WEBVTT - TechStuff Tidbits: Biometrics 0:00:04.440 --> 0:00:12.320 Welcome to tech Stuff, a production from iHeartRadio. Hey there, 0:00:12.360 --> 0:00:16.120 and welcome to tech Stuff. I'm your host, Jonathan Strickland. 0:00:16.160 --> 0:00:19.639 I'm an executive producer with iHeartRadio. And how the tech 0:00:19.720 --> 0:00:22.639 are you? You know, I'm sure you've all seen movies 0:00:22.800 --> 0:00:25.920 which someone has to get past a security system by 0:00:26.040 --> 0:00:29.960 having an eye scan, and maybe they have an unconscious 0:00:30.080 --> 0:00:35.440 or potentially unalive security guard laying nearby and they hold 0:00:35.440 --> 0:00:38.320 the person's head up to the scanner to gain access 0:00:38.360 --> 0:00:42.199 to the security door. Or maybe the main character has 0:00:42.280 --> 0:00:47.280 undergone some sort of incredibly disturbing surgical procedure to have 0:00:47.360 --> 0:00:50.559 their own eyes altered to give them access. But you 0:00:50.600 --> 0:00:53.320 get the idea. The eye scan verifies that the person 0:00:53.400 --> 0:00:57.320 trying to open the door has the authorization to do that. 0:00:57.880 --> 0:01:01.600 This is a form a biometric and I'm sure we're 0:01:01.640 --> 0:01:06.600 all familiar with biometrics. You probably use them regularly. If 0:01:06.600 --> 0:01:09.679 you have a phone that has a fingerprint scanner or 0:01:09.720 --> 0:01:14.440 a face scanner to unlock the phone, that's biometrics. I 0:01:14.520 --> 0:01:17.039 used to use a laptop that had a fingerprint scanner, 0:01:17.440 --> 0:01:20.120 which was kind of cool. It made it really easy 0:01:20.200 --> 0:01:25.600 for me to log in. Not necessarily the most secure technology. However, 0:01:25.760 --> 0:01:30.080 as I remember reading about how that particular fingerprint scanner 0:01:30.160 --> 0:01:34.240 had some issues, but it's still sure made logging into 0:01:34.240 --> 0:01:36.800 my computer easy back in those days. My current one 0:01:36.840 --> 0:01:40.200 does not have that. And there are actually a couple 0:01:40.240 --> 0:01:44.039 of different definitions for biometrics, but the one we're concerned 0:01:44.080 --> 0:01:49.000 about involves body and behavioral measurements for the purposes of 0:01:49.200 --> 0:01:55.240 authentication and verification or identification, because these are all related 0:01:55.320 --> 0:01:58.680 but different things. So for this to work, the stuff 0:01:58.840 --> 0:02:02.280 what we're measuring has to be unique to an individual. 0:02:02.680 --> 0:02:05.800 It needs to be something that cannot be easily replicated. 0:02:06.440 --> 0:02:08.679 That's the whole point, right. If it could be replicated, 0:02:09.600 --> 0:02:13.360 then it's no good for verifying someone or identifying someone 0:02:13.560 --> 0:02:16.960 because it could be one of many people. So you 0:02:17.000 --> 0:02:20.040 need to have something that is unique to that person. 0:02:20.760 --> 0:02:23.920 Spoiler alert. While biometrics can be used for the purposes 0:02:23.919 --> 0:02:28.760 of security, for verification or identification, it's also pretty darn 0:02:28.800 --> 0:02:32.320 handy if you just want to keep track of someone, right, 0:02:33.000 --> 0:02:35.720 so you get enough biometric data on a person, it 0:02:35.720 --> 0:02:38.600 becomes easier to track that person down to keep track 0:02:38.600 --> 0:02:42.000 of what they're doing, assuming that they're not like hold 0:02:42.080 --> 0:02:44.560 up in some off grid shack in the middle of 0:02:44.600 --> 0:02:47.840 nowhere and they never venture out of there, in which case, yeah, 0:02:47.960 --> 0:02:50.400 they can be under the radar then, but otherwise it 0:02:50.440 --> 0:02:54.000 gets pretty tricky. But before we get into all that 0:02:54.280 --> 0:02:57.280 and the concerns around biometrics, let's talk a little bit 0:02:57.440 --> 0:03:02.840 about the history of biometrics. This history is varied, It 0:03:02.880 --> 0:03:05.840 goes all over the place. It has lots of starts 0:03:05.840 --> 0:03:09.880 and stops and stutters, and that's because biometrics focuses on 0:03:09.960 --> 0:03:12.880 lots of different things, right Like, it's not all just 0:03:13.520 --> 0:03:16.720 one area of