WEBVTT - The Human Ethics of Artificial Intelligence 0:00:04.400 --> 0:00:12.840 Welcome to tech Stuff, a production from I Heart Radio. Okay, there, 0:00:13.039 --> 0:00:16.599 I'm Lauren Vogelbaum, sitting in for Jonathan Strickland today for 0:00:16.760 --> 0:00:21.479 I Heart Radio's International Women's Day podcast Takeover. Uh. Some 0:00:21.640 --> 0:00:24.239 of y'all might remember me from the way back I 0:00:24.280 --> 0:00:26.560 was Jonathan's co host here for a minute, when I 0:00:26.640 --> 0:00:29.480 was just a ton of little baby podcaster, or from 0:00:29.520 --> 0:00:32.559 Forward Thinking, another show that we worked on together, or 0:00:32.600 --> 0:00:35.680 from other podcasts that I work on myself, or you 0:00:35.760 --> 0:00:39.400 might find my voice new and strange. Um. But in 0:00:39.479 --> 0:00:43.520 any case, hello, thank you for existing and for being 0:00:43.560 --> 0:00:47.440 here today. In honor of International Women's Day, I wanted 0:00:47.440 --> 0:00:51.640 to do this episode about ethics in technology, and specifically 0:00:51.640 --> 0:00:55.080 an artificial intelligence. You know, about the ways that tech 0:00:55.200 --> 0:00:58.720 can can hurt or help the quest to make the 0:00:58.760 --> 0:01:03.560 world more equitable. And you might be going, aren't computers 0:01:03.680 --> 0:01:08.480 essentially or even quintessentially unbiased? Um? You know, a program 0:01:08.480 --> 0:01:11.560 doesn't have feelings, It only has code that it executes, 0:01:12.160 --> 0:01:16.039 And of course that's true, but the humans who write 0:01:16.040 --> 0:01:20.440 the code do have biases, some conscious, some unconscious, and 0:01:20.520 --> 0:01:23.080 so the ways that we tell programs to work can 0:01:23.120 --> 0:01:27.000 carry those biases. One example that I always think of. 0:01:27.200 --> 0:01:29.520 And you might have seen headlines about this back in 0:01:29.600 --> 0:01:35.039 like two about how digital cameras can behave in biased ways. 0:01:35.480 --> 0:01:38.920 There was this whole thing where some webcams weren't tracking 0:01:38.959 --> 0:01:42.640 and focusing on the faces of black users, and some 0:01:42.760 --> 0:01:47.360 other cameras were flagging photos of Asian subjects because the 0:01:47.440 --> 0:01:51.160 software insisted that their eyes were closed. In both of 0:01:51.200 --> 0:01:54.120 these cases, it was clear that the programs had been 0:01:54.160 --> 0:01:58.120 trained on photos of a vast majority of white faces. 0:01:58.440 --> 0:02:01.200 The programs didn't know what to do with skin that 0:02:01.240 --> 0:02:05.000 reflected light differently or with eyes that were a different shape. 0:02:05.800 --> 0:02:07.840 And I will say, like this is not a purely 0:02:08.240 --> 0:02:10.480 digital issue. UM, this sort of thing has been an 0:02:10.520 --> 0:02:14.320 issue in photography for as long as photography has existed. 0:02:14.520 --> 0:02:18.760 UM film stocks were originally created with only white subjects 0:02:18.760 --> 0:02:23.160 in mind. And it wasn't until like furniture and chocolate 0:02:23.240 --> 0:02:26.960 companies started lodging complaints with Kodak in the nineteen sixties 0:02:27.000 --> 0:02:29.880 and seventies. The Kodak started to adjust their films to 0:02:29.919 --> 0:02:35.200 better capture different shades of brown. But on this like 0:02:35.440 --> 0:02:39.000 small example scale, you know, like that sucks for the 0:02:39.080 --> 0:02:43.520 users of these cameras. But these problems are really compounded 0:02:43.639 --> 0:02:47.000 when you start getting into big data and machine learning 0:02:47.120 --> 0:02:52.520 and artificial intelligence. Artificial intelligence has the capacity to totally 0:02:52.680 --> 0:02:56.280 change our world for the better. Um. Everything from making 0:02:56.440 --> 0:03:00.640 our energy grid more efficient and more adaptable, preventing tragic 0:03:00.680 --> 0:03:04.920 outages like we saw in Texas recently, to helping farmers 0:03:05.040 --> 0:03:07.640 make the most of their resources and getting more fresh 0:03:07.680 --> 0:03:11.600 foods to people who currently don't have good access to that, 0:03:12.280 --> 0:03:16.840 to making autonomous vehicles possible, to letting your doctor just 0:03:16.919 --> 0:03:19.679 like real quick consult every case of a disease, ever 0:03:19.880 --> 0:03:22.919 while making a decision about how to proceed with your treatment. 