WEBVTT 1 00:00:02.600 --> 00:00:05.879 Hello, and welcome to This anthro Life, a podcast about the little things 2 00:00:05.879 --> 00:00:08.720 we do as people that shape the course of humanity. I'm your host, 3 00:00:08.839 --> 00:00:14.039 Adam Gamwell. Today we're going to talk about responsible AI, artificial intelligence, 4 00:00:14.199 --> 00:00:17.760 and the future of the technology industry. But before we get into it, 5 00:00:17.920 --> 00:00:21.559 let me share a story with you. Recently, a law enforcement agency in 6 00:00:21.559 --> 00:00:25.719 the US deployed an AI system to predict which of its officers might need counseling 7 00:00:25.879 --> 00:00:30.640 based on their social media posts. The system was meant to detect indicators of 8 00:00:30.679 --> 00:00:35.000 mental health issues and help get struggling officers the help that they need. At 9 00:00:35.039 --> 00:00:37.600 first, clans, it seemed like a great use of AI, but when 10 00:00:37.600 --> 00:00:41.000 the system was analyzed, it turned out that the vast majority of the at 11 00:00:41.079 --> 00:00:46.280 risk officers were people of color or women, were both expressing their concern about 12 00:00:46.280 --> 00:00:51.759 systemic racism and sexism within their departments. In other words, the AI system 13 00:00:51.799 --> 00:00:56.240 was flagging officers who were speaking out against injustices, rather than those who might 14 00:00:56.280 --> 00:01:00.000 need help in counseling for mental health issues. The story highlights the urgent need 15 00:01:00.039 --> 00:01:04.280 for responsible AI development and implementation, and that's precisely what we'll be talking about 16 00:01:04.280 --> 00:01:11.120 today with our guests David Graywitterer and Don Nephis. David Graywitterer is a technology 17 00:01:11.120 --> 00:01:15.040 and ethics researcher and PhD candidate at Carnegie Mellon University. He has conducted extensive 18 00:01:15.079 --> 00:01:19.359 research on the intersection of ethics, technology, and labor as well as works 19 00:01:19.359 --> 00:01:23.400 as a computer scientist. Don Nephis is a senior research scientist it Intel Labs, 20 00:01:23.840 --> 00:01:29.799 and her work focuses on designing and studying systems that support personal informatics and 21 00:01:30.040 --> 00:01:34.680 data driven decision making, as well as exploring the ethical implications of these systems. 22 00:01:34.319 --> 00:01:38.719 Today, Don and David are going to help us explore three questions. 23 00:01:38.120 --> 00:01:44.400 First, what are the limitations of the current modularity system in software development and 24 00:01:44.560 --> 00:01:49.359 what impact can it have on user experience, data privacy, and end user 25 00:01:49.400 --> 00:01:53.200 consumption. If you're not familiar with what a modular system is for development, 26 00:01:53.280 --> 00:01:55.359 no worries. We're going to get into that in the earlier part of the 27 00:01:55.400 --> 00:02:00.640 episode. Second, what are the implications of a siloed approach to user research 28 00:02:00.760 --> 00:02:06.400 and design on AI ethics and responsible development? And how might we ensure that 29 00:02:06.439 --> 00:02:10.639 the needs of a wider range of stakeholders are taken into account During development including 30 00:02:10.759 --> 00:02:16.120 end users. And finally, what are the potential solutions to the power imbalances 31 00:02:16.199 --> 00:02:21.360 between labor that's the folks that are doing the development and management, the folks 32 00:02:21.400 --> 00:02:25.120 that are directing how that labor happens, and how might we build frameworks of 33 00:02:25.120 --> 00:02:31.599 accountability and collective action across the technology industry landscape. So this conversation is fundamentally 34 00:02:31.639 --> 00:02:37.280 important as AI is becoming more and more ubiquitous in our everyday lives and touching 35 00:02:37.400 --> 00:02:40.879 more parts of how we move about our everyday worlds. We hope that this 36 00:02:40.960 --> 00:02:46.759 conversation will help you better understand the complex and multifaceted challenges of responsible AI development 37 00:02:46.840 --> 00:02:50.400 and implementation. We're going to be walking through some of the arguments that David 38 00:02:50.439 --> 00:02:53.759 and Don put together in a new paper. We cannot wait to dive into 39 00:02:53.759 --> 00:02:57.639 this with you, and so we'll jump right into it after a message from 40 00:02:57.639 --> 00:03:10.319 this episode sponsor. So I just want to say Don and David super excited 41 00:03:10.319 --> 00:03:14.560 to have you on the podcast today. It was fun to connect with David 42 00:03:14.599 --> 00:03:17.080 at the Scity for Applied Anthropology this year in February, and we actually just 43 00:03:17.159 --> 00:03:20.680 ran into each other and then we got to kind of talking about the work 44 00:03:20.719 --> 00:03:23.800 that David was doing and immediately caught my attention as this is a conversation we 45 00:03:23.840 --> 00:03:27.400 need to have on the podcast and helped bring to wider audiences. And I 46 00:03:27.439 --> 00:03:30.840 was delighted also find that he's working with the amazing Don Nephis and great to 47 00:03:30.840 --> 00:03:34.479 have you back on the program as well, and both the very timely topic. 48 00:03:34.520 --> 00:03:36.919 But then also I think the work that you're both doing in terms of 49 00:03:36.919 --> 00:03:42.439 bringing computer science into conversation with anthropology and anthropological thinking is super important. So 50 00:03:42.800 --> 00:03:44.960 first off, I want to say to you both, thank you for being 51 00:03:45.000 --> 00:03:49.319 on the show and excited to chat with you today. Excited to be here, 52 00:03:49.680 --> 00:03:51.960 great to be here. Right on, David, tell us to your 53 00:03:52.000 --> 00:03:54.560 superhero origin story, you know, how you found your way into this space, 54 00:03:54.599 --> 00:04:00.199 and then then we'll jump over to Don Sure. So I am Peter 55 00:04:00.240 --> 00:04:05.240 scientists by training. I come to AI from more of a technical perspective. 56 00:04:05.280 --> 00:04:10.360 My first class in IT was thinking about how to build neural networks and such, 57 00:04:10.560 --> 00:04:15.319 and in my research I tend to interview software engineers and try and understand 58 00:04:15.360 --> 00:04:17.959 how they work, how they use tools, you know, So I come 59 00:04:17.959 --> 00:04:23.560 at this more from a technical perspective. Having taken AI courses, but also 60 00:04:23.600 --> 00:04:26.720 trying to understand how software engineers work, how they use tools, how they 61 00:04:26.759 --> 00:04:29.600 make choices, how they make decisions, how they collaborate. And that led 62 00:04:29.639 --> 00:04:33.319 to a few different internships NASA at Microsoft and then also at Intel, where 63 00:04:33.360 --> 00:04:39.480 I met Don and the more I learned about the existing research and responsible AI, 64 00:04:39.519 --> 00:04:43.199 the more I at least saw that it relied on a particular idea of 65 00:04:43.240 --> 00:04:47.000 what developers thought was their problem or their responsibility and what kind of power they 66 00:04:47.079 --> 00:04:53.800 might have to action on what they thought their job was their sort of agency, 67 00:04:54.319 --> 00:04:58.800 and that then led to me conceiving of some of my dissertation research as 68 00:04:58.839 --> 00:05:03.560 thinking about will actually what kind of responsibility do individual developers feel over the downstream 69 00:05:03.600 --> 00:05:08.160 impact of what they create? And given that, what do they feel they 70 00:05:08.160 --> 00:05:11.920 can actually do about it? And the more I've worked across disciplinary lines such 71 00:05:11.920 --> 00:05:15.879 as it's done, the more I've tried to think about those more individual factors 72 00:05:15.920 --> 00:05:23.120 of responsibility and agency within sort of broader discourses and contexts that constrain and to 73 00:05:23.199 --> 00:05:28.000 guide those. So I'll pick up next. So I've been an anthropologist in 74 00:05:28.000 --> 00:05:35.959 industry for an embarrassingly long time, and I turned to responsible AI back when 75 00:05:36.000 --> 00:05:42.639 it was really just exploding around twenty seventeen, twenty eighteen, where gender shades 76 00:05:42.639 --> 00:05:45.639 had just come out. Folks were saying that this is a big deal, 77 00:05:45.759 --> 00:05:51.279 this is a real issue, and internally, my lab director Lama Knoxman had 78 00:05:51.360 --> 00:05:56.680 started leading the charge about, Okay, what do we need to do in 79 00:05:56.839 --> 00:06:00.800 house to sort ourselves out? So we were looking in only what do we 80 00:06:00.839 --> 00:06:03.360 need to do? You know, Intel does put out AI models, we 81 00:06:03.399 --> 00:06:09.120 also have a lot of customers that use AI for various purposes, and so 82 00:06:09.199 --> 00:06:13.399 the question of both how do we have to govern what kinds of models we 83 00:06:13.439 --> 00:06:17.000 put out in the world, which is kind of where the developer experienced perception 84 00:06:17.120 --> 00:06:24.040 understanding slash notion to supply chain became salient. And where does Intel sit in 85 00:06:24.079 --> 00:06:28.439 all of this? Because you know, we're not Google, we're not Meta, 86 00:06:28.639 --> 00:06:31.879 we don't really build the sort of stuff that you commonly see in the 87 00:06:31.920 --> 00:06:36.720 news, So we're do our responsibilities begin and end? Became this question that 88 00:06:36.879 --> 00:06:42.680 became part of my job to answer. So here I am here to answer 89 00:06:42.720 --> 00:06:46.199 that question. Boy, folks, do we have the answer for you today? 