study. Typically we're talking about things that 0:03:16.760 --> 0:03:22.840 people were interested in studying or advancing that collectively we 0:03:22.919 --> 0:03:26.040 would categorize under biometrics, but they were looking at a 0:03:26.120 --> 0:03:30.079 very specific version of it. So you could argue that 0:03:30.280 --> 0:03:33.680 biometrics traces its history to more than twenty five hundred 0:03:33.720 --> 0:03:38.000 years ago. Because Babylonians back in the day, which I'm 0:03:38.040 --> 0:03:42.240 told was a Tuesday, would sign off on business transactions 0:03:42.600 --> 0:03:46.440 by placing a thumb print on a clay tablet. Right 0:03:46.600 --> 0:03:50.480 that the clay tablet would be pliable, you would PLoP 0:03:50.480 --> 0:03:53.560 your thumb down and make an impression. The tablet would 0:03:53.640 --> 0:03:56.000 dry and your print would be there to authenticate that 0:03:56.040 --> 0:04:00.080 you had, in fact completed whatever the transaction was. Bo 0:04:00.640 --> 0:04:04.440 that's biometrics, and Tom Cruise wasn't even involved a little bit. 0:04:05.080 --> 0:04:07.280 I always think of Tom Cruise in movies like the 0:04:07.320 --> 0:04:10.800 Mission Impossible series, or even in Minority Report. When I 0:04:10.840 --> 0:04:13.080 think about biometrics, those are the ones that pop up 0:04:13.080 --> 0:04:16.280 in my head. Of course, it's not like there was 0:04:16.320 --> 0:04:20.720 a direct line of evolution from the ancient Babylonians to 0:04:20.720 --> 0:04:23.440 Tom Cruise, at least not that I am aware of. 0:04:23.960 --> 0:04:27.560 Biometrics weren't always a common means of marking documents or 0:04:27.560 --> 0:04:31.600 anything like that. But by the late nineteenth century, some 0:04:31.640 --> 0:04:34.760 folks started to figure out that the Babylonians could have 0:04:34.800 --> 0:04:38.720 really been onto something. Actually, I'm kidding. They probably didn't 0:04:38.720 --> 0:04:41.160 even know about the Babylonian stuff. They just thought they 0:04:41.200 --> 0:04:44.359 came up with it themselves. Because typically that's how we 0:04:44.440 --> 0:04:48.640 see the smarty pants of the eighteenth and nineteenth centuries. 0:04:49.200 --> 0:04:52.080 Some of them were a little more humble than that, 0:04:52.600 --> 0:04:56.560 but not all of them. Anyway, Folks in the late 0:04:56.640 --> 0:04:59.440 nineteenth century, so we're talking the eighteen hundreds here, they 0:04:59.480 --> 0:05:02.720 observe that fingerprints appeared to be unique to the individual, 0:05:03.040 --> 0:05:05.640 and if you got two different people to make prints 0:05:05.680 --> 0:05:08.680 of their fingertips. You would see the differences in the 0:05:08.720 --> 0:05:11.200 patterns of those fingerprints, and you would be able to 0:05:11.240 --> 0:05:14.960 tell which person made which set of prints, assuming that 0:05:15.000 --> 0:05:17.360 you had a reference right, as long as you're able 0:05:17.400 --> 0:05:20.919 to refer it back to something on record. This logically 0:05:21.480 --> 0:05:24.200 led to the concept of using fingerprints as a way 0:05:24.240 --> 0:05:27.200 to identify if a suspect had been present at the 0:05:27.240 --> 0:05:29.760 scene of a crime. Of course, folks had to figure 0:05:29.800 --> 0:05:34.159 out how to detect and lift prints at the scene. Typically, 0:05:34.240 --> 0:05:38.000 they would use a very fine powder or dust which 0:05:38.000 --> 0:05:41.840 would adhere to the oils left behind from the fingertips 0:05:41.880 --> 0:05:44.960 on different surfaces, and then they would lift those prints 0:05:45.000 --> 0:05:49.080 and then compare that to a record of fingerprints so 0:05:49.080 --> 0:05:51.480 that they could reference the prints that were lifted at 0:05:51.480 --> 0:05:55.160 the scene with ones that were in their database the 0:05:55.279 --> 0:05:58.799 reference system, or if they had the suspect in custody, 0:05:59.040 --> 0:06:02.599 they would have the suspect, you know, get printed, and 0:06:02.600 --> 0:06:04.880 they would compare the prints against whatever was found at 0:06:04.880 --> 0:06:07.440 the scene and then try to determine if the suspect 0:06:07.480 --> 0:06:10.520 was actually present at the scene of the crime. You 0:06:10.600 --> 0:06:14.840 know how this works, But the science of that and 0:06:14.960 --> 0:06:18.040 the systems based around that took a while to develop. 