0:03:23.560 --> 0:03:27.000 Uh to I don't know, stopping the thing where you 0:03:27.160 --> 0:03:33.560 always get served ads for the thing you just bought. Yes, 0:03:33.639 --> 0:03:35.760 I like that T shirt, That's why I just bought 0:03:35.760 --> 0:03:39.400 it up. Okay, Uh anyway, it's it's not It's not 0:03:39.480 --> 0:03:43.880 as simple as um Asimov's laws of robotics when you 0:03:44.000 --> 0:03:47.920 start getting into the wider consequences of AI UM and 0:03:48.240 --> 0:03:51.320 you know, um, a robot may not injure a human 0:03:51.360 --> 0:03:54.920 being or through inaction, allow a human being to come 0:03:54.960 --> 0:03:57.680 to harm, which of course didn't even work in the 0:03:57.680 --> 0:04:01.480 fictional world that that asthmov was set up. And robot 0:04:01.480 --> 0:04:04.920 ethics is totally a thing that is also not simple. 0:04:05.040 --> 0:04:07.080 And of course it is ideal if you're if your 0:04:07.160 --> 0:04:11.960 roomba or your autonomous car does not kill you. UM, 0:04:11.960 --> 0:04:15.880 but we're talking about designing these AI systems that will 0:04:16.000 --> 0:04:19.680 change the way of life for whole societies. UM, it 0:04:19.839 --> 0:04:23.400 is a big deal and it could lead to some 0:04:23.520 --> 0:04:28.599 big problems. So we need to talk about how to 0:04:28.720 --> 0:04:35.000 train your algorithm. Um. These big systems start small with 0:04:35.120 --> 0:04:39.640 designers training algorithms with data sets. So right from the start, UM, 0:04:39.720 --> 0:04:42.040 you have the issue of what data is going in 0:04:42.480 --> 0:04:46.279 and within that data, what's being paid attention to and 0:04:46.320 --> 0:04:50.240 what's being ignored. And now I'm not saying that these 0:04:50.279 --> 0:04:54.080 designers are like all mustache twirling villains out there to 0:04:54.160 --> 0:04:56.800 do evil, but they are human. You know, We're each 0:04:56.839 --> 0:04:59.760 moving through the world with our own set of experiences. 0:05:00.040 --> 0:05:02.800 There are so many other experiences that were bound to 0:05:02.960 --> 0:05:05.840 fail to take some of them into consideration, or to 0:05:06.160 --> 0:05:10.479 or to misunderstand some of those circumstances. Which is why 0:05:10.480 --> 0:05:12.640 it's so important to have people from a variety of 0:05:12.680 --> 0:05:16.320 backgrounds and on these projects. And right now, diversity in 0:05:16.360 --> 0:05:20.800 tech is uh not great. UM. Back inten a whole 0:05:20.839 --> 0:05:23.919 bunch of big tech companies got together and pledged to 0:05:24.240 --> 0:05:28.279 increase diversity in their workforces UM and make their results public. 0:05:28.839 --> 0:05:32.400 And every year these reports come out and the numbers 0:05:32.440 --> 0:05:35.880 haven't changed that much in some of these categories in 0:05:36.600 --> 0:05:41.359 six years. UH. Women are better represented now UM or 0:05:41.400 --> 0:05:46.360 as rather having gone from around fift of the workforce 0:05:46.400 --> 0:05:50.960 to around of the workforce at places like Google and Facebook. 0:05:51.680 --> 0:05:55.279 But the only company that showed a comparable jump in 0:05:55.560 --> 0:06:00.839 UM black employees was Amazon, and they're including their distribution 0:06:00.920 --> 0:06:05.679 center employees. And other categories of underrepresented people like people 0:06:05.680 --> 0:06:09.279 with disabilities aren't even being reported in all of these 0:06:09.320 --> 0:06:13.279 public results UM, but studies show that they're dramatically underrepresented 0:06:13.279 --> 0:06:16.520 in the workforce, which just isn't great. You know, when 0:06:16.560 --> 0:06:20.080 we're designing technology like self driving cars that will need 0:06:20.120 --> 0:06:25.360 to take into consideration the movement of wheelchair users. And 0:06:25.480 --> 0:06:29.840 there are unfortunately instances of programs being made specifically with 0:06:29.960 --> 0:06:34.159 bias UM, like in to sixteen when the Boston Police 0:06:34.160 --> 0:06:39.239 Department used this social media surveillance system to flag posts 0:06:39.279 --> 0:06:44.440 made by regular citizens who who