90 00:06:46.360 --> 00:06:48.720 I appreciate you both showing that and it's a really interesting arena again to 91 00:06:48.759 --> 00:06:53.720 think about the importance of interdisciplinary approaches to these kinds of problems, right, 92 00:06:53.720 --> 00:06:57.920 because one, as we're recording this in twenty twenty three, AI is becoming 93 00:06:58.000 --> 00:07:00.279 more ubiquitous. I mean, the thing is, it's already been kind of 94 00:07:00.279 --> 00:07:02.959 stealthily ubiquitous and a lot of technologies we might have used the word automation a 95 00:07:03.000 --> 00:07:06.439 little bit more, but now with things like you know, chat GBT and 96 00:07:06.480 --> 00:07:11.680 open AI and propic getting more getting more pressed, and there's just the stupid 97 00:07:11.680 --> 00:07:15.160 explosion of AI chatbots on the iOS and Android app stores now that if they're 98 00:07:15.160 --> 00:07:18.879 becoming just more common tools on the consumer side. And so I think what's 99 00:07:19.160 --> 00:07:23.279 really interesting and I'm excited to explore with with both of you today, is 100 00:07:23.879 --> 00:07:26.759 a lot of the work that you're looking at it is on the production the 101 00:07:26.759 --> 00:07:30.160 developer side, right, And so oftentimes it's like how do things get shaped 102 00:07:30.199 --> 00:07:33.000 before they even find their way over to consumer facing products for features? So 103 00:07:33.120 --> 00:07:38.839 I think it might be useful for our listeners take zoom out a little bit 104 00:07:38.879 --> 00:07:43.639 and describe the world that David and I have been working in, you know, 105 00:07:43.680 --> 00:07:46.399 at a very broad level. One of the things you'll see regularly and 106 00:07:46.680 --> 00:07:56.120 responsible AI reports and publications and even policies is developers should blah blah blah, 107 00:07:56.199 --> 00:08:01.399 right, they should test for bias or developers should be transparent about la la 108 00:08:01.480 --> 00:08:03.519 la la la, And it goes on and on. There's about a million 109 00:08:03.560 --> 00:08:07.000 inventories you can find online that's sort of like have you done X Y and 110 00:08:07.120 --> 00:08:13.600 say okay. And the problem that we're poking at is we're sort of saying, 111 00:08:13.600 --> 00:08:16.680 okay, wait, which ones? Like which ones do you mean exactly 112 00:08:16.680 --> 00:08:22.040 when you say developers should blah blah blah. And that's a problem that arises 113 00:08:22.480 --> 00:08:26.639 in many different contexts. Right. So, for example, even in policymaking 114 00:08:26.720 --> 00:08:31.279 right now, one of the things that many policymakers are struggling with or trying 115 00:08:31.279 --> 00:08:33.440 to figure out how do you have a policy around, is, for example, 116 00:08:33.440 --> 00:08:39.639 what happens when a model AI model gets open sourced? Right? This 117 00:08:39.759 --> 00:08:46.360 discussions happened before the generative AI kind of became hot in the news. Right, 118 00:08:46.440 --> 00:08:50.720 Even something as simple as does it detect a face? Like is it 119 00:08:50.759 --> 00:08:52.440 a face or not? Never mind facial recognition, just is it a face 120 00:08:52.559 --> 00:08:56.039 or not? Right? What's the responsibility when you put this out into open 121 00:08:56.080 --> 00:09:00.440 source? And all of a sudden. You know, anybody in their brother 122 00:09:00.559 --> 00:09:03.919 can do this, that and the other thing with it. And so that's 123 00:09:03.919 --> 00:09:09.159 why we started to reframe. Okay, you know there's the notion of developers, 124 00:09:09.200 --> 00:09:15.120 but who developers are actually situated in a political economy of sorts. If 125 00:09:15.159 --> 00:09:18.559 you're an anthropologist, that's how you describe it. If you're a business person, 126 00:09:18.679 --> 00:09:22.440 you describe it as an ecosystem, or you might also describe into supply 127 00:09:22.519 --> 00:09:24.960 chain terms, which is what we did, both because that invokes kind of 128 00:09:26.039 --> 00:09:33.080 corporate social responsibility, but also because that's how developers themselves see the world like 129 00:09:33.120 --> 00:09:37.519 they see it in terms of a chain where they made a little part and 130 00:09:37.639 --> 00:09:43.639 the little part is modularized and exactly the same way as you'd have a container 131 00:09:43.799 --> 00:09:46.120 on a ship. Right, the language isn't coincidental at all. And another 132 00:09:46.159 --> 00:09:50.720 scholars have been writing about how that way of imagining the world kind of inflects 133 00:09:50.759 --> 00:09:54.600 what people do. Right, So we've said, okay, fine, you've 134 00:09:54.679 --> 00:10:00.759 organized yourselves. There are many different actors, both companies and individuals, assembled 135 00:10:00.759 --> 00:10:03.679 along what they see as a chain, and what we're saying is, okay, 136 00:10:05.600 --> 00:10:09.320 how do you act that's a social circumstance so how do you act in 137 00:10:09.000 --> 00:10:13.279 that to get the results that's going to actually work out for people? So 138 00:10:13.639 --> 00:10:16.039 that's why we frame the problem in the way that we did. I think 139 00:10:16.120 --> 00:10:22.039 that we've seen, at least in my view, responsible AI emerge as a 140 00:10:22.519 --> 00:10:24.600 largely a set of principles we might have heard of, you know, fair 141 00:10:24.639 --> 00:10:30.799 AI, accountable AI, transparent AI, and that sort of in a lot 142 00:10:30.840 --> 00:10:33.799 of company context, but in a lot of like policy and regulatory context too, 143 00:10:35.200 --> 00:10:39.720 seems to outline the boundaries of what constitutes responsible AI. So we want 144 00:10:39.720 --> 00:10:41.720 to build these things in certain ways. We want to build what we're building 145 00:10:41.799 --> 00:10:45.919 in a in a fair way, make sure the AI is fair, doesn't 146 00:10:45.919 --> 00:10:50.399 discriminate. But what that does, in my view, and some of my 147 00:10:50.519 --> 00:10:54.600 research has shown, is it sort of scopes what people see as their their 148 00:10:54.600 --> 00:10:58.320 responsibility. You know, if we're going to build a fair AI, we 149 00:10:58.360 --> 00:11:01.519 will build it in such a way that it is you know, doesn't have 150 00:11:01.679 --> 00:11:05.720 bias in the model perhaps, But once that leaves our hands, that's where 151 00:11:05.879 --> 00:11:09.960 perhaps some of our responsibility might stop. You know, the design attributes, 152 00:11:09.000 --> 00:11:15.080 the fair, accountable, transparent design attributes, or what scope responsibility In many 153 00:11:15.080 --> 00:11:20.759 conceptions of responsible aim and less thinking downstream less thinking about how the thing we 154 00:11:20.799 --> 00:11:24.960 build may be used, and perhaps also thinking less upstream too, like you 155 00:11:26.000 --> 00:11:28.159 know, oh, well, our data that we're getting from the web or 156 00:11:28.200 --> 00:11:31.799 whatever has biased in it, but we didn't create that data, so that's 157 00:11:31.840 --> 00:11:35.759 not on us in some sense, and that's where we came from, you 158 00:11:35.799 --> 00:11:39.559 know, thinking about where I at least came from thinking about these principles and 159 00:11:39.600 --> 00:11:43.840 what their perhaps shortcomings or limitations are, and then began thinking about how they 160 00:11:43.919 --> 00:11:50.360 might limit the kind of scope that individual developers told to and believing certainly that 161 00:11:50.519 --> 00:11:56.360 responsible I is important, then may find the limits or the scope of what 162 00:11:56.440 --> 00:12:00.200 they see their job as being. And that's then where the AI supplied shape 163 00:12:00.440 --> 00:12:03.320 or value chain sort of notion became helpful, and that it sort of helped 164 00:12:03.399 --> 00:12:11.039 us zoom out from one particular developers context or sense of responsibility to seeing how 165 00:12:11.159 --> 00:12:16.840 they relate to the many other developers because the many other modules that are necessary 166 00:12:16.039 --> 00:12:20.080 to create finished