uh, 0:06:14.240 --> 0:06:17.159 social meds kind of coming on the scene right in 0:06:17.160 --> 0:06:22.280 that wheelhouse was a big deal. Um, Facebook, this is staggering. 0:06:22.880 --> 0:06:28.320 Facebook just by itself processes three hundred and fifty million 0:06:29.120 --> 0:06:34.400 new photos through its uh facial recognition software every day, 0:06:34.560 --> 0:06:39.359 and every time one comes through. Mark Zuckerberg goes like, 0:06:39.440 --> 0:06:41.640 you think it's neat when you go when you put 0:06:41.640 --> 0:06:43.640 a picture up and it says like would you like 0:06:43.680 --> 0:06:46.920 to tag Emily your wife because that's her? And you think, oh, well, 0:06:46.920 --> 0:06:49.880 that's super easy, thanks Facebook. But then you don't think, like, 0:06:49.880 --> 0:06:51.840 wait a minute, all right, how do they know that's 0:06:51.920 --> 0:06:56.719 my wife? And you know, it's like with everything else, Uh, 0:06:56.800 --> 0:06:59.680 there's privacy people that were like, whoa do you guys 0:06:59.680 --> 0:07:03.279 realize what's going on? And then of the sheep they're 0:07:03.320 --> 0:07:05.919 like huh no, no, they're like no, it's great, Like 0:07:05.960 --> 0:07:08.279 I don't have to go in and like make click 0:07:08.360 --> 0:07:12.360 two links or two buttons to take the way easier. 0:07:12.560 --> 0:07:15.400 So that was one thing. There's way more photos out 0:07:15.440 --> 0:07:19.400 there for those machines to learn one like good high 0:07:19.480 --> 0:07:24.920 quality photos right uty million a day just on Facebook alone, um, 0:07:25.080 --> 0:07:27.520 which means the machines were getting smarter, They were getting 0:07:27.560 --> 0:07:32.680 better and better at at training themselves. And then lastly, um, 0:07:32.760 --> 0:07:37.000 that has led to a ubiquity in in facial recognition. 0:07:37.440 --> 0:07:41.000 That the better the machines have gotten, the easier it 0:07:41.040 --> 0:07:43.280 has been to put together data sets for them to 0:07:43.320 --> 0:07:46.920 train on, which is lots and lots of pictures of people. Um, 0:07:46.960 --> 0:07:49.840 the cheaper the technology has gotten, which means the more 0:07:49.880 --> 0:07:54.440 people that are now using facial recognition than ever. Yeah. 0:07:54.480 --> 0:07:59.320 Amazon has a service called Recognition with a K, which 0:07:59.360 --> 0:08:02.080 is not a good look. No, it looks very German. 0:08:02.240 --> 0:08:06.360 There's something about replacing at sea with the K. It 0:08:06.480 --> 0:08:09.280 just looks creepy, Like when you spell America with a K, 0:08:10.280 --> 0:08:15.240 it means something. It means like bad America. Yet they 0:08:15.240 --> 0:08:20.240 went right full steam ahead and called it recognition. You 0:08:20.240 --> 0:08:21.720 have to say it like that, you do, I think, 0:08:21.720 --> 0:08:23.960 and you have to be like squeezing the air out 0:08:24.000 --> 0:08:27.560 of a syringe while you're saying it too. Um, so 0:08:27.600 --> 0:08:29.960 they have this. I didn't even know about this, but 0:08:30.160 --> 0:08:34.880 it's ubiquitous and it's not super expensive, and that means 0:08:34.920 --> 0:08:38.960 that the law enforcement agencies agencies they don't have to 0:08:39.040 --> 0:08:41.560 like create their own they can just say, well, just 0:08:41.559 --> 0:08:45.640 to sign up for a recognition right exactly. Because it's there, 0:08:45.720 --> 0:08:48.839 and