used certain terms, for example, 0:06:44.920 --> 0:06:51.160 colloquial Arabic Muslim words or um words like ferguson or protest, 0:06:51.920 --> 0:06:55.440 and we are all in this, whether we like it 0:06:55.560 --> 0:06:58.960 or not. UM Again, just as one example, everything that 0:06:59.000 --> 0:07:01.599 we do online, from you know, what we type into 0:07:01.640 --> 0:07:05.880 search engines and social media sites, to our location data, 0:07:06.040 --> 0:07:09.680 to how we move our mice or like tap at 0:07:09.680 --> 0:07:12.760 our smartphones, all of that has the potential to be 0:07:12.840 --> 0:07:17.440 recorded and collected and sold and referenced and cross referenced 0:07:17.480 --> 0:07:20.360 and used to track us in any number of ways. 0:07:21.000 --> 0:07:25.679 But AI systems are a huge industry worldwide. Business spending 0:07:25.760 --> 0:07:30.240 on artificial intelligence hit about fifty billion dollars in and 0:07:30.280 --> 0:07:36.960 it's expected to more than double that by retail, banking, media, governments, 0:07:37.120 --> 0:07:41.560 All kinds of industries are investing in this, and all 0:07:41.600 --> 0:07:45.000 of this is fairly new, But of course the field 0:07:45.080 --> 0:07:48.560 of ethics, and even computer ethics is not new at all. 0:07:49.480 --> 0:07:55.760 So ethics and technology have always been tied together because 0:07:56.040 --> 0:07:59.840 every time that that we humans create some new technology 0:07:59.840 --> 0:08:02.640 that changes our world and how we interact with it 0:08:02.760 --> 0:08:06.720 and each other, we have to reconsider our world and 0:08:06.880 --> 0:08:11.239 our interactions. You could even argue that in that way, 0:08:11.400 --> 0:08:16.400 like like philosophically, ethics is itself a type of technology. 0:08:16.640 --> 0:08:18.600 But I'm not going to go that deep today. I'm 0:08:18.600 --> 0:08:23.120 backing away from that precipice. Uh So, let's skip ahead 0:08:23.200 --> 0:08:26.200 from you know, the beginning of human consciousness UM to 0:08:26.520 --> 0:08:30.800 the nineteen forties, because that's when digital computers were being 0:08:30.840 --> 0:08:34.520 invented and in the field of cybernetics got started. That's 0:08:34.559 --> 0:08:39.280 the science of information feedback systems, right. UM. Cybernetics was 0:08:39.320 --> 0:08:43.160 pioneered by m T mathematician Norbert Wiener and some of 0:08:43.200 --> 0:08:45.960 his colleagues as they were working during World War Two 0:08:46.320 --> 0:08:49.840 to develop an anti aircraft cannon that could that could 0:08:49.880 --> 0:08:53.720 a detect and track a fast moving airplane and then 0:08:53.960 --> 0:08:58.640 be extrapolate the airplane's probable location in the immediate future 0:08:58.920 --> 0:09:02.520 and aim um and then see signal the firing mechanism 0:09:02.559 --> 0:09:08.840 to fire. And that internal communication that the machine was 0:09:08.960 --> 0:09:13.680 doing really got Weener thinking. In he published a book 0:09:13.720 --> 0:09:18.000 called Cybernetics, or Control and Communication in the Animal and 0:09:18.120 --> 0:09:21.840 the Machine, and in it he mused that these new 0:09:22.000 --> 0:09:27.000 computing machines could very easily become central nervous systems for 0:09:27.120 --> 0:09:30.080 processing all kinds of data from all kinds of instruments, 0:09:30.600 --> 0:09:34.360 and that that potential was huge. He compared it to 0:09:34.440 --> 0:09:38.920 nuclear weapons um. He wrote, long before Nagasaki and the 0:09:38.960 --> 0:09:41.920 public awareness of the atomic bomb, it had occurred to 0:09:41.960 --> 0:09:44.679 me that we were here in the presence of another 0:09:44.960 --> 0:09:50.040 social potentiality, of unheard of importance for good and for evil. 0:09:51.720 --> 0:09:54.200 Oh Weener expounded on this in a book he published 0:09:54.240 --> 0:09:58.119 in nineteen fifty called The Human Use of Human Beings, 0:09:58.640 --> 0:10:03.600 which which basically predicted that integrating computer technology into society 0:10:03.920 --> 0:10:07.840 was going to be another revolution, just as sweeping and 0:10:08.000 --> 0:10:12.400 messy as the Industrial Revolution, and he tried to lay 0:10:12.440 --> 0:10:14.600 out a bit of groundwork for how to not like 0:10:14.920 --> 0:10:20.120 totally bork it up. Spoiler alert, people borked it up anyway. 