AI systems, and how responsibility does or does not sort of 167 00:12:20.080 --> 00:12:24.440 transfer throughout that those chains. I think that's that's a really important framing as 168 00:12:24.440 --> 00:12:28.840 you're inquiring about this, because there is, of course the literal business and 169 00:12:28.919 --> 00:12:33.000 lived reality of how developers work in a system, right, and I think 170 00:12:33.039 --> 00:12:35.879 I think of echoes too of I mean this sounds weird to say this, 171 00:12:35.919 --> 00:12:39.399 but like assembly line factory labor too right that you have your part of the 172 00:12:39.440 --> 00:12:41.240 car and Henry Ford's factory, you build this part of the car and you 173 00:12:41.279 --> 00:12:43.600 do that over and over again, and so you're not responsible for when the 174 00:12:43.600 --> 00:12:46.840 car gets on the road or are you as one of the questions that you'll 175 00:12:46.879 --> 00:12:50.799 are poking as and I think you know you're absolutely right there too done where 176 00:12:50.879 --> 00:12:54.120 it's the language that we use shapes how we approach the process, what we 177 00:12:54.159 --> 00:12:58.480 think about what it is that we're quote unquote supposed to be doing. And 178 00:12:58.519 --> 00:13:01.240 so that kind of question I was putting us up with to his like responsible 179 00:13:01.279 --> 00:13:03.799 AI as again a set of practices, as like where did it come from? 180 00:13:03.919 --> 00:13:07.200 Versus a sense of responsibility that an individual hasn't And I think it's really 181 00:13:07.240 --> 00:13:11.879 important that you've exploded that, say, we have to look between these two 182 00:13:11.960 --> 00:13:16.679 in terms of if we're talking about developers should do something, which developers, 183 00:13:16.879 --> 00:13:20.639 who's making it and where David to kind of drawn on some of the language 184 00:13:20.639 --> 00:13:22.720 you brought up to in terms of Are we upstream? Are we downstream? 185 00:13:22.720 --> 00:13:26.559 Where are we in the supply chain? Are we very close to consumer end 186 00:13:26.679 --> 00:13:31.399 products and features? Are we at the first base code right where a consumer 187 00:13:31.399 --> 00:13:33.799 will never see that unless there's some get a nerd that wants to dive into 188 00:13:33.799 --> 00:13:37.000 it if it becomes open source later. Right? And so one thing I'm 189 00:13:37.039 --> 00:13:39.840 curious to think about, two is this kind of typical process, you know 190 00:13:39.879 --> 00:13:43.279 down you begin walking us into this space, and so thinking about what is 191 00:13:43.320 --> 00:13:48.240 that development process typically look like? So we have the both the metaphor and 192 00:13:48.240 --> 00:13:50.840 I think the reality right of working in these modules, right, these like 193 00:13:50.960 --> 00:13:54.759 these sections of ship as it were that they end of help move part one 194 00:13:54.799 --> 00:13:56.320 to Part two to Part three to part four. But then also I think 195 00:13:56.320 --> 00:13:58.600 the other piece that I want to hear from you two is like how developers 196 00:13:58.639 --> 00:14:03.279 also see themselves within these these modules. So David, your perspective, Howard, 197 00:14:03.240 --> 00:14:07.480 developers thinking about themselves in the development process as part of this chain, 198 00:14:07.519 --> 00:14:09.960 as part of these different modules, what does that look like for them? 199 00:14:09.399 --> 00:14:13.600 Starting with something you sort of said a little bit ago thinking about modularity. 200 00:14:13.639 --> 00:14:18.879 I mean I sit in a PhD in software engineering program where we read sort 201 00:14:18.919 --> 00:14:22.000 of the papers that said, Hey, one day, systems are going to 202 00:14:22.039 --> 00:14:24.799 become so big we can't think about them all at once. We have to 203 00:14:24.799 --> 00:14:28.159 think about modularity, and this will enable all these wonderful benefits such as division 204 00:14:28.200 --> 00:14:33.240 of labor in the factory context you just talked, and so really, you 205 00:14:33.240 --> 00:14:37.279 know, modularity is held up in software engineering as sort of an unquestioned core 206 00:14:37.360 --> 00:14:43.720 principle that allows us to build scale, to build large systems and have someone 207 00:14:43.960 --> 00:14:48.279 be responsible for one thing and then someone else be responsible for using that thing, 208 00:14:48.320 --> 00:14:52.639 but not knowing how it works. So in terms of how does AI 209 00:14:52.799 --> 00:14:56.399 finished AI systems today get built, modularity allows it so that not each developer 210 00:14:56.440 --> 00:15:01.320 has to understand the inner workings of the whole system, so that we can, 211 00:15:01.440 --> 00:15:03.840 for example, oh, this is the standard data set. I got 212 00:15:03.840 --> 00:15:05.480 it off the shelf. I don't know exactly how it was collected, and 213 00:15:05.519 --> 00:15:09.919 I don't know exactly you know the issues at play, but we agree that 214 00:15:09.960 --> 00:15:13.320 this is the standard module for this kind of thing. Or this is the 215 00:15:13.360 --> 00:15:18.000 facial recognition model that we're going to use, and really don't need to know 216 00:15:18.039 --> 00:15:20.879 the minute details of how it works, but we can tune it to to 217 00:15:22.080 --> 00:15:26.240 work in this particular context. Or whatever for this particular product. So none 218 00:15:26.279 --> 00:15:28.840 of this is terribly controversial. I mean, people use these sort of kickstarts 219 00:15:30.000 --> 00:15:35.080 to make software engineering and building ani systems faster, to make it so they 220 00:15:35.120 --> 00:15:41.159 can be built larger without having to intimately understand the inner workings, and developers 221 00:15:41.200 --> 00:15:43.559 would of see this as part of being a good developer. You ought not 222 00:15:43.759 --> 00:15:48.480 make the person using your module understand too much about how you built the thing 223 00:15:48.480 --> 00:15:52.639 because it ought to work and just you know, render its interface easily understandable 224 00:15:52.639 --> 00:15:54.720 to the next person. You know, as David is talking, well, 225 00:15:54.759 --> 00:15:58.000 I'll say two things. First is that I think doing this work with David 226 00:15:58.039 --> 00:16:03.120 helped me appreciate that there are times and places where it actually does make sense 227 00:16:03.799 --> 00:16:08.200 to not have to get into the inner workings of every single little thing, 228 00:16:08.240 --> 00:16:14.120 particularly when you're building complex systems. I think I've come to appreciate that more. 229 00:16:15.480 --> 00:16:22.360 At the same time, it has all the hallmarks of what Charles Perraut 230 00:16:22.360 --> 00:16:27.919 would call normal accidents. Right, you build up system that is sufficiently complex 231 00:16:29.840 --> 00:16:36.039 where no one person necessarily understands the whole because you can't cognitively, you just 232 00:16:36.159 --> 00:16:44.759 can't. That's a pretty good scenario for opening the door to failure, right, 233 00:16:44.799 --> 00:16:48.240 because you have systems that layer upon systems upon systems, and then all 234 00:16:48.279 --> 00:16:52.080 of a sudden you in that layer and you have a through line to badness 235 00:16:52.080 --> 00:16:56.600 happening of some kind. His example is nuclear power accidents, right, Why 236 00:16:56.679 --> 00:17:00.840 they continue to happen despite all this guards thrown on top and thrown on top 237 00:17:00.879 --> 00:17:06.640 and thrown on top. Right. In this context, we have something similar 238 00:17:06.799 --> 00:17:10.920 where everything things do layer over time and it is actual work. I mean 239 00:17:10.960 --> 00:17:14.359 part of the problem here is it takes work to actually go digging through. 