0:06:18.720 --> 0:06:21.440 Folks had to identify the elements and fingerprints that are 0:06:21.560 --> 0:06:27.280 definable and identifiable. They had to make classifications and taxonomies. 0:06:27.440 --> 0:06:29.560 You know, you had to have a way to communicate 0:06:30.240 --> 0:06:33.400 the qualities of a print when talking with someone else, 0:06:33.720 --> 0:06:35.640 or to be able to look for a match or 0:06:35.720 --> 0:06:38.880 even just a partial match. And it took a couple 0:06:38.880 --> 0:06:42.880 of decades to really catch on. And while fingerprints are 0:06:43.080 --> 0:06:47.320 the most famous example of early biometrics and law enforcement, 0:06:47.720 --> 0:06:50.880 police were trying different stuff too. In France, around the 0:06:50.880 --> 0:06:54.159 same time as fingerprints coming into vogue, there was a 0:06:54.200 --> 0:06:59.880 system called the Bertilline process or Bertillion process. The police 0:07:00.120 --> 0:07:04.120 would measure suspects and they would write down their measurements 0:07:04.160 --> 0:07:07.120 on a card and keep that in an index. And 0:07:07.200 --> 0:07:09.880 I'm talking all sorts of measurements. You know. It's kind 0:07:09.880 --> 0:07:12.120 of like if you were to go and get fitted 0:07:12.200 --> 0:07:16.720 for custom clothing, except you're being fitted for crimes, I guess. 0:07:16.880 --> 0:07:19.680 So the police would take down stuff like how tall 0:07:19.720 --> 0:07:22.680 you were, you know, how wide were your shoulders, how 0:07:22.760 --> 0:07:26.440 long are your arms, all that kind of stuff. This 0:07:26.560 --> 0:07:30.000 approach had some big drawbacks, however, the biggest one being 0:07:30.000 --> 0:07:33.440 that there was a lack of standardization in technique and metrics, 0:07:33.960 --> 0:07:37.520 which meant it wasn't terribly useful. Because if one police 0:07:37.520 --> 0:07:42.240 officer measures arm length one way, let's say, like they 0:07:42.280 --> 0:07:45.960 measure the arm just hanging down from the shoulder, and 0:07:46.040 --> 0:07:48.000 someone else measures it in a different way, like you're 0:07:48.000 --> 0:07:51.160 holding your arm up at an angle, then you might 0:07:51.240 --> 0:07:55.280 end up getting very different you know, metrics measurements. And 0:07:55.320 --> 0:07:57.720 that meant that you would look at the index card 0:07:57.720 --> 0:07:59.680 and say, oh, well, this person's not a match. That's 0:07:59.720 --> 0:08:02.720 not who who this person is or whatever, And so 0:08:03.240 --> 0:08:07.040 it was not terribly useful, and you could get false 0:08:07.080 --> 0:08:11.000 positives or false negatives. So this particular method wouldn't stand 0:08:11.000 --> 0:08:15.680 the test of time. Fingerprints definitely did. In the nineteen thirties, 0:08:15.880 --> 0:08:18.520 a guy named Frank Birch had another idea for using 0:08:18.560 --> 0:08:23.400 biometrics as a way to identify people. He was an ophthalmologist, that, 0:08:23.440 --> 0:08:25.360 by the way, is a great word that trips me 0:08:25.440 --> 0:08:29.520 up in spelling bees, and you might imagine. His idea 0:08:29.720 --> 0:08:34.280 was that an individual's iris contains complex patterns, and those 0:08:34.320 --> 0:08:37.479 patterns are unique to that person. It's like a fingerprint, 0:08:37.480 --> 0:08:40.280 but it's the pattern that's in the iris of your eye, 0:08:40.800 --> 0:08:43.840 and so like a fingerprint, a person's iris could serve 0:08:43.920 --> 0:08:47.280 as a way to identify that person and subsequently verify 0:08:47.320 --> 0:08:50.120 their identity in the future. You just needed a way 0:08:50.160 --> 0:08:53.640 to capture those patterns and then a technology capable of 0:08:53.720 --> 0:08:58.160 scanning and matching patterns against those that are stored in 0:08:58.160 --> 0:09:02.120 a database. So in twenty six when Frank Birch came 0:09:02.200 --> 0:09:05.080 up with this idea, none of that was really practical 0:09:05.160 --> 0:09:09.480 or possible, but his basic concept using the eye as 0:09:09.480 --> 0:09:13.959 a way to verify