0:10:20.679 --> 0:10:22.840 There wasn't a whole lot more work done in the 0:10:22.880 --> 0:10:26.320 field of computer ethics until the nineteen sixties and seventies, 0:10:26.720 --> 0:10:31.359 when computer based crime and information security and privacy concerns 0:10:31.559 --> 0:10:35.640 had already become a problem. So by the mid sixties, 0:10:35.720 --> 0:10:40.480 corporations and the government had both begun collecting just giant 0:10:40.559 --> 0:10:46.120 amounts of personal data about US citizens in literally massive computers. 0:10:47.120 --> 0:10:50.520 I guess, I guess all computers are literally massive in 0:10:50.520 --> 0:10:53.720 that they have mass. But okay, you know what I mean. Um, 0:10:54.080 --> 0:10:57.280 From you know, medical records to military records to legal 0:10:57.320 --> 0:11:01.240 documents to shopping habits. And this journalist by the name 0:11:01.280 --> 0:11:04.959 of Vance Packard wrote a book called The Naked Society 0:11:05.000 --> 0:11:09.600 published in nineteen sixty four about the inherent privacy issue 0:11:09.679 --> 0:11:16.040 of having that information collected and available for instantaneous reference. UH. 0:11:16.200 --> 0:11:20.959 Something of an uproar ensued UM. Focusing as Packard's book did, 0:11:21.400 --> 0:11:25.880 on the US government's use of citizens information UM, and 0:11:25.920 --> 0:11:29.000 it led to just a whole bunch of data transparency 0:11:29.080 --> 0:11:32.520 legislation over the next decade. UH. The Freedom of Information Act, 0:11:32.600 --> 0:11:36.080 the Fair Credit Reporting Act, all designed to make sure 0:11:36.120 --> 0:11:39.440 that citizens are able to know what data is being 0:11:39.480 --> 0:11:43.560 collected about them by the government and how it's being used, 0:11:43.960 --> 0:11:47.319 And to be fair, at that time, most computational power 0:11:47.480 --> 0:11:50.600 was in the hands of the government. But this legislature 0:11:50.800 --> 0:11:56.120 and conversation really ignored the activities of private corporations, and 0:11:56.160 --> 0:11:59.600 it never really questioned the ethics of collecting all of 0:11:59.600 --> 0:12:02.679 that at to in the first place. UM. And this 0:12:02.760 --> 0:12:05.400 is apparently like a very American thing, the concept that 0:12:05.760 --> 0:12:09.360 information is inherently good and like more is better UM. 0:12:09.400 --> 0:12:12.200 But that is a rabbit hole for another day. But 0:12:12.679 --> 0:12:16.800 it's not that people weren't thinking about the ethics. One 0:12:16.880 --> 0:12:19.240 important thing that happened around the same time was that 0:12:19.280 --> 0:12:22.080 a researcher by the name of Don Parker, who was 0:12:22.160 --> 0:12:26.840 looking into crime being committed via computers. Parker proposed to 0:12:26.920 --> 0:12:32.360 this leading industry professional organization, the Association for Computing Machinery, 0:12:32.400 --> 0:12:35.000 that they develop a code of ethics for their members, 0:12:35.520 --> 0:12:38.079 and they were like, yeah, cool, you you do that. UM. 0:12:38.120 --> 0:12:40.120 So he headed up a committee and the a c 0:12:40.280 --> 0:12:43.800 M adopted their first code of Ethics in ninety three. 0:12:44.559 --> 0:12:46.880 They've updated it, I think like about once a decade 0:12:46.920 --> 0:12:49.640 since then, UM, with the most recent update being in 0:12:51.160 --> 0:12:54.120 It's really thoughtful. UM. One of my favorite bits from 0:12:54.120 --> 0:12:57.439 the intro its specifies that it's quote not an algorithm 0:12:57.520 --> 0:13:01.000 for solving ethical problems. Rather, it's as a basis for 0:13:01.120 --> 0:13:04.160 ethical decision making. UM. You can read it if you're 0:13:04.160 --> 0:13:05.599 into that sort of thing by going to a c 0:13:05.800 --> 0:13:09.439 M dot org and then back to our timeline. In 0:13:09.520 --> 0:13:14.400 nineteen seventy six, Joseph Weisenbaum, who had created the psychotherapy 0:13:14.440 --> 0:13:18.880 mimicking chat, bought Eliza. A decade earlier, he published his 0:13:18.920 --> 0:13:23.240 book Computer Power and Human Reason From Judgment to Calculation, 0:13:24.480 --> 0:13:28.280 and this book was a response to the response that 