240 00:17:14.440 --> 00:17:18.960 Oh wait, is the problem actually and the training data side, right, 241 00:17:18.039 --> 00:17:22.319 what the models take in and learn from? Is it in Oh? Actually 242 00:17:22.400 --> 00:17:26.359 something went wrong with the model itself, Like they didn't tune in the right 243 00:17:26.400 --> 00:17:30.119 way? Is it in way the entity of the company putting it out in 244 00:17:30.160 --> 00:17:34.599 the world actually could have made better choices about what context? Or is it 245 00:17:34.640 --> 00:17:37.480 in the end? Use? Right? One of the things we know from 246 00:17:37.480 --> 00:17:41.039 the anthropology is, of course using technology is a creative endeavor, right, 247 00:17:41.200 --> 00:17:48.279 That's just true. So so there's this distance that you've got to travel through, 248 00:17:48.559 --> 00:17:49.799 and it takes a lot. It takes both a lot of work and 249 00:17:49.839 --> 00:17:55.400 a lot of kind of intellectual agility to go through all of that difference, 250 00:17:55.519 --> 00:17:59.519 right, you know, modularities there for a reason, and really what we're 251 00:17:59.559 --> 00:18:03.880 calling for is an apparatus to help folks traverse through it. Right. So 252 00:18:03.920 --> 00:18:07.720 it might be that instead of just saying, look, okay, here's my 253 00:18:07.720 --> 00:18:12.480 box, I did my work right, maybe it's things like asking the person 254 00:18:12.680 --> 00:18:15.640 from whom you received your parts, right, your data whatever it is, 255 00:18:15.759 --> 00:18:18.920 or your open source model, what's in this box? Like what do I 256 00:18:18.960 --> 00:18:22.319 need to know about this box? Not necessarily what's in it, but what 257 00:18:22.359 --> 00:18:26.160 do I need to know about how it behaves? Right? And building that 258 00:18:26.240 --> 00:18:30.279 reflex a bit better, like things like that that kind of deepening the connections 259 00:18:30.359 --> 00:18:33.799 or sort of the thing we're after here, and not necessarily just like what 260 00:18:33.880 --> 00:18:36.839 do I need to know what's in this box to use it? Which is 261 00:18:36.920 --> 00:18:40.720 kind of the state of affairs now, but like for purposes of thinking about 262 00:18:40.759 --> 00:18:44.319 ethical questions, like what do I need to know about what's in this box? 263 00:18:44.599 --> 00:18:48.599 For thinking about my own relation to the next person who might use what 264 00:18:48.640 --> 00:18:52.680 I'm building and taking away, oh this is a standard component as sort of 265 00:18:52.680 --> 00:18:56.880 a common refrain, and thinking more no, like I'm building a system that 266 00:18:56.960 --> 00:19:00.319 is using a standard component. But I still want what I'm building to not 267 00:19:00.359 --> 00:19:04.200 cause harm down for the next user or for the people who deploy it downstream, 268 00:19:04.359 --> 00:19:08.119 or on whom it's deployed downstream. You know, I want to think 269 00:19:08.200 --> 00:19:11.319 more about like what I'm depending on, you know, who I'm depending on, 270 00:19:11.519 --> 00:19:15.759 and what the not just will it work? Will it fulfill the immediate 271 00:19:15.799 --> 00:19:18.880 goal of what I'm trying to build, but which is already fulfilled through modularity 272 00:19:18.920 --> 00:19:23.880 and sort of the interfaces that permit functionality, but thinking more about asking those 273 00:19:23.880 --> 00:19:29.039 same kind of questions that don't outlined for ethical reasons, as like an analog. 274 00:19:29.319 --> 00:19:33.160 If you're writing a dissertation or a research article, you are in essence 275 00:19:33.200 --> 00:19:37.680 doing something similar of pulling a library sometimes literally of previous research and ideas and 276 00:19:37.720 --> 00:19:41.640 thinkers to the amalgam them in a different way that can then be consumed in 277 00:19:41.680 --> 00:19:45.359 a new way for different audiences. And we may not think about maybe we 278 00:19:45.359 --> 00:19:48.559 should causing harm the same way when we're when we're putting a paper, dissertation 279 00:19:48.640 --> 00:19:51.519 or book out into the world. But I mean, if we are putting 280 00:19:51.519 --> 00:19:56.480 something that people can consume, obviously a book doesn't make decisions or maybe shapeway 281 00:19:56.559 --> 00:20:00.359 traffic light works the way that generative AI may at some point. But the 282 00:20:00.599 --> 00:20:03.559 modularity example, I think it's really interesting. Then I do think that there's 283 00:20:03.599 --> 00:20:06.400 there's good ways to echo that. I think in a typical research process too. 284 00:20:06.519 --> 00:20:08.519 And just again, so if folks either don't work in computer science or 285 00:20:10.039 --> 00:20:11.440 like it still feels like a black box, then part of it is this 286 00:20:11.480 --> 00:20:15.000 too, like what resources are we drawing from ahead of time to then build 287 00:20:15.359 --> 00:20:18.880 into our system now to then see where where it goes next. And we 288 00:20:18.920 --> 00:20:22.720 may find ourselves at one point and when a book itself can also then you 289 00:20:22.720 --> 00:20:25.960 know, write it's next chapter. But that's I think we're not quite there 290 00:20:26.359 --> 00:20:30.039 yet. I mean, maybe we are. I want to read exactly. 291 00:20:32.279 --> 00:20:33.720 You know. It's funny you mentioned writing, Adam. I mean, I 292 00:20:33.759 --> 00:20:38.200 think one of the things I learned in writing with a computer science student is 293 00:20:38.240 --> 00:20:45.519 that, like mondularity is really epistemological. It's not just I know what my 294 00:20:45.640 --> 00:20:48.319 job is and I do my job. It's like it gets into how you 295 00:20:48.359 --> 00:20:55.559 think, to the point where computer scientists actually don't use word processors. This 296 00:20:55.640 --> 00:21:00.039 is a thing. It's called overleaf, latex or late I've heard all the 297 00:21:00.319 --> 00:21:04.039 you know, many different ways of pronouncing it. And it's wild, because 298 00:21:04.079 --> 00:21:07.480 okay, it's wild if you're an anthropologist, right, it's totally normal if 299 00:21:07.519 --> 00:21:11.160 you're computer scientists. But I am an anthropologist, so I'm gonna call it 300 00:21:11.200 --> 00:21:15.039 wild. And what happens is like the way you write is you code and 301 00:21:15.079 --> 00:21:21.839 then you render, right, So it's lines, I mean, it's lines 302 00:21:21.880 --> 00:21:25.160 of texts. It's legible. There are a couple of commands that if you're 303 00:21:25.960 --> 00:21:30.400 smart, where you could learn potentially I did not. But other than that, 304 00:21:30.519 --> 00:21:33.920 it's like like the final text is a separate thing. It's like you 305 00:21:33.960 --> 00:21:37.359 know, compiling code, and it's meant to be like code is meant to 306 00:21:37.400 --> 00:21:44.359 be built in modules like good Apparently texting computer science is also meant to be 307 00:21:44.440 --> 00:21:48.119 in clean modules. And I couldn't do it. I could not write this 308 00:21:48.200 --> 00:21:53.680 paper like because there were these parts and there was no there were no visual 309 00:21:53.720 --> 00:22:00.000 assets to actually build an interlocking argument in a way that I normally would. 310 00:22:00.799 --> 00:22:07.279 And Jeremy Bonkers, you're very confident, you know, happy with with Lake 311 00:22:07.359 --> 00:22:11.559 textbook. Well, again, it's the unquestioned norm in my field. Like 312 00:22:11.599 --> 00:22:15.400 I also had to learn this early on in my PhD, and you know, 313 00:22:15.920 --> 00:22:19.079 didn't have fun either um, but like to collaborate with anyone, and 314 00:22:19.160 --> 00:22:22.799 at least within computer science is sort of expected to go back to the point 315 00:22:22.839 --> 00:22:26.839 you know, Don was making um like, yeah, this is designed just 316 00:22:26.000 --> 00:22:30.759 to with particular like rational goals in mind, like it's supposed to save labor, 317 00:22:30.799 --> 00:22:33.440 and that you can literally import package, use package, like, oh, 318 00:22:33.480 --> 00:22:37.640 you want a package, it's going to make really fancy tables and format 319 00:22:37.640 --> 00:22:41.119 them nicely, so you ideally can save labor or at least if you know 320 00:22:41.160 --> 00:22:44.720 how to use them. You import a package, you import someone else's fancy 321 00:22:44.759 --> 00:22:49.119 table module and and then use it. But once something goes wrong with one 322 00:22:49.119 --> 00:22:52.519 of those packages, good luck trying to fix the problem. Or also then 323 00:22:52.559 --> 00:22:56.960 once you try and sort of render the thing or work with people who are 324 00:22:56.200 --> 00:22:59.839 used to what computer sciences were called busy lake, what you see is what 325 00:22:59.880 --> 00:23:03.440 you get editors like word, that's where problems and or hilarity ensues. The 326 00:23:03.519 --> 00:23:10.440 modularity is sort of seen as a technical practice for people in my field. 327 00:23:10.759 --> 00:23:12.880 It's seen as just simply the way we build code. And we can have 328 00:23:12.960 --> 00:23:18.559 debates about the right way of modularizing things, you know, to fulfill one 329 00:23:18.759 --> 00:23:22.759 value or one goal or another. But one of what what Don practly pointed 330 00:23:22.759 --> 00:23:26.200 out is she as she spoke there was it functions more as an epistemic culture. 