because it's it's relatively cheap, you can just get 0:08:48.880 --> 0:08:52.200 a subscription, not just law enforcement agencies. If you have 0:08:52.280 --> 0:08:56.079 a photo sharing app or whatever and you want some 0:08:56.160 --> 0:09:00.880 facial recognition technology, you just contract with with Amazon, and 0:09:00.960 --> 0:09:04.000 Amazon goes here, you go, here's our data, and you 0:09:04.040 --> 0:09:06.520 can or um like our code, and you can put 0:09:06.559 --> 0:09:10.200 it onto your platform. Anybody can use it. Um. So 0:09:10.280 --> 0:09:12.760 it is kind of everywhere, and that makes a lot 0:09:12.800 --> 0:09:18.120 of people, including me, very very nervous because as this 0:09:18.200 --> 0:09:22.880 guy wood Row hurt Zog, if he's not Werner hurt Zog, brother, 0:09:23.640 --> 0:09:28.040 I'll be disappointed. But but a wood Row and a 0:09:28.120 --> 0:09:31.800 Werner come on, you know anyway. Um. Wood Row hurt 0:09:31.880 --> 0:09:38.319 Zog as a professor of computer science, and I believe privacy, 0:09:38.840 --> 0:09:43.960 civil liberty. Um. He basically says, Look, there is no 0:09:44.120 --> 0:09:47.720 way we're going to reap the benefits of facial recognition 0:09:48.080 --> 0:09:55.520 without ultimately sliding irreversibly into a dystopian surveillance state where 0:09:55.800 --> 0:09:58.320 it's happening right now, and if we don't do something 0:09:58.360 --> 0:10:01.640 about it, it's never going to change back. We're about 0:10:01.679 --> 0:10:04.559 to fully give up our privacy because it's one thing 0:10:04.600 --> 0:10:07.240 to have your phone tracked you can give up your phone, 0:10:07.840 --> 0:10:10.920 get yourself a burner phone, like you're Jesse Pinkman or 0:10:10.920 --> 0:10:13.520 something like that, and then you just throw that phone away. 0:10:13.559 --> 0:10:16.319 You can't be tracked anymore. You can't get a burner face. 0:10:16.840 --> 0:10:19.320 And if that does become a thing down the line, 0:10:19.600 --> 0:10:22.040 it'll be very, very expensive. So the average person can't 0:10:22.040 --> 0:10:26.200 get a burner face. Will be tracked by our face 0:10:26.760 --> 0:10:28.880 everywhere we go. And as we add more and more 0:10:29.120 --> 0:10:34.880 cameras and this technology becomes cheaper and cheaper, it's we 0:10:34.920 --> 0:10:36.640 will be living in a world where there will be 0:10:36.760 --> 0:10:39.920 zero privacy and will be monitored and tracked because it 0:10:39.920 --> 0:10:42.679 will be so easy, and it will be sold to 0:10:42.760 --> 0:10:45.160 us like it's being sold to us now that it's 0:10:45.200 --> 0:10:48.600 a law enforcement tool to get the bad guys. But 0:10:48.720 --> 0:10:51.760 it's eventually going to extend to include everybody. But what 0:10:51.840 --> 0:10:54.240 do you have to worry about. You're an upstanding citizen. 0:10:54.320 --> 0:10:57.440 It doesn't matter if you're tracked. That's not true. That's 0:10:57.520 --> 0:11:00.680 just not the case, everybody, It's not case. All right, 0:11:00.679 --> 0:11:05.559 we're gonna call that soap soapbox soliloquy number one of 0:11:06.160 --> 0:11:10.240 what I guarantee will be probably three or four. Uh, 0:11:10.360 --> 0:11:12.400 let's talk of a little bit about how it works. 