identity, that was solid. Now, it 0:09:13.960 --> 0:09:16.520 would take more than fifty years before someone would come 0:09:16.559 --> 0:09:19.880 up with a computer algorithm that could identify iris patterns 0:09:20.360 --> 0:09:23.920 and moreover, match an IRIS pattern that is collected to 0:09:24.000 --> 0:09:27.120 one in a database. That came to us courtesy of 0:09:27.200 --> 0:09:32.160 Cambridge University professor John Dogman in nineteen eighty seven, and 0:09:32.240 --> 0:09:35.400 it took a few more years to actually capitalize on 0:09:35.440 --> 0:09:38.640 that development. It wasn't until the mid nineteen nineties that 0:09:38.720 --> 0:09:42.760 a company called Iridium Technologies became the first to commercialize 0:09:42.800 --> 0:09:46.720 iris scanning. Now there are lots of companies that do this, 0:09:47.160 --> 0:09:49.920 many with their own proprietary approach to doing it, but 0:09:50.160 --> 0:09:53.360 it's all rooted in Frank Birch's suggestions back in the 0:09:53.440 --> 0:09:57.080 nineteen thirties. Okay, we've got a lot more to talk 0:09:57.120 --> 0:10:00.240 about with biometrics before we get into that. Let's take 0:10:00.280 --> 0:10:13.439 a quick break. You know, not long ago I talked 0:10:13.440 --> 0:10:15.720 a little bit about the early days of researchers trying 0:10:15.760 --> 0:10:21.720 to use computers to develop facial recognition technology, another example 0:10:21.760 --> 0:10:26.160 of biometrics. So the pioneers in that space included Woody Bledsoe, 0:10:26.440 --> 0:10:31.000 a mathematician and computer scientist, Helen chan Wolf, a scientist 0:10:31.080 --> 0:10:35.360 specializing in early artificial intelligence work, and Charles Bisson, another 0:10:35.640 --> 0:10:40.320 computer scientist and the three word for panoramic research in California. 0:10:40.440 --> 0:10:42.200 And they set about trying to figure out how to 0:10:42.320 --> 0:10:46.200 break down facial features into data points that a computer 0:10:46.320 --> 0:10:50.040 could analyze and match. So, you know, you imagined that 0:10:50.080 --> 0:10:52.760 at first, this just looks and involves looking at pictures 0:10:52.760 --> 0:10:56.280 of people and saying, well, where are some ways that 0:10:56.320 --> 0:10:59.320 we can we can point out, like landmarks that a 0:10:59.400 --> 0:11:03.480 computer would be able to recognize and measure against, right, 0:11:04.080 --> 0:11:07.080 like the corners of eyes compared to the bridge of 0:11:07.080 --> 0:11:09.160 the nose, that kind of stuff. And there are all 0:11:09.200 --> 0:11:12.480 these different images of pictures of people with all these 0:11:12.559 --> 0:11:16.439 different lines geometric lines drawn on the faces in an 0:11:16.440 --> 0:11:21.480 effort to kind of establish these standards. And as you 0:11:21.520 --> 0:11:24.560 might imagine, in the early days, this technology hit some 0:11:24.600 --> 0:11:28.960 pretty tough limitations. A computer couldn't necessarily match two different 0:11:28.960 --> 0:11:32.880 pictures of the same person's face. If the lighting and 0:11:32.920 --> 0:11:36.040 the shadows or the angle of the picture were different enough, 0:11:36.360 --> 0:11:38.920 the computer system would have trouble determining that both images 0:11:38.920 --> 0:11:41.360 were of the same face. Yes, if you had two 0:11:41.400 --> 0:11:45.000 pictures of the same person where it was under identical 0:11:45.080 --> 0:11:48.240 lighting conditions and the camera was at the same distance 0:11:48.280 --> 0:11:51.680 in the same angle, then it was easier, But with 0:11:51.840 --> 0:11:55.880 any deviation from that it became much more difficult. It 0:11:55.880 --> 0:11:59.920 would take many years for the algorithms, the computer technology, 0:12:00.160 --> 0:12:03.280 camera technologies to improve to a point that made facial 0:12:03.320 --> 0:12:07.840 recognition a possibility. So we're going to skip way up 0:12:07.880 --> 0:12:10.719 to the present day, and we've talked extensively on this 0:12:10.840 --> 0:12:14.360 show about how facial recognition technology often has major issues, 0:12:14.400 --> 0:12:18.000 particularly when it comes to false positives and false negatives 0:12:18.000 --> 0:12:21.319 with