331 00:23:26.240 --> 00:23:32.440 It sort of transcends the purely technical purpose and then begins to be used 332 00:23:32.440 --> 00:23:37.079 as a way we think for the way we organize different concerns, separate concerns 333 00:23:37.079 --> 00:23:41.200 to use the lingo in our field. And the hardest thing about is trying 334 00:23:41.200 --> 00:23:45.960 to explain the point of this paper to people within my field is drawing the 335 00:23:45.000 --> 00:23:52.359 connection between modularity as a technical practice, which we've talked about the implications and 336 00:23:52.599 --> 00:23:57.119 benefits and disadvantages of here, and also to explain how that sort of then 337 00:23:57.160 --> 00:24:02.079 transcends and it becomes an epistemic culture where it's seen as the right way to 338 00:24:02.200 --> 00:24:04.359 organize relations, or the right way to do work, or the right way 339 00:24:04.400 --> 00:24:07.960 to do much broader thing. And so as I try to explain that the 340 00:24:08.000 --> 00:24:12.759 implications of modularity, it's hard to get people who see it as a technical 341 00:24:12.799 --> 00:24:18.759 practice and purely that to see some of it's perhaps what we might call like 342 00:24:18.799 --> 00:24:23.680 ethical side effects or implications in the way we think as as computer scientists. 343 00:24:25.039 --> 00:24:30.160 Yeah, at the end of the paper, we have these three different proposals, 344 00:24:30.279 --> 00:24:33.880 right, so everybody recognizes that this is a real problem. You know, 345 00:24:33.039 --> 00:24:36.680 what do you do about it? And one of the things we say 346 00:24:36.799 --> 00:24:41.480 is, look what you do about it depends precisely on how committed you are 347 00:24:41.519 --> 00:24:45.640 to modularity in the first place. And so if you are like, this 348 00:24:45.720 --> 00:24:48.720 is your bread and butter, this is what you do, right, Like, 349 00:24:48.759 --> 00:24:55.599 the action you take is even better modularity. That solves the problem. 350 00:24:55.680 --> 00:25:02.119 Right, So you assign somebody to do the biases, right, you frog 351 00:25:02.160 --> 00:25:07.160 marks your user experience people in to anticipating the usages, so you know, 352 00:25:07.400 --> 00:25:11.359 and then you can define like what the right usages should be going forward, 353 00:25:11.400 --> 00:25:15.640 and what's otoscope right. Even more, you just divide them the modules according 354 00:25:15.799 --> 00:25:21.119 to meeting the requirements of the chain, right, and then all the way 355 00:25:21.160 --> 00:25:26.359 the other side. We essentially had a kind of a just reject the notion 356 00:25:26.400 --> 00:25:30.720 of modularity entirely, right. So there are examples at this where if we 357 00:25:30.759 --> 00:25:34.119 look at Mary Gray's work, we look at what the folks over in dare 358 00:25:34.160 --> 00:25:38.960 are doing, which was a research institute recently set up after the Google got 359 00:25:40.039 --> 00:25:42.359 rid of it. Many of its ethicis right. What they're doing is they're 360 00:25:42.359 --> 00:25:48.480 saying that we do deep community work. Right, we take our time and 361 00:25:48.599 --> 00:25:52.920 we make people as much of a part of the development process, so much 362 00:25:52.920 --> 00:25:56.039 so that it doesn't matter what technology we end up with or not. The 363 00:25:56.079 --> 00:25:59.680 point is to meet people's needs period. Right. If this is not a 364 00:25:59.720 --> 00:26:03.160 technic analogy problem project, this is a problem solving project. And if we 365 00:26:03.240 --> 00:26:07.000 have the technical skills that actually helped, great, if not also great, 366 00:26:07.000 --> 00:26:10.240 we're going to solve the problem. And then in between we have a notion 367 00:26:10.279 --> 00:26:14.559 that types some more advantage of the institutional forms that businesses often use, right, 368 00:26:15.000 --> 00:26:18.559 might be contracts, a bunch of other stuff that is at neither end. 369 00:26:18.920 --> 00:26:22.079 The amazing part was the people who read this paper who come from David's 370 00:26:22.079 --> 00:26:26.440 part of the world, We're like, you can't reject modularity, Like, 371 00:26:26.519 --> 00:26:30.559 how would that even work? Like you can't kild anything, right, That's 372 00:26:30.559 --> 00:26:36.440 just not even possible. But then on the other side we had sts reviewers 373 00:26:36.440 --> 00:26:40.440 saying the implication was, I know, you're just putting in this improved modularity 374 00:26:40.480 --> 00:26:42.640 thing, is like this giveaway, but really the point is to reject it. 375 00:26:42.920 --> 00:26:48.880 Obviously, because the whole point of the papers leading up to rejecting modularity, 376 00:26:48.960 --> 00:26:49.799 right, just softballing in the middle, Right, yeah, yeah, 377 00:26:49.839 --> 00:26:53.240 yeah, we're just softballing in I'm a come from industry, you know, 378 00:26:53.319 --> 00:26:57.720 like you know those people, you know, so as like, wow, 379 00:26:57.920 --> 00:27:03.119 two highly different points folks got from the paper was wild? Is that a 380 00:27:03.119 --> 00:27:07.599 function of the fact that anthropologists and computer scientists are approaching this together, that 381 00:27:08.000 --> 00:27:11.519 there are these multiple interpretations. I think that the responses like I mentioned that 382 00:27:11.519 --> 00:27:18.000 I've gotten from computer scientists to speak to why I think it's important to think 383 00:27:18.039 --> 00:27:22.759 about how we might improve ethics within dominant logics and modularity that exists today. 384 00:27:22.839 --> 00:27:27.160 I mean, like, it's both funny that the first question people tend to 385 00:27:27.200 --> 00:27:30.720 ask me when I in computer science, when I talk about this paper is 386 00:27:30.720 --> 00:27:33.240 oh, well, how can we modularize ethics, Like what are the interfaces 387 00:27:33.279 --> 00:27:37.759 we can reveal from our modules that will like can you do an API call 388 00:27:37.839 --> 00:27:41.160 for bias or you know, like people will start start trying to then see 389 00:27:41.200 --> 00:27:47.920 it as a modularized or modularizeable problem, essentially presenting modularity is a dominant ideology 390 00:27:48.079 --> 00:27:53.000 that sort of subsumes and organizes other concerns like ethics beneath it. At the 391 00:27:53.079 --> 00:27:59.680 same time, modular systems are what structure large systems today and other people in 392 00:27:59.720 --> 00:28:02.240 my field will ask, well, like where does it end? Like if 393 00:28:02.240 --> 00:28:06.400 we can't depend on modules, can we not depend on compilers? So modularity 394 00:28:06.440 --> 00:28:08.920 is designed to manage complexity? How else would we build large systems? Now, 395 00:28:08.920 --> 00:28:11.200 again, that reveals sort of an idea that, you know, we 396 00:28:11.240 --> 00:28:15.319 ought to build large systems, and that's sort of the natural goal we must 397 00:28:15.400 --> 00:28:19.720 do that, the natural goal we must hold m So you know, a 398 00:28:19.799 --> 00:28:23.759 practical critique there. But and the more I think about it, the more 399 00:28:23.839 --> 00:28:29.920 like I think this speaks to sort of a different orientation from at least computer 400 00:28:29.960 --> 00:28:34.799 science and anthropology, which is like computer science tends to see the value in 401 00:28:36.039 --> 00:28:41.119 building new things, like building systems. And I actually had someone asked me 402 00:28:41.200 --> 00:28:45.240 recently like how long did you have to build things before you were allowed to 403 00:28:45.279 --> 00:28:51.359 just comment, you know, and reading this article or and of course like 404 00:28:51.400 --> 00:28:55.279 they were joking with me, they're a friend, you know. But the 405 00:28:55.359 --> 00:28:57.319 first goal of this kind of work, if this paper at least is not 406 00:28:57.400 --> 00:29:03.119 to say this is how we might build things better. The tagline at my 407 00:29:03.160 --> 00:29:04.640 department that's printed art t shirts is build it better. So that sort of 408 00:29:04.680 --> 00:29:10.480 shows you something about the orientation of software engineering departments. Perhaps, whereas thinking 409 00:29:10.519 --> 00:29:15.079 more carefully about sort of the epistemological questions embedded in the way we build things 410 00:29:15.720 --> 00:29:18.960 is seen as much more in anthropology at least, you know, Don can 411 00:29:18.000 --> 00:29:22.119 speak more to that. But I see that as more what I've learned from 412 00:29:22.119 --> 00:29:26.559 working with Don, what I've learned from collaborating with an anthropologist here. But 413 00:29:26.039 --> 00:29:29.319 like Vie, the common refrain of like, well, you pointed out a 414 00:29:29.319 --> 00:29:32.160 bunch of problems, Now what spurred us? Just think about the menu of 415 00:29:32.200 --> 00:29:37.039 options approach? Like okay, So you'll have some readers that are very much 416 00:29:37.240 --> 00:29:41.200 wanting to know what they can do better within the ethos of mondularity. But 417 00:29:41.240 --> 00:29:45.400 then of course we'll have other readers who are like not necessarily added to or 418 00:29:45.599 --> 00:29:52.359 sort of schooled in mondularity, who want to imagine other different futures that might 419 00:29:52.440 --> 00:29:59.599 have no immediately applicable within many organizational context sort of practical ability to be implemented. 