0:11:12.559 --> 0:11:16.280 It is biometric authentication. Um. It's like a fingerprint or 0:11:16.320 --> 0:11:19.840 a retina scan. And basically what it does is, uh, 0:11:20.240 --> 0:11:25.200 it is precise measurements of a face to calculate every 0:11:25.240 --> 0:11:29.400 person's very unique visual geometry, like how far apart your 0:11:29.400 --> 0:11:32.600 eyes are, how how far apart your pupils are from 0:11:32.600 --> 0:11:36.079 your nostrils. Yeah, your facial geometry, how your face is 0:11:36.120 --> 0:11:38.440 all set up? I think. Yeah, it's even gotten into 0:11:38.440 --> 0:11:43.000 things like facial hair, skin tone, um, skin texture. Yeah, 0:11:43.040 --> 0:11:45.240 I'm sure it'll get just more and more specific. Yeah. 0:11:45.280 --> 0:11:48.040 And then you know, because they're getting the machines are 0:11:48.040 --> 0:11:50.199 getting better and better and easier and easier to train 0:11:50.280 --> 0:11:52.520 on this stuff, you can just add more and more 0:11:52.600 --> 0:11:56.760 data to it, and the the the recognition will just 0:11:56.840 --> 0:12:02.839 become increasingly um um good. Yeah. And if you want 0:12:02.840 --> 0:12:06.600 to throw off facial recognition software and freak out every 0:12:06.679 --> 0:12:10.320 human you meet to shave your eyebrows, oh yeah, that 0:12:10.360 --> 0:12:13.080 would be a little freaky. Have you ever seen that. 0:12:13.400 --> 0:12:15.360 You've seen it before, I'm sure in movies and stuff. 0:12:15.400 --> 0:12:17.440 It's an interesting thing. I remember a kid in industrial 0:12:17.520 --> 0:12:19.440 arts class did that one year. He was like a 0:12:19.480 --> 0:12:21.839 little you know, kind of a ninth grade burnout, and 0:12:21.840 --> 0:12:24.480 he just showed up one day with no eyebrows, like 0:12:24.720 --> 0:12:27.559 you would I think like not having a nose would 0:12:27.559 --> 0:12:32.000 be more easily accepted. There's something just uncanny when someone 0:12:32.040 --> 0:12:34.920 shaves their eyebrows, Like one day they have them, the 0:12:34.920 --> 0:12:39.920 next day they don't. Was it like immediately recognizable what 0:12:39.960 --> 0:12:41.880 the thing was, or was it like that's the thing 0:12:42.760 --> 0:12:46.199 off today? Whereas if you, you know, you came in 0:12:46.240 --> 0:12:48.079 the next day without a nose, the first thing you 0:12:48.080 --> 0:12:49.600 would say is what happened to your nose? What happened 0:12:49.600 --> 0:12:51.880 to your nose? Todd? Yeah, and Todd would be like, 0:12:52.200 --> 0:12:56.560 I can't rent of it. I fell off. So those 0:12:56.600 --> 0:12:58.880 measurements we were talking about. What happens then is they 0:12:58.920 --> 0:13:05.080 compare that just like a fingerprint, with a database of images. 0:13:05.200 --> 0:13:08.480 And depending on what this is for, it could be 0:13:08.520 --> 0:13:11.080 like just within your company, or it could be the 0:13:11.120 --> 0:13:14.760 FBI's database of mugshots, or it could be the d 0:13:14.840 --> 0:13:17.959 m VS database of driver's license photos yeh, which we'll 0:13:17.960 --> 0:13:21.000 get into. Yeah. What's interesting is each stage of the way, 0:13:21.040 --> 0:13:24.760 there's a different algorithm that does you know, each increasingly 0:13:24.840 --> 0:13:29.760 sophisticated step, until you finally have basically like all of 0:13:29.800 --> 0:13:32.880 the different data points for that, you know, what makes 0:13:32.960 --> 0:13:35.880 up that facial geometry, and then you can compare it 0:13:35.920 --> 0:13:38.320 to all the other data points. We think of like 0:13:38.360 --> 0:13:43.199 a computer running like a you know, a picture, You've 0:13:43.240 --> 0:13:46.000 got your input picture and then running you know, all 0:13:46.040 --> 0:13:48.200 the pictures next to it. That's not what it's doing. 