certain populations. Now, I'm not going to dive into 0:12:21.320 --> 0:12:23.280 that whole can of worms yet again, because I'm sure 0:12:23.280 --> 0:12:24.960 most of you have heard me talk about it a lot, 0:12:25.440 --> 0:12:30.080 but it is an inescapable fact that most facial recognition 0:12:30.160 --> 0:12:33.680 systems are prone to making errors due to biases that 0:12:33.720 --> 0:12:35.840 are in the system. And just to be clear, I 0:12:35.880 --> 0:12:39.800 don't mean to imply that there was ever any intentional bias, 0:12:40.400 --> 0:12:45.920 but intentional or not, that bias still has an effect. Meanwhile, 0:12:45.960 --> 0:12:49.760 other computer scientists were working in the mid twentieth century 0:12:50.120 --> 0:12:54.520 on the challenge of developing speech recognition technologies. Now, like 0:12:54.600 --> 0:12:58.800 facial recognition, this tech had a steep barrier of entry. 0:12:59.320 --> 0:13:04.599 Voices come in all sorts of timbers, volumes, pitches, dialects, 0:13:04.679 --> 0:13:08.240 or accents. So two people might say the very same 0:13:08.280 --> 0:13:12.600 word in very different ways. You know, it may sound 0:13:12.760 --> 0:13:14.640 very different, and yet it's the same word, And that 0:13:14.760 --> 0:13:18.520 means the system needs to be capable of recognizing that 0:13:18.559 --> 0:13:21.480 word no matter who says it or how they say it. 0:13:21.720 --> 0:13:25.079 Otherwise you're not going to have a satisfying experience. Well, 0:13:25.080 --> 0:13:27.760 folks in Bell Labs were working on speech recognition tech 0:13:27.800 --> 0:13:30.880 as early as the nineteen fifties, but that work focused 0:13:30.880 --> 0:13:34.119 on training a machine to recognize when someone was speaking 0:13:34.320 --> 0:13:38.640 out numbers. It took a decade of hard work to 0:13:38.679 --> 0:13:41.800 get to a point where the computer systems could recognize 0:13:42.000 --> 0:13:45.600 words of a certain complexity. The nineteen seventies would see 0:13:45.600 --> 0:13:49.360 the technology advanced significantly, helped in large part by a 0:13:49.559 --> 0:13:53.600 DARPA initiative. Y'all remember DARPA, right, that's the division within 0:13:53.679 --> 0:13:56.679 the US Department of Defense that awards contracts to companies 0:13:56.960 --> 0:13:59.800 that engage in R and D, and it's all in 0:13:59.840 --> 0:14:03.359 the purpose of trying to fund projects that could potentially 0:14:03.400 --> 0:14:07.559 benefit US defense initiatives in the future. DARPA has funded 0:14:07.640 --> 0:14:12.560 research into everything from autonomous vehicles to robots capable of 0:14:12.559 --> 0:14:16.040 performing half a dozen different tasks, and in the seventies, 0:14:16.320 --> 0:14:21.440 the division funded work in speech recognition. But speech recognition 0:14:22.000 --> 0:14:28.680 is really related to using voice technology for the purposes 0:14:28.720 --> 0:14:32.960 of a computer understanding what someone is saying. There's also 0:14:33.880 --> 0:14:40.200 voice recognition, or some people prefer speaker recognition, because that's 0:14:40.240 --> 0:14:46.480 all about recognizing a specific person. So speech recognition is 0:14:46.520 --> 0:14:49.800 really more about computer systems that can parse what you're 0:14:49.840 --> 0:14:53.880 saying and glean instructions from that and respond properly so 0:14:53.920 --> 0:14:56.960 that when you ask your smart speaker about the weather, 0:14:57.120 --> 0:14:59.160 it can give you the information you want instead of 0:14:59.200 --> 0:15:01.640 I don't knowtaneously turning off all the lights in your 0:15:01.640 --> 0:15:07.040 house or whatever, not that I've had that experience. Voice 0:15:07.080 --> 0:15:11.240 recognition or speaker recognition involves breaking a specific voice into 0:15:11.320 --> 0:15:14.360 data points for the purposes of identification, just like with 0:15:14.480 --> 0:15:18.000 facial recognition. Right. So, the idea of developing a system 0:15:18.040 --> 0:15:22.440 to do serious analysis on voices dates back to the 0:15:22.560 --> 0:15:26.400 early twentieth century. However, a lot of that early work 0:15:26.520 --> 0:15:31.840 lacked scientific value or rigor. So while you had people saying, yes, 