420 00:30:00.680 --> 00:30:04.880 I would not call very grades project not practical to implement. It clearly 421 00:30:06.039 --> 00:30:10.359 was within certain organizational structures. Yeah, like that's all I mean, you 422 00:30:10.400 --> 00:30:12.119 know, all props to Microsoft, but it did happen. You know, 423 00:30:12.799 --> 00:30:17.319 she is that Microsoft. So there are times in places where you can do 424 00:30:17.359 --> 00:30:22.680 that, certainly nonprofits and geo's foundations, Right, I mean there's a whole 425 00:30:22.720 --> 00:30:29.680 world out there that, with proper science policy, could actually proliferate and could 426 00:30:29.720 --> 00:30:37.359 actually really develop and elaborate the kinds of broader social and technical technological trajectories that 427 00:30:37.440 --> 00:30:41.440 we might want to have. Right, That's not the world we live in 428 00:30:41.559 --> 00:30:45.319 now because we have an unequal situation. But I think they're entirely practical for 429 00:30:45.480 --> 00:30:48.759 that side of them. Yeah, And we get to that a little bit 430 00:30:48.799 --> 00:30:52.160 towards the end when we talk about the idea of like there's this unquestioned idea 431 00:30:52.200 --> 00:30:56.319 of like general purpose modules, like well, general purpose for who like the 432 00:30:56.400 --> 00:31:00.640 standard bricks or the standard shipping containers or modules that you can take off the 433 00:31:00.680 --> 00:31:06.440 shelf to do certain things. Well, do the existing general purpose modules exist 434 00:31:06.640 --> 00:31:10.720 to serve all of the kinds of things we might use software for equally or 435 00:31:10.759 --> 00:31:15.559 do they tend to push towards particular ways of using software for particular people, 436 00:31:15.599 --> 00:31:19.559 for particular needs business needs? So like what would you know? I don't 437 00:31:19.599 --> 00:31:25.319 know, As Don mentioned, perhaps like public funding for software maintenance and creation 438 00:31:25.400 --> 00:31:29.599 and sort of thinking of it as a public good approach. Perhaps or you 439 00:31:29.640 --> 00:31:34.640 know, implications for policy in that direction. We're going to take a quick 440 00:31:34.680 --> 00:31:37.000 break. Just wanted to let you know that we're running ads to support the 441 00:31:37.000 --> 00:31:48.359 show down. We'll be right back. Is there something about A in general 442 00:31:48.759 --> 00:31:52.400 that sparks this kind of conversation of what is my responsibility? If I back 443 00:31:52.480 --> 00:31:56.720 up ten years and I'm developing the code for Facebook's social graph, we think 444 00:31:56.720 --> 00:32:00.720 about that a bit differently in terms of a knowledge graph. And if I'm 445 00:32:00.759 --> 00:32:02.839 putting my notes together in Obsidian or something, right, I guess something about 446 00:32:02.880 --> 00:32:06.799 AI that makes us say we actually need this pause. I mean, obviously 447 00:32:06.839 --> 00:32:08.839 there's been you know, Donn You mentioned the problems at Google with with the 448 00:32:08.880 --> 00:32:14.759 firing of Tina Jibru and other issues like silencing or firing ethicists for raising questions. 449 00:32:15.200 --> 00:32:16.920 And I'm just thinking out out here, wondering both from your paper and 450 00:32:16.920 --> 00:32:21.960 through our conversation and just other things that we've seen recently. Sam Alt and 451 00:32:21.960 --> 00:32:23.680 the CEO of Open Ai, was testifying for a congress. You know, 452 00:32:23.720 --> 00:32:28.799 I was also perusing Lex Fridman's podcast and he had Max Techmark was talking about 453 00:32:28.839 --> 00:32:30.400 the case for halting AI development, or at least just slowing it. Down 454 00:32:30.720 --> 00:32:34.720 stop being in a quote unquote arms race to be the first out of the 455 00:32:34.759 --> 00:32:38.119 gate, because GPT four is actually just the baby program. We think it's 456 00:32:38.119 --> 00:32:42.839 a very pretty baby, but it's the first iteration of many more complex things 457 00:32:42.839 --> 00:32:45.839 that are to come. And so this begs the bigger question, how will 458 00:32:45.880 --> 00:32:50.039 AI change society? Or are we just more conscious now that we have social 459 00:32:50.079 --> 00:32:54.039 media that's changed society in many ways, mostly for worse, some for better, 460 00:32:55.039 --> 00:33:00.799 that we're thinking more intentionally about the impact AI might have. In other 461 00:33:00.839 --> 00:33:04.839 words, is there something about AI that begs us to rethink the development process? 462 00:33:05.680 --> 00:33:08.400 Such as modularity? It seems that there's a sense of urgency from the 463 00:33:08.400 --> 00:33:12.960 examples that we're talking through that A will get out of control at some point 464 00:33:13.000 --> 00:33:16.279 if we don't put safeguards in place. And so I'm curious for your perspective, 465 00:33:16.319 --> 00:33:21.799 how is AI reshaping how we approach the problem or process of development. 466 00:33:22.319 --> 00:33:27.319 You know, the issue is always what counts as AI is a moving kind 467 00:33:27.359 --> 00:33:30.359 of goalposts, and you know, when you get into the weeds of it. 468 00:33:30.519 --> 00:33:34.079 As far as I understand it, very few people actually use the term 469 00:33:34.119 --> 00:33:38.160 AI, although recently it's become more popular as as it becomes the term of 470 00:33:38.200 --> 00:33:42.839 the popular press, and I think you're right to point out that if you 471 00:33:43.039 --> 00:33:46.720 look at technologies and harm right, there's a long list of them right through 472 00:33:46.759 --> 00:33:52.720 the ethnographic and sociological record that we could point to, and arguably responsibility for 473 00:33:52.799 --> 00:33:59.680 that should have been taken well before machine learning came on the scene, right. 474 00:34:00.200 --> 00:34:02.759 I would love for somebody to actually dig into this properly. My guess 475 00:34:02.880 --> 00:34:08.920 is that when something like gender shades comes up, where it's so palpably odious 476 00:34:09.239 --> 00:34:16.079 that you have somebody not being seen literally because of who they are, you 477 00:34:16.119 --> 00:34:21.599 can no longer deny or say that, oh, tech is neutral and the 478 00:34:21.639 --> 00:34:27.920 only problem is how you use it, right, because clearly not so that 479 00:34:28.199 --> 00:34:34.559 might have been part of why we're here. I will say that I think 480 00:34:34.599 --> 00:34:40.719 the current sort of elaborations of the moment where folks have varying reasons to be 481 00:34:40.760 --> 00:34:46.840 concerned about generative AI shall we say personally as an anthropologist, I would say 482 00:34:46.880 --> 00:34:52.559 that some of them are more realistic than others. When I am concerned about, 483 00:34:52.880 --> 00:35:00.119 again as an individual, is being distracted by made up harms at the 484 00:35:00.199 --> 00:35:04.320 expense of, you know, the actual arms that we exist, And not 485 00:35:04.360 --> 00:35:08.400 only that, but for the flavors of machine learning that aren't generative. Are 486 00:35:08.400 --> 00:35:13.760 we going to suddenly give up because now we're worried about something that puts speech 487 00:35:13.800 --> 00:35:17.719 together, all of a sudden, we're anthropomorphizing, putting an imagination of personhood 488 00:35:17.760 --> 00:35:22.199 where it arguably doesn't belong, And so that's just fueling the fire. I 489 00:35:22.199 --> 00:35:27.760 think also, I have friends who are getting PhDs and more of the technical 490 00:35:28.239 --> 00:35:31.559 aspects behind AI, and there's a huge disconnect with between what they do and 491 00:35:31.599 --> 00:35:38.280 what, like most systems marketed as AI are are doing. I think like 492 00:35:38.760 --> 00:35:44.840 almost any software is now labeled AI software, at least in the public way 493 00:35:44.880 --> 00:35:49.320 it's message or the way it's marketed. And yeah, I think that can 494 00:35:49.360 --> 00:35:54.800 be distracting and sort of lead towards conceiving of AI ethics and questions of superhuman 495 00:35:54.800 --> 00:36:00.440 possible like sort of science fiction ey scenarios like Terminator and the AI that might 496 00:36:00.559 --> 00:36:04.880 kill us or might you know AGI. What I'm drawing, what I'm going 497 00:36:04.920 --> 00:36:10.320 towards here artificial general intelligence singularly, and with that is the very lifelike chat 498 00:36:10.360 --> 00:36:15.920 GPT, sort of human appearing convincingly putting together strings of words that sound very 499 00:36:15.960 --> 00:36:20.440 much like a human. But when you actually have domain expertise in what it's 500 00:36:20.760 --> 00:36:24.440 loviating about our less san But also in some of my more recent work, 501 00:36:24.480 --> 00:36:30.519 I've sort of begun to question the impact of the conceptual category of AI in 502 00:36:31.039 --> 00:36:37.440 responsible AI discussions, because, as we talked about earlier, like there's a 503 00:36:37.440 --> 00:36:39.400 certain kind of question that's like, how do we build this thing ethically? 