0:13:48.280 --> 0:13:51.880 It's it's running the numbers. Basically, it's doing computer stuff. Yeah. 0:13:51.880 --> 0:13:55.360 I love that first step, which is you have to 0:13:55.400 --> 0:13:59.840 teach the computer what a face is. So, I mean, 0:14:00.040 --> 0:14:02.960 seems silly, but of course that's what it is. Well, yeah, 0:14:02.960 --> 0:14:04.640 because I mean if you show it a picture of 0:14:04.840 --> 0:14:07.839 a person standing next to a fire hydra shaming on 0:14:07.880 --> 0:14:10.679 the fire Hydran, Yeah, and say hello, handsome, So this 0:14:10.720 --> 0:14:13.199 is what a human face looks like. Yeah, or no, 0:14:13.400 --> 0:14:16.600 that's a butt, And then it starts, you know, closer, closer. 0:14:16.640 --> 0:14:18.280 All right, now that's a face. You know what a 0:14:18.320 --> 0:14:21.200 face is. Now move on to step two, which is 0:14:21.560 --> 0:14:24.080 stop screwing around. Yeah, so now you know what a 0:14:24.120 --> 0:14:26.040 face is. You've got to normalize it for the photo, 0:14:26.120 --> 0:14:29.400 which means there are not that many. Well, let's you 0:14:29.440 --> 0:14:34.600 have to put it in the dockers. That normalizes, Um, 0:14:34.680 --> 0:14:37.000 you isolate that face, and then you have to make 0:14:37.040 --> 0:14:41.680 sure that it's normalized as far as looking at the camera. 0:14:42.280 --> 0:14:44.160 So if it's a if you get a photo of 0:14:44.240 --> 0:14:47.440 someone from a CCTV, let's say, and it's sort of 0:14:47.440 --> 0:14:50.600 a three quarter, they have the ability to make it 0:14:50.680 --> 0:14:52.480 as if it is looking straight at you. Yeah, the 0:14:52.520 --> 0:14:56.080 computer can pretty accurately predict what the rest of the 0:14:56.120 --> 0:15:00.720 face looks like face you know it on, I guess 0:15:00.720 --> 0:15:05.520 face on um. And when it normalizes it like that, 0:15:05.560 --> 0:15:08.600 it makes it much easier to compare to other pictures because, 0:15:09.000 --> 0:15:11.680 as we'll see, most of the pictures or most of 0:15:11.680 --> 0:15:14.160 the data points that it's comparing it to, are taken 0:15:14.200 --> 0:15:17.440 from databases of pictures that have been taking of people 0:15:17.640 --> 0:15:20.040 face on. Right. So that's why it wants to go 0:15:20.080 --> 0:15:24.040 to driver's license. Yeah, well just spoiled it. But yes, 0:15:24.200 --> 0:15:26.720 we already said that. Oh we did, Yeah, I did, Okay, 0:15:27.120 --> 0:15:32.440 I missed it, I know. So uh from there, do 0:15:32.440 --> 0:15:38.280 you have more algorithms still that isolate parts of the face. Um. 0:15:38.360 --> 0:15:40.160 And this is where my old theory that like, there 0:15:40.200 --> 0:15:43.400 are only so many sort of facial combinations, so that's 0:15:43.400 --> 0:15:45.960 why you have doppelgangers. We got to do an episode 0:15:45.960 --> 0:15:47.840 on doppel Gan. There's only so many things you can 0:15:47.880 --> 0:15:49.920 do with two eyes to eyebrows and nose in a 0:15:49.960 --> 0:15:53.880 mouth and cheekbones and in chin. Well, okay, what else? 0:15:54.560 --> 0:15:57.200 I mean, there's not a whole lot. There's lips, lips, sure, 0:15:57.800 --> 0:16:01.720 what about um uh uh uh, that's about it. These 0:16:01.720 --> 0:16:04.840 are called elevens, the ridges between your eyebrows. Well, if 0:16:04.840 --> 0:16:07.320 you want to get super specific, but that's what I'm saying. 