0:15:31.880 --> 0:15:35.920 I can get a print out of what someone's voices 0:15:36.000 --> 0:15:38.040 and thus be able to compare two print outs and 0:15:38.040 --> 0:15:39.960 tell you if it's the same person speaking or not, 0:15:40.440 --> 0:15:43.520 in truth, it was a bit more complicated than that. 0:15:43.720 --> 0:15:46.760 I mean, you know, it required more than just having 0:15:46.880 --> 0:15:54.600 a long strip of paper tape pulled across a pencil that, 0:15:54.720 --> 0:15:58.080 in turn was connected to a diaphragm that would cause 0:15:58.160 --> 0:16:01.720 vibrations to move through when sound would go through like 0:16:02.880 --> 0:16:07.800 a horn or a microphone. Yeah, you could create a 0:16:07.920 --> 0:16:11.280 visual representation of someone's voice that way, but it wasn't 0:16:11.480 --> 0:16:14.320 something that was specific enough for you to be able 0:16:14.360 --> 0:16:19.320 to differentiate that voice from another voice or even sometimes 0:16:19.400 --> 0:16:23.440 other noises entirely, like you might get a record of 0:16:23.960 --> 0:16:26.520 noise that is not a voice at all, but because 0:16:26.560 --> 0:16:30.160 it looks kind of like what someone else produced when 0:16:30.160 --> 0:16:31.720 they were speaking into it, you might think, oh, it 0:16:31.760 --> 0:16:35.600 was this person. It would take years to get to 0:16:35.680 --> 0:16:39.920 a more sophisticated approach to speech analysis and voice analysis 0:16:40.440 --> 0:16:45.480 to approach the possibility of identifying someone based on their voice. 0:16:45.920 --> 0:16:50.280 In the nineteen forties, folks learned about a sound spectrograph technology. 0:16:50.440 --> 0:16:52.680 So this is a device that would create a visual 0:16:52.720 --> 0:16:56.880 representation of a signal. Sound spectrograph being sound. We're talking 0:16:56.880 --> 0:16:59.280 about audio in this case, but you can have spectrographs 0:16:59.280 --> 0:17:02.440 that create a visual representation of lots of different types 0:17:02.480 --> 0:17:06.320 of signals. This is just specifically about audio. Some people 0:17:06.359 --> 0:17:09.400 would refer to these as voice prints. They would compare 0:17:09.440 --> 0:17:13.159 this to a fingerprint, saying, oh, this representation shows the 0:17:13.240 --> 0:17:15.920 quality of this person's voice. If we find a match, 0:17:16.040 --> 0:17:18.080 then that's the same person. It's just like a fingerprint. 0:17:18.440 --> 0:17:21.600 The people who were actually working in the field really 0:17:21.640 --> 0:17:24.640 didn't use the term voice print very much, if at all, 0:17:25.040 --> 0:17:26.600 but it was a term that was used a lot 0:17:26.640 --> 0:17:30.280 in the media, and I think a lot of folks 0:17:30.320 --> 0:17:33.159 who were working in the field were frustrated because they 0:17:33.160 --> 0:17:36.040 felt it was a bit reductive and an oversimplification of 0:17:36.080 --> 0:17:39.640 what they were doing. So a slightly more acceptable term 0:17:39.760 --> 0:17:43.679 is what I've referenced earlier, speaker recognition. This implies the 0:17:43.680 --> 0:17:47.560 technology isn't necessarily trying to understand what someone is saying. 0:17:48.320 --> 0:17:52.200 Instead is trying to identify the person who is saying it, 0:17:52.760 --> 0:17:55.359 or verifying that the person who is saying it is 0:17:55.400 --> 0:17:58.400 who they claimed to be. You might use this technology 0:17:59.119 --> 0:18:04.159 to have someone get access to something secret, right, like 0:18:04.520 --> 0:18:07.000 the voice print analysis. You've seen this in movies too, 0:18:07.000 --> 0:18:09.880 where someone walks up to a door and they say 0:18:09.920 --> 0:18:14.920 a phrase and then apparently the computer inside the door 0:18:15.359 --> 0:18:18.600 and analyzes the voice and then either allows entry or 0:18:18.640 --> 0:18:21.760 denies it. Or you might use it to try and 0:18:21.800 --> 0:18:25.560 identify somebody, right, maybe you've got a recording of someone. 0:18:25.600 --> 0:18:29.439 Maybe someone calls into a television station and makes a threat. 