0:13:28.400 --> 0:13:32.479 he had gotten from people to his chat Bought Eliza, 0:13:33.040 --> 0:13:35.640 and I know that Jonathan has talked about this chat 0:13:35.679 --> 0:13:37.960 Bought on the show before. UM. It comes up a 0:13:38.000 --> 0:13:41.599 lot and discussions about the Turing test and how convincingly 0:13:41.720 --> 0:13:45.480 computers can approximate human communication because it was one of 0:13:45.480 --> 0:13:48.760 the first that was effective. UM. The thing is, though, 0:13:48.880 --> 0:13:54.280 that Weisenbaum designed Eliza as a demonstration of how bad 0:13:54.679 --> 0:13:59.680 computers inherently are at this type of communication, but he 0:13:59.760 --> 0:14:03.920 got the opposite response from people who tried the program 0:14:03.920 --> 0:14:07.280 out in talking to I mean, you know, typing with 0:14:07.559 --> 0:14:12.720 Eliza about their psychological problems. People felt like Eliza understood them, 0:14:12.760 --> 0:14:17.000 even when they knew it was a robot. UM. So 0:14:17.880 --> 0:14:21.960 Weisenbaum was kind of like, whoe wait uh. And so 0:14:22.000 --> 0:14:25.000 he wrote this book to to really explain the differences 0:14:25.120 --> 0:14:30.560 between computation and human intelligence and to assert that ethics 0:14:30.600 --> 0:14:34.880 are imperative in the design of artificial intelligence because people 0:14:34.920 --> 0:14:39.440 will forget the computers do not have the understanding, the wisdom, 0:14:39.560 --> 0:14:44.880 the moral and emotional consideration of human beings. And this 0:14:44.960 --> 0:14:48.080 is when things really started picking up in the field 0:14:48.080 --> 0:14:51.840 of computer ethics. UM. Also in ninety this professor who 0:14:51.880 --> 0:14:55.560 was teaching medical ethics at the time. Walter Byner noticed 0:14:55.680 --> 0:15:00.880 how computers were complicating that field, and he became the 0:15:00.880 --> 0:15:04.920 first person in academia to really like decide computer ethics 0:15:04.960 --> 0:15:08.880 should be its own field of research and application UM. 0:15:08.880 --> 0:15:13.280 He wrote that computer technology was creating whole new ethical 0:15:13.320 --> 0:15:17.040 concerns that needed to be taken into consideration. UM, so 0:15:17.080 --> 0:15:20.560 he popularized the term computer ethics, did speeches and workshops 0:15:20.560 --> 0:15:24.920 and everything and jumping ahead. The first major textbook on 0:15:24.960 --> 0:15:28.480 the subject was published called Computer Ethics, edited by one 0:15:28.560 --> 0:15:32.960 Deborah Johnson, and Johnson disagreed with maner Um. She argued 0:15:33.000 --> 0:15:38.400 the computers weren't creating new problems, but rather exacerbating old 0:15:38.480 --> 0:15:44.680 problems around things like privacy, ownership, power, and responsibility. We 0:15:44.760 --> 0:15:47.560 will get into what the future may hold after we 0:15:47.640 --> 0:15:58.080 get back from a quick break. Welcome back. As the 0:15:58.160 --> 0:16:03.160 computer industry group with a consumer adoption of home computing, 0:16:03.240 --> 0:16:06.640 with with internet access growing in both the capacity and 0:16:06.880 --> 0:16:11.640 the ubiquity of computers, just absolutely snowballing UM. The field 0:16:11.640 --> 0:16:15.320 of computer ethics exploded UM, and more and more groups 0:16:15.400 --> 0:16:19.440 formed to to help everyone makes sense of all of this. 0:16:20.040 --> 0:16:24.880 For example, the Electronic Frontier Foundation was founded in there, 0:16:25.160 --> 0:16:29.080 that advocacy and activism group that's dedicated to defending civil 0:16:29.160 --> 0:16:33.560 liberties as technology advances. And okay, this is kind of 0:16:33.560 --> 0:16:36.000 a side quest, but I did not notice. They actually 0:16:36.000 --> 0:16:39.240 formed up in response to the federal seizure of a 0:16:39.240 --> 0:16:44.080 bunch of computer equipment belonging to Steve Jackson Games, yep, 0:16:44.120 --> 0:16:46.960 the company that brings us Apples to Apples and Munchkin 0:16:47.520 --> 0:16:50.160 lots of other good stuff. So what happened here? What 0:16:50.200 --> 0:16:52.720 was that was that there was this Bell South digital 0:16:52.760 --> 0:16:57.880 document that explained how the emergency telephone system worked, and 0:16:57.960 --> 0:17:01.520 it leaked, and the U. S. Secret Service was concerned 0:17:01.560 --> 0:17:04.280 that having that info out there was a security concern, 0:17:04.359 --> 0:17:07.359 you know, that hackers might overwhelm the system, um, something 0:17:07.400 --> 0:17:11.880 like that. So so the Secret Service was conducting raids 0:17:12.200 --> 0:17:17.879 tracking this documents digital distribution. Steve Jackson was innocent, um, 0:17:17.920 --> 0:17:20.520 but they got this warrant, conducted this raid, took all 0:17:20.560 --> 0:17:24.719 this stuff, um, and wound up accessing and the leading 0:17:24.920 --> 0:17:28.240 a bunch of personal bulletin board messages from the company's 0:17:28.280 --> 0:17:32.439 website in the process. And now the Secret Service, you know, 0:17:32.480 --> 0:17:35.280 didn't find anything, so they gave the equipment back and 0:17:35.320 --> 0:17:39.600 didn't press charges, but Steve Jackson was like, no, no, no, 0:17:40.359 --> 0:17:44.080 y'all almost tanked my business. You violated the privacy of 0:17:44.119 --> 0:17:47.159 my bulletin board users. We are pressing charges against you. 0:17:48.160 --> 0:17:52.439 But there was really no civil rights organization that was 0:17:52.480 --> 0:17:55.960 prepared to take on the case due to the technological 0:17:55.960 --> 0:17:59.159 complexity of the issue. So the e f F formed 0:17:59.200 --> 0:18:01.960 in order to bring that suit to court. Um, and 0:18:02.000 --> 0:18:04.480 that was the first time that the court recognized the 0:18:04.560 --> 0:18:07.720 email should have equal protection to to to to phone 0:18:07.720 --> 0:18:10.800 calls or any other kind of communication. So so thanks, 0:18:10.960 --> 0:18:15.199 thanks Steve Jackson for for everything. Um, the ff has 0:18:15.240 --> 0:18:18.239 done a whole lot of important stuff. Um, they got 0:18:18.359 --> 0:18:22.639 encryption technology taken off the list of nationally regulated weapons. 0:18:23.520 --> 0:18:26.840 That's a separate episode though, anyway, So side quest over 0:18:27.000 --> 0:18:30.200 back to the main quest. Um, all of this computer 0:18:30.240 --> 0:18:34.760 stuff exploded and the field of computer ethics specialized. Um, 0:18:34.880 --> 0:18:39.760 so now you've got internet ethics, information systems ethics, robot ethics. 0:18:40.720 --> 0:18:42.080 And I and I do want to say, all of 0:18:42.119 --> 0:18:45.160 this does coincide with work being done in in other 0:18:45.359 --> 0:18:48.640 fields engine engineering and information science. Like I don't want 0:18:48.640 --> 0:18:52.320 to imply that computer technology was the only field that 0:18:52.359 --> 0:18:55.640 had been working with and contributing to these to these 0:18:55.880 --> 0:18:59.840 ethical theories and the practical application of them. But like 0:18:59.840 --> 0:19:01.720 I was saying at the top of this episode, one 0:19:01.720 --> 0:19:05.480 of the really interesting specialties to me is in AI 0:19:05.560 --> 0:19:09.320 ethics because it does have such potentially sweeping effects, and 0:19:09.440 --> 0:19:12.280 also because it keeps coming up in the news for 0:19:12.760 --> 0:19:16.879 less than rad reasons. So you have probably seen headlines 0:19:16.920 --> 0:19:22.119 over the past few years about algorithms gone wrong. Um, 0:19:22.359 --> 0:19:26.639 there's there's the story from where Pro Publica found that 0:19:26.680 --> 0:19:30.560 the software used by some courts to determine the risk 0:19:30.680 --> 0:19:35.359 of criminal defendants committing further offenses and therefore to determine 0:19:35.480 --> 0:19:39.840 whether to detain those defendants until trial or or what 0:19:39.960 --> 0:19:42.840 kind of bail to set for them. Um. At least 0:19:42.840 --> 0:19:47.359 one of those algorithms, called Compass, regularly found black people 0:19:47.520 --> 0:19:52.480 riskier and white people less risky, even when everything else 0:19:52.560 --> 0:19:56.640 about the defendants cases were comparable. And this kind of 0:19:56.680 --> 0:20:00.760 issue crops up in discussions about hiring software as well. 0:20:00.840 --> 0:20:03.880 Because of a lack of care in their design, these 0:20:03.920 --> 0:20:08.880 programs that automatically sort through resumes have ranked applicants who 0:20:08.920 --> 0:20:13.240 have