504 00:36:39.800 --> 00:36:44.199 And there's another kind of question that's like, how do we use this or 505 00:36:44.199 --> 00:36:46.960 how will we permit this system to be used ethically? In what ways? 506 00:36:47.000 --> 00:36:50.360 Can we sell the system in what ways? And who should we sell the 507 00:36:50.400 --> 00:36:57.559 system too? And scoping the question to specifically responsible AI scopes the question at 508 00:36:57.639 --> 00:37:00.079 least in my view, to a design question like Okay, well we're using 509 00:37:00.119 --> 00:37:05.119 AI to do this thing. Now we figure out how to design it ethically, 510 00:37:05.360 --> 00:37:07.440 and now we figure out how to build it in an ethical way. 511 00:37:07.719 --> 00:37:09.199 We want we want to be fair, we want to be accountable, transparent, 512 00:37:09.239 --> 00:37:14.840 whatever, right now, That to me is another kind of trick that 513 00:37:15.000 --> 00:37:17.440 distracts us from well the question of whether we're not we need to use AI 514 00:37:17.480 --> 00:37:21.280 in the first place, machine learning or statistical techniques in the first place to 515 00:37:21.320 --> 00:37:25.960 do this thing, but then also renders out of scope some of the more 516 00:37:27.400 --> 00:37:31.559 difficult to answer questions when you're an individual software developer in a business. And 517 00:37:31.599 --> 00:37:35.519 this speaks to our study too. Do I want my software being used by 518 00:37:35.599 --> 00:37:38.000 the military? Do I want my software being used in this way for this 519 00:37:38.920 --> 00:37:45.119 purpose that I might have concerns with? So AI in itself can scope the 520 00:37:45.199 --> 00:37:49.400 kind of questions in a more narrow sense to questions of design rather questions of 521 00:37:49.559 --> 00:37:53.280 use, And that might be something we want to rethink as we think about 522 00:37:53.320 --> 00:37:57.559 AI ethics, because AI is certainly the term of the moment and the term 523 00:37:57.599 --> 00:38:00.360 of art used in both business and in a lot of development context too. 524 00:38:00.559 --> 00:38:05.519 But I think a lot of the same questions can be asked about software ethics 525 00:38:05.599 --> 00:38:10.280 or about technology ethics more broughtly and and the and not doing so can have 526 00:38:10.880 --> 00:38:15.199 certain implications for what is legitimate discuss or No. That's an incredibly helpful framework 527 00:38:15.239 --> 00:38:20.760 to think with because as an outsider to software developments, as I think about 528 00:38:20.760 --> 00:38:25.039 what we're talking through here, there is this question of the role that AI 529 00:38:25.119 --> 00:38:29.079 is also helping us. Then reflect back on to your point, David, 530 00:38:29.119 --> 00:38:31.800 like the broader elements of software development in the first place, right, Like, 531 00:38:31.840 --> 00:38:36.760 how is something that's like hot on people's minds, helping us reflect back 532 00:38:36.800 --> 00:38:39.280 on the entire production process. I think it's really valuable to ask this idea 533 00:38:39.320 --> 00:38:45.400 of if we're falling into a moment of we're trying to design something, then 534 00:38:45.480 --> 00:38:47.760 we might miss or scope out the question of ethics because we're not talking about 535 00:38:47.800 --> 00:38:51.719 it's use case. As someone who has taught in a design program of taught 536 00:38:51.719 --> 00:38:55.320 design thinking methodologies and research, I think about this because teaching UX classes user 537 00:38:55.320 --> 00:38:58.840 experience classes. You know, we always talk about remember one of your first 538 00:38:58.840 --> 00:39:00.400 things of doing user research is to find out who it is that you're designing 539 00:39:00.400 --> 00:39:05.039 for. And when we are talking about designed versus research, right, is 540 00:39:05.039 --> 00:39:07.000 there a level of research or user research that kind of goes into how we 541 00:39:07.079 --> 00:39:12.480 might think about this too, of either how quickly or at one point ethical 542 00:39:12.559 --> 00:39:15.920 questions do get asked? Like do they tend to get outsourced to the research 543 00:39:15.000 --> 00:39:19.119 department? I know in the paper you notice in the development chain it's like 544 00:39:19.239 --> 00:39:22.800 they tend to get pushed into different model cards later. But is the research 545 00:39:22.840 --> 00:39:25.480 team also an area that we tend to see the ethics questions get siloed or 546 00:39:25.519 --> 00:39:30.519 moved over? Yeah, it's like twofold right, So often user experience research 547 00:39:30.559 --> 00:39:37.000 comes in at the end of the chain and there's an actual interface, there's 548 00:39:37.039 --> 00:39:42.159 an actual product, like at that level, then that's when the user experienced 549 00:39:42.159 --> 00:39:46.519 folks typically have something to say. I think it's also the case that it's 550 00:39:46.800 --> 00:39:53.639 useful to do research kind of at that innovation level, which is one of 551 00:39:53.639 --> 00:39:57.320 the reasons I do state until for as long as I do, is that 552 00:39:57.320 --> 00:40:00.719 that is what we do. You know, we do food social research in 553 00:40:00.800 --> 00:40:08.840 the process of doing AI innovation itself, specifically to set trajectories in a way 554 00:40:08.840 --> 00:40:14.559 that are going to have social utility and uh, you know, a sense 555 00:40:14.559 --> 00:40:17.920 of responsibility from the start, right. That's something a big company can do 556 00:40:19.639 --> 00:40:24.480 that arguably is harder for smaller companies. What I am concerned about is being 557 00:40:24.519 --> 00:40:30.039 a nerd and understanding other people's nerdiness. And at some level I do get 558 00:40:30.039 --> 00:40:36.320 this, oh like like chasing the technological curiosity, like oh wait, if 559 00:40:36.320 --> 00:40:38.760 I could just make it do this, you know, like, but then 560 00:40:38.800 --> 00:40:46.400 you end up in a situation where your user is like the technological trajectory itself. 561 00:40:46.760 --> 00:40:52.599 And that's that's why we're here, right, right, I mean, 562 00:40:52.599 --> 00:40:59.239 we have people publicly making up uses on behalf of new forms of AI for 563 00:40:59.320 --> 00:41:05.079 no good reason and rights. I've seen statements to the effect of well generative 564 00:41:05.119 --> 00:41:09.320 AI of course will positively impact x y z, but there are negative implications. 565 00:41:09.320 --> 00:41:15.440 It's like, have you demonstrated that, No, Like you're just saying 566 00:41:15.480 --> 00:41:22.039 things now. There's a lot of just saying stuff and arguably more research is 567 00:41:22.199 --> 00:41:28.800 necessary. But then you get that flip side of when research as a whole 568 00:41:29.000 --> 00:41:32.280 enterprise. Right, Universities, think tanks all the rest of it are sitting 569 00:41:32.320 --> 00:41:40.559 around chasing the trajectory that's essentially been set by a combination of very powerful, 570 00:41:40.880 --> 00:41:45.480 usually white men, and you know the way that the engineering is unfolding. 571 00:41:46.440 --> 00:41:51.320 Right then, like at some certain level, those wise minds might be better 572 00:41:51.360 --> 00:41:54.679 spent looking at other, more socially productive things. Right, So we're in 573 00:41:54.679 --> 00:42:00.519 this kind of vicious loop of chasing the things that you know, really need 574 00:42:00.880 --> 00:42:07.440 significant levels of control at the expense of things that could potentially be better. 