0:18:29.880 --> 0:18:33.199 This has happened, and then what you're trying to do 0:18:33.320 --> 0:18:37.480 is identify a suspect you have found and to determine 0:18:37.480 --> 0:18:39.000 whether or not they were the person who made the 0:18:39.040 --> 0:18:41.240 phone call, and you're trying to match the voices together. 0:18:41.800 --> 0:18:44.440 This is something that's been going on in law enforcement 0:18:44.520 --> 0:18:48.320 for decades and for a very long time. Courts would 0:18:48.800 --> 0:18:55.760 reject the evidence presented because there was a lack of 0:18:55.800 --> 0:19:00.920 scientific studies showing the accuracy and reliability of this kind 0:19:00.920 --> 0:19:04.119 of approach, and you needed to show that there was 0:19:04.160 --> 0:19:09.439 a scientific basis for this and not just a claim 0:19:09.600 --> 0:19:13.800 that these two voices must be identical. The technology involves 0:19:14.000 --> 0:19:18.960 sophisticated pattern analysis and it is really tricky. So those 0:19:19.040 --> 0:19:21.520 early court cases, it's probably a good idea to throw 0:19:21.560 --> 0:19:25.640 those cases out. Now, maybe some guilty people went free. 0:19:26.160 --> 0:19:30.359 It's hard to say, but the fact is we just 0:19:30.400 --> 0:19:33.359 weren't at a level of sophistication a pattern analysis to 0:19:33.359 --> 0:19:37.560 be able to have really reliable identification. These days, there 0:19:37.560 --> 0:19:41.920 are examples of speaker recognition technology built into consumer products. 0:19:42.359 --> 0:19:47.840 My smart speaker at home supposedly recognizes my voice, for example, 0:19:48.080 --> 0:19:50.160 and this should make it possible for me to ask 0:19:50.200 --> 0:19:54.640 about my daily schedule, and because the speaker recognizes that 0:19:54.680 --> 0:19:58.359 the user is yours, truly, it could then cross reference 0:19:58.440 --> 0:20:01.520 my calendar and then tell me what my schedule is, 0:20:01.960 --> 0:20:03.679 or at least it's supposed to be able to do that. 0:20:04.200 --> 0:20:06.720 I don't know. I mean, it's probably because I never 0:20:06.760 --> 0:20:09.000 put anything in my calendar. I'm really bad about doing that. 0:20:09.160 --> 0:20:11.600 So really, my speaker just gets fed up with me 0:20:11.840 --> 0:20:15.760 because I'm asking it to do something it really cannot do. 0:20:16.560 --> 0:20:18.359 All right, we're gonna take another quick break. When we 0:20:18.400 --> 0:20:20.200 come back, I'm going to talk a little bit more 0:20:20.200 --> 0:20:25.760 about some other biometric approaches and also about a current 0:20:25.840 --> 0:20:30.720 story that really inspired this entire episode and the different 0:20:30.880 --> 0:20:34.600 sides to that story. But first let's take another quick break. 0:20:43.720 --> 0:20:45.720 Before the break, I alluded to the fact that there 0:20:45.760 --> 0:20:50.639 are other approaches to biometric verification or identification. So another 0:20:50.680 --> 0:20:55.679 one is gate recognition GAI. This is the way someone 0:20:55.800 --> 0:20:59.760 moves through a space, like things like the length of 0:20:59.800 --> 0:21:04.480 their stride, the position of their body as they are walking, 0:21:05.040 --> 0:21:08.000 the location of various body parts in relation to one another, 0:21:08.119 --> 0:21:12.119 like how far are your hips from your knees or 0:21:12.160 --> 0:21:14.919 your knees from your ankles, that sort of thing, what 0:21:15.080 --> 0:21:18.119 sort of how much do you bend when you're moving. 0:21:19.280 --> 0:21:22.600 So it is possible to analyze how a person naturally 0:21:22.800 --> 0:21:27.560 walks or moves through space and then to use that 0:21:27.600 --> 0:21:32.880 information to identify someone. So if you have reference data 0:21:33.119 --> 0:21:36.800 where you know how a person typically walks, then you 0:21:36.880 --> 0:21:39.639 might be able to use that in when you're searching 0:21:39.640 --> 0:21:42.520 for somebody. A person could be trying to evade surveillance, 0:21:42.560 --> 0:21:46.359 for example, through the use of disguises, but gate analysis 0:21:46.440 --> 0:21:49.359 might reveal who they really are, assuming you've got reference 0:21:49.480 --> 0:21:52.720 data as well. I've seen movies that do this where 0:21:52.720 --> 0:21:55.960 it's a person who's just watching like a video