0:16:07.320 --> 0:16:09.840 I think they're getting more and more specific. Oh yeah, yeah. 0:16:09.880 --> 0:16:12.840 But my whole point is, and we'll learn and here 0:16:12.880 --> 0:16:16.920 and facial recognition. They do use Apple gangers, but put 0:16:16.920 --> 0:16:19.960 a pin in that so they recognize all these features 0:16:20.480 --> 0:16:23.200 and then each feature becomes what's called a nodal point 0:16:24.320 --> 0:16:28.600 or no Dowell point. I think, nodal nodal, I think. 0:16:28.840 --> 0:16:31.240 And this is where you're gonna get your super exact 0:16:32.080 --> 0:16:36.040 uh angles and distances between all these parts as a 0:16:36.120 --> 0:16:41.320 flat two dimensional thing. Which my question was because below here, 0:16:41.400 --> 0:16:43.680 you know, it talks about Apple and their iPhone have 0:16:43.720 --> 0:16:47.720 a three D facial recognition. Is that is two dimensional 0:16:47.760 --> 0:16:50.360 superior to three D? I don't know, or is it 0:16:50.400 --> 0:16:53.360 just because that's what all the pictures are in the databasis, 0:16:53.360 --> 0:16:55.000 so that's what they do. I don't know. All I 0:16:55.040 --> 0:16:57.440 know is my phone usually unlocks when I look at it. 0:16:59.560 --> 0:17:00.960 You know what? I Hey, that's having to take off 0:17:01.000 --> 0:17:06.000 my sunglasses. Worst so I found I've got some wayfarers 0:17:06.480 --> 0:17:09.560 that I don't have to take, but my aviators I 0:17:09.560 --> 0:17:11.920 do have to take off. Interesting do they keep trying 0:17:11.920 --> 0:17:16.399 to make you into mav when you have on the 0:17:16.400 --> 0:17:20.399 the go ahead? What is that? That was Tom Cruise 0:17:20.640 --> 0:17:26.760 laughing and chewing gum. Okay, wow thanks, I feel like okay, 0:17:26.880 --> 0:17:29.159 um we we we gotta keep going because I was 0:17:29.200 --> 0:17:34.680 about to take a break unnecessarily. So um when the 0:17:34.800 --> 0:17:36.840 when the computer is running through the pictures, it just 0:17:36.840 --> 0:17:40.000 sis soon goes like no, no, no, no, no, millions 0:17:40.000 --> 0:17:42.720 and millions of times and then finally goes yes. But 0:17:42.800 --> 0:17:46.280 when it says yes and it spits out another picture, 0:17:46.480 --> 0:17:49.080 it's not like this is that person? No you want 0:17:49.119 --> 0:17:51.639 it to be. Because we all watch n c I S, 0:17:51.840 --> 0:17:54.760 we all watch c S I, we all watch um 0:17:55.880 --> 0:17:59.880 Law and Order, we all watch um uh party down 0:18:00.400 --> 0:18:07.239 uh Andy Griffith, that Mattlock, the whole CHAMAI um, so 0:18:07.320 --> 0:18:09.159 we wanted to just spit out and be like, here's 0:18:09.160 --> 0:18:12.720 your here's your your person of interest, right. But what 0:18:12.760 --> 0:18:16.320 it's really doing is it's it's producing a similarity score 0:18:17.520 --> 0:18:21.439 that is probabilistic. It's saying this is there's this percent 0:18:21.600 --> 0:18:24.200 chance that this is the same person as the picture 0:18:25.040 --> 0:18:27.239 or the person in the picture that you uploaded. It's 0:18:27.240 --> 0:18:30.200 a bit of a guess. It is, so a sophisticated 0:18:30.200 --> 0:18:33.880 guess it is, and the better computers get at this, 