woman sounding or or black sounding names as lower 0:20:13.800 --> 0:20:18.920 in consideration. And then there's Google's search algorithms UM from 0:20:20.359 --> 0:20:23.800 eighteen there were all of these headlines. Reverse image searches 0:20:24.080 --> 0:20:28.919 using photos of black people were returning images of guerrillas 0:20:29.680 --> 0:20:32.000 because no one had taught the system how to consider 0:20:32.440 --> 0:20:36.399 dark skin tones in people. Or if you searched the 0:20:36.680 --> 0:20:42.080 seemingly innocuous term black girls, the first page results out 0:20:42.080 --> 0:20:46.600 of trillions of web indexed pages included porn. Or there 0:20:46.640 --> 0:20:49.720 was research into the search history of the white guy 0:20:49.880 --> 0:20:53.400 who killed nine black people in a church in Charleston, 0:20:53.480 --> 0:20:57.920 South Carolina. It's highly likely that he was radicalized, at 0:20:57.960 --> 0:21:01.119 least in part due to the way that Google search 0:21:01.160 --> 0:21:05.280 works UM. It's taking into account location and demographics about 0:21:05.359 --> 0:21:09.240 him and other searchers in his area. Google search returning 0:21:09.440 --> 0:21:13.720 white supremacist propaganda when he searched the term black on 0:21:13.800 --> 0:21:17.160 white crime. By the way, related to this, you can 0:21:17.240 --> 0:21:21.320 make Google give you less specialized search results. And I 0:21:21.359 --> 0:21:23.359 am looking into that and doing it as soon as 0:21:23.400 --> 0:21:27.120 I finished recording this this because it is continually infuriating 0:21:27.119 --> 0:21:30.000 to me, just just when I'm doing my reading for 0:21:30.000 --> 0:21:32.760 for podcast episodes like this and trying to find stuff 0:21:32.800 --> 0:21:36.120 that isn't like a restaurant in my area. If I'm 0:21:36.119 --> 0:21:41.639 talking about a larger concern anyway. UM. Also, remember how 0:21:41.640 --> 0:21:44.600 I was talking about the flaws in programming of digital 0:21:44.640 --> 0:21:49.400 cameras and how they sometimes have trouble discerning specific features 0:21:49.400 --> 0:21:52.560 of people of color. UM. You know, extrapolate that out 0:21:52.680 --> 0:21:57.040 to how facial recognition software is used with surveillance footage 0:21:57.160 --> 0:22:01.199 by police departments. Um, the software is more likely to 0:22:01.240 --> 0:22:05.440 make a mistake in identifying a black person's face because 0:22:05.440 --> 0:22:08.480 the software just isn't as good as seeing that face 0:22:08.720 --> 0:22:11.480 because of how it was programmed, which can lead to 0:22:11.680 --> 0:22:16.960 false identifications and thus wrongful harassments and arrests. UM. Georgetown 0:22:17.040 --> 0:22:20.639 University put out a whole report on this in called 0:22:20.720 --> 0:22:24.560 the Perpetual Lineup. That report found that half of Americans 0:22:24.760 --> 0:22:28.240 have photos and police facial recognition databases, by the way, UM, 0:22:28.280 --> 0:22:32.080 which includes just lots of people with no criminal backgrounds. 0:22:32.400 --> 0:22:35.000 And of course, even if we fix those algorithms, that 0:22:35.000 --> 0:22:37.560 that isn't going to fix the fact that communities of 0:22:37.600 --> 0:22:41.399 color are subject to more surveillance in the first place. 0:22:42.920 --> 0:22:46.600 More recently, there have been headlines and a whole discussion 0:22:46.680 --> 0:22:50.840 about this new rule that was issued in September by 0:22:51.000 --> 0:22:54.840 the U S Department of Housing and Urban Development. And 0:22:54.880 --> 0:22:59.280 this rule essentially makes it super difficult for banks or 0:22:59.400 --> 0:23:03.960 landlord or homeowners insurance companies to be sued for denying 0:23:04.000 --> 0:23:07.840 housing two people of color if an algorithm was used 0:23:08.000 --> 0:23:12.160 to make that determination. Based on the concept that algorithms 0:23:12.160 --> 0:23:16.720 cannot be racist, the rule was immediately challenged, UM, and 0:23:17.000 --> 0:23:21.679 in January, President Biden issued an executive order directing the 0:23:21.720 --> 0:23:25.000 Department of Housing and Urban Development to to examine the 0:23:25.080 --> 0:23:31.240 rules effects. But hoof, hoof, UM, And you know it's 0:23:31.280 --> 