feed. 0:21:56.000 --> 0:21:58.720 They're saying, ah, I see they have a limp, and 0:21:59.040 --> 0:22:01.679 you know, we know so and so favors their right legs, 0:22:01.680 --> 0:22:04.240 so I think we've got the person here. Like we've 0:22:04.240 --> 0:22:06.639 seen that sort of thing. It's the same basic idea, 0:22:06.680 --> 0:22:08.800 except you don't have to have a limp. It's really 0:22:08.920 --> 0:22:13.040 just all the basic movements you make that end up 0:22:13.280 --> 0:22:16.600 kind of being unique to you. You can try and 0:22:16.640 --> 0:22:20.000 disguise that. Obviously, you can purposefully give yourself like a 0:22:20.119 --> 0:22:23.800 limp or something in an effort to throw off any 0:22:24.600 --> 0:22:27.640 surveillance techniques that would be looking to match your GIT 0:22:27.880 --> 0:22:33.120 with you. And that's a possibility, but this is an 0:22:33.160 --> 0:22:38.560 actual thing that's being used, this gate analysis and gait verification. Really, 0:22:38.600 --> 0:22:41.239 when it comes to biometrics, again, as I said at 0:22:41.240 --> 0:22:43.959 the beginning, the important quality is that whatever thing you 0:22:44.000 --> 0:22:48.719 are measuring needs to be unique to individuals. So it 0:22:48.720 --> 0:22:51.119 could be physical, it could be behavioral. It could be 0:22:51.160 --> 0:22:53.040 a combination of the two, but it has to be 0:22:53.160 --> 0:22:56.600 unique or else it doesn't do you any good whether 0:22:56.720 --> 0:23:00.159 you're using this technology to verify someone's identity before or 0:23:00.200 --> 0:23:02.840 letting them through a secure checkpoint, or if you're just 0:23:02.880 --> 0:23:06.240 trying to use it to determine an unknown person's identity 0:23:06.240 --> 0:23:09.960 by comparing it against the database of known individuals. Now, 0:23:11.080 --> 0:23:14.600 there are a lot of reasonable concerns about biometrics. It's 0:23:14.600 --> 0:23:18.440 totally understandable, right, like, this is information that is unique 0:23:18.480 --> 0:23:22.240 to you, that is a very private info. And we're 0:23:22.680 --> 0:23:27.600 seeing more and more implementations of biometric systems around us. 0:23:27.600 --> 0:23:30.920 Like I mentioned before, your very phone may rely on it. 0:23:31.400 --> 0:23:34.919 Mine has a fingerprint sensor for example. A lot of 0:23:34.920 --> 0:23:38.960 iPhones use facial recognition to activate, to turn on without 0:23:39.000 --> 0:23:41.239 you having to put in some sort of code. So 0:23:41.560 --> 0:23:46.840 we're seeing biometrics rolled out very in very wide deployments 0:23:46.960 --> 0:23:50.600 in all sorts of different applications. We also have seen 0:23:50.640 --> 0:23:53.480 it for things that are very official, not just use 0:23:53.520 --> 0:23:56.600 of consumer products, but things like you know, maybe getting 0:23:56.640 --> 0:23:59.600 an iris scan as part of identification like in a 0:23:59.640 --> 0:24:03.520 past or or something, or you might have a biometric 0:24:03.600 --> 0:24:07.840 scan that allows you to bypass the normal process of 0:24:07.880 --> 0:24:11.040 getting on a plane. I've had that happen a couple 0:24:11.040 --> 0:24:12.640 of times, and it freaked me out the first time 0:24:12.680 --> 0:24:17.080 because I don't remember ever submitting to the initial scan, 0:24:18.560 --> 0:24:21.240 but it knew who I was when I walked up, 0:24:21.280 --> 0:24:23.680 and I thought, well, that's weird because I didn't actually 0:24:24.400 --> 0:24:26.840 I didn't do I didn't sit down or agree to 0:24:26.880 --> 0:24:31.439 do a process I didn't knowingly agree. I probably did agree, 0:24:31.640 --> 0:24:34.560 I just didn't read the fine print, because nobody ever does, right, 0:24:35.119 --> 0:24:37.919 But that I walked up to the gate and it 0:24:37.960 --> 0:24:40.080 recognized who I was and I didn't even have to 0:24:40.080 --> 0:24:43.400 present a boarding pass or anything. I just got waved in. 0:24:43.960 --> 0:24:48.520 And that's a little creepy, and I can see why 0:24:48.560 --> 0:24:53.439 people would be hesitant about it. And you think about 0:24:53.520 --> 0:24:57.000 the possibilities of this technology and the possibilities of abusing 0:24:57.000 --> 0:24:59.800