0:18:35.320 --> 0:18:38.000 the likelier it is that if they say this is 0:18:38.040 --> 0:18:41.159 a there's a chances the same person that it's the 0:18:41.200 --> 0:18:45.159 same person. But it's as we'll see, it's up to 0:18:45.200 --> 0:18:50.200 the human user to determine what is an acceptable threshold 0:18:50.359 --> 0:18:55.600 of a confidence Is it no? Is it no? Frankly, 0:18:55.600 --> 0:18:58.240 it really should be about or higher should be the 0:18:58.280 --> 0:19:02.600 confidence setting for the confidence level. Didn't that what Amazon's 0:19:02.680 --> 0:19:07.560 recognition says the threshold should be. I'm glad you said that, man, 0:19:07.640 --> 0:19:10.240 because it really is creepy and I couldn't put my 0:19:10.280 --> 0:19:13.280 finger on it. And it's exactly I mean, I knew 0:19:13.280 --> 0:19:15.360 the k looked weird or whatever, but it hadn't hit 0:19:15.440 --> 0:19:18.000 me that just how creepy it is and just how 0:19:18.080 --> 0:19:23.439 off the mark or potentially on the mark that name is. Like, Oh, like, 0:19:23.480 --> 0:19:26.200 if my name was spelled c h u K, I'm 0:19:26.200 --> 0:19:29.080 sinister a little bit, you'd be more sinister. Yeah, I 0:19:29.119 --> 0:19:32.320 don't think you could ever be truly sinister. I appreciate that. Alright, 0:19:32.400 --> 0:19:35.119 let's take a break. I'm gonna go work on sinistering 0:19:35.200 --> 0:19:37.919 up a bit, and we'll talk a little bit more 0:19:37.960 --> 0:20:04.840 about some of the uses of f R right after this. So, 0:20:04.880 --> 0:20:07.840 as with all technology, it has to be abbreviated into 0:20:07.920 --> 0:20:10.800 two letters, the second of which is are do they 0:20:10.840 --> 0:20:14.359 call it f R? I've seen it, Yes, I was 0:20:14.400 --> 0:20:17.280 just silly, but it doesn't surprise me. Nope. So in 0:20:17.800 --> 0:20:22.400 f R facial recognition technology, there are there are some 0:20:22.400 --> 0:20:25.159 some beneficial uses for it. Yeah, Like we said, you 0:20:25.280 --> 0:20:30.119 got to tag people is chief among them. The for 0:20:30.440 --> 0:20:33.600 people like you and me, that's the pinnacle as it stands. 0:20:34.040 --> 0:20:37.040 You don't have to tag people yourself. Facebook does it 0:20:37.119 --> 0:20:40.760 for you. That's what we're trading everything for. I gotta 0:20:40.800 --> 0:20:45.440 calm down. Okay. There are some other, like genuinely beneficial 0:20:45.520 --> 0:20:50.200 uses too. There's a nonprofit company called Thorn that scans 0:20:50.480 --> 0:20:56.879 missing person's pictures against um pictures of children in child 0:20:56.920 --> 0:21:01.960 porn videos, um or suspected human and trafficking to to 0:21:02.080 --> 0:21:04.879 get matches, and apparently they've rescued a hundred kids so 0:21:04.960 --> 0:21:08.560 far from using that technology. There's a pretty beneficial use 0:21:08.600 --> 0:21:13.000 of facial recognition software. Uh dating apps. Let's say you 0:21:13.119 --> 0:21:15.959 want to you can get pretty specific on what kind 0:21:16.000 --> 0:21:20.919 of face you find attractive, which is interesting. But you 0:21:20.960 --> 0:21:24.840 can say I really think, um, I like guys with 0:21:24.920 --> 0:21:28.040 high cheekbones, and but no, you would go find mall lips. 0:21:28.280 --> 0:21:32.120 It would be more like, um, somebody could be like, oh, 0:21:32.160 --> 0:21:34.679 I really find Christian Bale attractive, and they get a 0:21:34.800 --> 0:21:37.600 picture of Christian Bale into this dating app and I 0:21:37.640 --> 0:21:39.680 would come up. But I wouldn't because I wouldn't be 0:21:39.720 --> 0:21:41.720 in the dating app because I'm happily married. Do you 0:21:41.720 --> 0:21:44.879 think you look like Christian Bale? I'm told that a lot. Really, Yeah, 0:21:45.080 --> 0:21:46.760 that's weird. Don't think you look anything like it. I 0:21:46.800 --> 0:21:50.960 don't either, but people say interesting. I'm I don't know 0:21:51.000 --> 0:21:52.520 what I would do if I was dating now. I 0:21:52.520 --> 0:21:54.320 guess I would just go to a service and say 0:21:54.440 --> 0:21:57.920 and aliety type sure for more games era, But they'd 0:21:57.920 --> 0:22:00.399 be like, okay, sir, you would just upload the picture. 0:22:00.440 --> 0:22:02.720 You don't have to come into the office, which is 0:22:02.760 --> 0:22:05.000 really not even open to the public, and just tell 0:22:05.080 --> 0:22:07.919 us you're interested in Ali sheety type like a weirdo. 0:22:08.160 --> 0:22:10.239 You mean, dating apps don't have offices where they just 0:22:10.600 --> 0:22:13.920 feel complaints and interested parties. You sit down and they 0:22:13.960 --> 0:22:17.879 they videotape you with the VHS camera, put you on 0:22:17.960 --> 0:22:19.560 with some other guys on the tape. That's how they 0:22:19.640 --> 0:22:21.679 used to do it. Oh yeah, that was one of 0:22:21.720 --> 0:22:26.720 the subplots of Singles, the Cameron Crow movie. Oh yeah, 0:22:26.880 --> 0:22:29.520 it was Expect the Best was the name of the 0:22:29.600 --> 0:22:32.720 dating service, and you would make a videotape and like 0:22:32.720 --> 0:22:35.200 a watch video tapes of people you know, saying who 0:22:35.240 --> 0:22:37.840 they are. How do you remember that? I was a 0:22:37.840 --> 0:22:40.200 big singles fan. That's not a bunch I got you. Yeah, 0:22:40.280 --> 0:22:44.480 Expect the Best. You me pack of cigarettes and some coffee. 0:22:45.040 --> 0:22:49.440 We don't need anything else becausezoom tight. So what else here? 0:22:49.560 --> 0:22:53.480 This was? UM Taylor Swift and and her security team 0:22:53.480 --> 0:22:56.680 on tour used it to scan the audience to see 0:22:56.680 --> 0:23:00.240 if any of the creeps who have harassed and doctor 0:23:00.359 --> 0:23:03.399 were in the audience. That's super beneficial. No one should 0:23:03.400 --> 0:23:06.119 have to go through that. UM also, cops use it 0:23:06.320 --> 0:23:10.960 in myriad ways, but in particular especially beneficial or um 0:23:11.040 --> 0:23:14.680 when they use facial recognition to identify people who can't 0:23:14.720 --> 0:23:18.840 identify themselves. Somebody in the midst of a psychotic break, 0:23:18.880 --> 0:23:25.159 perhaps somebody wasted on troom's um, somebody you're not Jesus Christ, 0:23:25.359 --> 0:23:30.959 who who has uh amnesia, Our friend Benjamin Kyle, who 0:23:31.040 --> 0:23:34.359 apparently knows who he is now but he's decided not 0:23:34.400 --> 0:23:37.480 to disclose it publicly. Remember the guy he was found 0:23:37.480 --> 0:23:41.399 behind a burger king near a dumpster, had zero recollection 0:23:41.480 --> 0:23:43.719 of who he was or how he got there, and 0:23:43.800 --> 0:23:49.800 like there was this international publicity publicizing like who he 0:23:49.880 --> 0:23:52.800 wasn't that he couldn't remember who he was, and somebody 0:23:52.840 --> 0:23:55.600