0:23:35.560 it's not it's not easy. None of this is easy, um. 0:23:35.600 --> 0:23:40.120 In order to build a non biased artificial intelligence system, 0:23:40.280 --> 0:23:43.560 we um, I mean like like humans, not like you know, 0:23:43.640 --> 0:23:46.840 you and me, dear listener. We need to change the 0:23:46.880 --> 0:23:51.280 systems that lead to the building of artificial intelligence. We 0:23:51.320 --> 0:23:56.399 need to examine how the design and programming is taught, um, 0:23:56.440 --> 0:24:01.640 how companies conduct their business, how policy is written, and 0:24:01.640 --> 0:24:05.119 and who has access to seats at all of those tables. 0:24:05.520 --> 0:24:08.280 I mean it's also not technically easy, Like when you 0:24:08.320 --> 0:24:11.520 build and train these systems, just adding more diverse data 0:24:11.760 --> 0:24:15.800 isn't magically going to make the system create better, less 0:24:15.840 --> 0:24:19.800 biased rules. It might create conflicting rules. UM. You know 0:24:20.000 --> 0:24:23.320 it is expensive in terms of time and effort and 0:24:23.320 --> 0:24:26.720 and just pure physical energy to to do this work. 0:24:27.359 --> 0:24:31.159 The pitfalls of not doing this work are tremendous. You know, 0:24:31.200 --> 0:24:35.720 it can cause measurable hurt in people's lives. And as 0:24:35.760 --> 0:24:41.080 one doctor Debi Chatra, material science engineer, has said, any 0:24:41.160 --> 0:24:47.320 sufficiently advanced neglect is indistinguishable from malice. But the benefits 0:24:47.480 --> 0:24:53.600 to doing this work are equally measurable and tremendous. One 0:24:53.760 --> 0:24:57.840 consideration moving into the future is how to square the 0:24:57.960 --> 0:25:03.920 very concept of ethics with an increasingly multicultural digital world. Um. 0:25:03.960 --> 0:25:06.160 You know, not not everyone on the planet grew up 0:25:06.160 --> 0:25:09.440 with European philosophy, going back to the ancient Greek as 0:25:09.440 --> 0:25:13.439 the basis of their ethical conception. And we also have 0:25:13.520 --> 0:25:17.440 to acknowledge that, um, you know, whatever a culture's philosophical basis, 0:25:18.040 --> 0:25:21.240 it's probably rooted in some biases of its own. Um. 0:25:21.280 --> 0:25:24.320 Just for example, just throwing it out there, if your 0:25:24.320 --> 0:25:29.720 society has coded emotions as feminine and feminine as bad, um, 0:25:29.840 --> 0:25:34.040 then you're probably not giving emotional harm as much weight, 0:25:34.119 --> 0:25:38.680 if any weight, as physical harm in your considerations of justice. Um. 0:25:38.720 --> 0:25:41.800 And you can see the effects of this in things 0:25:41.880 --> 0:25:44.720 like the care that we give our veterans with physical 0:25:44.760 --> 0:25:49.600 injury versus veterans with PTSD UM or just the general 0:25:49.600 --> 0:25:53.240 ways that our society handles mental health versus physical health, 0:25:53.520 --> 0:25:57.920 or or any kind of neurodivergence. All of this work 0:25:58.080 --> 0:26:02.719 in artificial intelligence and the ethics thereof is really requiring 0:26:02.800 --> 0:26:08.080 us to redefine intelligence, to fully consider what we mean 0:26:08.359 --> 0:26:12.359 by human intelligence, what logic and emotion and experience go 0:26:12.440 --> 0:26:16.440 into that, and the ways in which machine intelligence might differ. 0:26:16.880 --> 0:26:22.480 The Stanford Encyclopedia Philosophy references Minsky's book The Society of 0:26:22.480 --> 0:26:26.760 the Mind, saying we do not wish to restrict intelligence 0:26:27.000 --> 0:26:32.280 to what would require intelligence if done by humans. And 0:26:32.280 --> 0:26:35.800 and of course that's true, um A, I can do 0:26:35.920 --> 0:26:40.080 stuff that we can't that that's arguably the whole point, um. 0:26:40.119 --> 0:26:42.439 But it all has to be done with the best 0:26:42.640 --> 0:26:47.560 of what human intelligence can be in mind, and I'm 0:26:47.560 --> 0:26:49.439 just now realizing that, like I might have written a 0:26:49.520 --> 0:26:52.560 forward thinking episode instead of a text stuff episode. But 0:26:52.560 --> 0:26:54.920 but that's what I've got for you today. So if 0:26:55.000 --> 0:26:57.600 you have enjoyed this episode and would like to hear 0:26:57.640 --> 0:27:01.280