575 00:42:07.840 --> 00:42:13.079 That's a great point that how are different people's expertise like have to be deployed, 576 00:42:13.159 --> 00:42:15.079 right, Like, we have finite amounts of time and energy, and 577 00:42:15.119 --> 00:42:19.360 so if we're chasing these rabbit holes, sometimes sometimes legit threat sometimes not right, 578 00:42:19.519 --> 00:42:21.840 you know, then it does take away from these other these other projects 579 00:42:21.840 --> 00:42:23.840 that have incredible value. That's a really interesting question too. Yeah, there 580 00:42:23.840 --> 00:42:27.760 are a few points that done accidentally laid lay down there that I just want 581 00:42:27.760 --> 00:42:30.320 to pick up on. One we make in the in the paper, which 582 00:42:30.360 --> 00:42:35.559 is the idea that like user research leaves out actually a lot of people nowadays 583 00:42:35.559 --> 00:42:40.880 because software systems are deployed on people who are not necessarily users in the sense 584 00:42:40.960 --> 00:42:45.960 we think about them. There, people who are subject to algorithmic systems are 585 00:42:45.079 --> 00:42:49.840 rarely the people that we're sort of seen as the pain user. And so 586 00:42:49.880 --> 00:42:52.920 when we frame it as a user research problem, where it's probably thinking about 587 00:42:52.920 --> 00:42:58.360 our customers, are probably thinking about a particular frame of person who willfully engages 588 00:42:58.360 --> 00:43:00.800 with the system, or or at least knowingly engages with the system, rather 589 00:43:00.840 --> 00:43:06.760 than the much more pervasive effects that happen or that can happen for people who 590 00:43:06.800 --> 00:43:10.159 aren't in the user role. You can think about like maybe the police department 591 00:43:10.199 --> 00:43:15.880 are the users of the predicted policing software, but the data subjects are the 592 00:43:15.920 --> 00:43:20.599 ones whose thought crime is being predicted or future crime is being predicted. Right, 593 00:43:21.320 --> 00:43:23.000 those are data subjects, and those are not, at least with what 594 00:43:23.039 --> 00:43:27.760 I've been exposed to in my classes and some of my work is Yeah, 595 00:43:27.880 --> 00:43:30.639 users are thought of as one way, and we're hoping to think about the 596 00:43:30.679 --> 00:43:37.360 broader stakeholders, the broader sort of people implicated in how systems are deployed and 597 00:43:37.400 --> 00:43:40.840 developed, and along with the broad many more other researchers focusing on that question 598 00:43:40.880 --> 00:43:45.880 specifically. I also think, like when we you know, is this a 599 00:43:45.000 --> 00:43:49.800 user research problem? I mean I think this quite rightly. It's just pointing 600 00:43:49.800 --> 00:43:52.599 out speaks to what is you valued by different sort of roles, different sort 601 00:43:52.639 --> 00:43:59.639 of titles within orcs, Like software engineering has a particular history and a particular 602 00:43:59.679 --> 00:44:02.400 culture from coming from that history of like what is valued as a software engineer, 603 00:44:02.480 --> 00:44:06.800 like making working code, making it faster, making it more efficient, 604 00:44:06.880 --> 00:44:09.719 making it more modular, the various like ways that people have written about this. 605 00:44:09.800 --> 00:44:13.920 But code is good things, and we talked about in the paper house. 606 00:44:13.920 --> 00:44:15.679 I think it was a technical lead was glad that that they were hoping 607 00:44:15.719 --> 00:44:20.480 to soon modularize out the parts where they had to deal with people. We 608 00:44:20.559 --> 00:44:22.079 have sales leads to do that, we have user experience researchers do that, 609 00:44:22.119 --> 00:44:28.159 so sort of a conception of what is within scope for articular role or not 610 00:44:28.519 --> 00:44:31.480 can make it harder to then reassemble the various sort of partial knowledges that are 611 00:44:31.559 --> 00:44:36.320 necessary, the deep technical knowledges that are necessary to understand the technic, you 612 00:44:36.320 --> 00:44:39.159 know, the implications of the technical choice on the wider world, and then 613 00:44:39.159 --> 00:44:45.519 the concerns about the wider world that are then implicated or that have implications for 614 00:44:45.559 --> 00:44:49.360 how we design the technical thing. So that's part of why we frame sort 615 00:44:49.400 --> 00:44:53.199 of the paper as well. If we have dislocated accountabilities, how can we 616 00:44:53.519 --> 00:44:59.880 actually reassemble partial knowledges using the broader frame of the supply chain of the values 617 00:45:00.679 --> 00:45:04.320 and on the point of chat GPT and all that. And we see this 618 00:45:04.440 --> 00:45:08.280 like modularization of concerns. It's not my problem of concerns into the way that 619 00:45:08.199 --> 00:45:13.519 that has been talked about. Like maybe some folks at least listening are familiar 620 00:45:13.559 --> 00:45:17.119 with the sparks of AGI paper, the paper that said that chat GPT is 621 00:45:17.599 --> 00:45:22.440 demonstrating sparks of artificial intelligence and because it can draw unicorn well or something. 622 00:45:22.800 --> 00:45:27.760 But one of the claims they made in the papers that since we don't have 623 00:45:27.800 --> 00:45:32.679 access to the vast full details of GPT for its training data. We have 624 00:45:32.719 --> 00:45:36.800 to assume it's seen every benchmark. So in this sense, it's sort of 625 00:45:36.880 --> 00:45:40.039 saying, well, it's probably seen everything on the web, and so carefully 626 00:45:40.039 --> 00:45:44.280 inspecting, you know, it's data set to make sure that it is appropriate 627 00:45:44.280 --> 00:45:47.440 for use or doesn't have bias. And then, as Don says, disconnected 628 00:45:47.480 --> 00:45:52.639 from any sense of what in particular it will be used for, for whom 629 00:45:52.639 --> 00:45:54.239 it will be used, what problem is it trying to solve. We see 630 00:45:54.239 --> 00:45:58.639 that disavowal as like, the data set is out there, the data is 631 00:45:58.679 --> 00:46:00.119 not our problem. We are building this. I think I think it's a 632 00:46:00.159 --> 00:46:05.639 super valuable point across both of what you're saying that it's easy, I mean, 633 00:46:05.800 --> 00:46:08.239 upon reflection to see where and why some of those disconnects can take place, 634 00:46:08.320 --> 00:46:10.559 right in terms of what is my responsibility, what's the scope of what 635 00:46:10.599 --> 00:46:15.960 I'm producing and making based on my expertise of technical knowledge? You know that's 636 00:46:15.960 --> 00:46:19.639 either a user problem or like an outsource a question of like how will it 637 00:46:19.679 --> 00:46:23.480 be used? You're both approaching this. I think that's really important and challenging 638 00:46:23.559 --> 00:46:30.119 question. That is taking stock of the existing epistemological framework of how developers do 639 00:46:30.159 --> 00:46:32.239 what they do, Like how we make software today, right, we use 640 00:46:32.280 --> 00:46:35.719 modula systems in order to understand them. Taking part at a part. Be 641 00:46:35.719 --> 00:46:37.239 I'm in a barrow from this library. I'm going to hand my box off 642 00:46:37.239 --> 00:46:39.079 to to David. He's gonna put his pieces in there, do his thing 643 00:46:39.079 --> 00:46:40.719 to hand it off the dawn a babbabod. It's going to go on the 644 00:46:40.840 --> 00:46:44.000 chain. At some point, some users are going to try it out. 645 00:46:44.280 --> 00:46:45.440 But the police, example, say it was actually really great, Like when 646 00:46:45.440 --> 00:46:50.000 you say users were actually like the police may be the users of the technology, 647 00:46:50.039 --> 00:46:52.000 but actually the subjects of the data are an entirely different group of people 648 00:46:52.000 --> 00:46:55.360 that have no input or you know, they probably didn't give permission to be 649 00:46:55.519 --> 00:46:59.280 recorded in the ways that they were recorded, right, I mean what happens 650 00:46:59.320 --> 00:47:00.960 with their data? So this is I think, I think one of the 651 00:47:00.000 --> 00:47:05.880 big challenges. And also I appreciate why you kind of present three menu options, 652 00:47:06.000 --> 00:47:08.480 right of like what could we do with this system in terms of do 653 00:47:08.599 --> 00:47:12.639 we just burn it and get rid of modularity? Do we like actually increase 654 00:47:12.760 --> 00:47:15.559 relationships between them? You know, what is the kind of the way the 655 00:47:15.599 --> 00:47:19.679 way of finding our path forward? So I think I'm kind of coming away 656 00:47:19.719 --> 00:47:22.320 from this feeling a lot more optimistic, but at least that the contours of 657 00:47:22.360 --> 00:47:25.360 the problem feel clearer. So asking about that, you know, what do 658 00:47:25.400 --> 00:47:28.920 we need to be paying attention to, you know going forward? What's like 659 00:47:29.000 --> 00:47:30.440 caught your attention because of doing this paper, Like what do you now see 660 00:47:30.440 --> 00:47:34.000 that you didn't see before? You know? I appreciate your pointing out that 661 00:47:34.039 --> 00:47:37.920 part of the job of the researcher is to identify the contours of the problem. 662 00:47:37.400 --> 00:47:44.360 And I will say that in the course of drafting and putting out preprint, 663 00:47:44.599 --> 00:47:50.679 finally putting in proper publication, that just through time, the acknowledgement of 664 00:47:51.360 --> 00:47:53.800 that this is a real issue is coming from a lot more quarters, and 665 00:47:53.880 --> 00:48:00.840 I think it did when we started. So in that sense, on the 666 00:48:00.920 --> 00:48:04.239 optimistic side, that is what I would be looking for, is people really 667 00:48:04.320 --> 00:48:10.079 digging into what are the relationships between different institutions, between different developers. Then 668 00:48:10.480 --> 00:48:15.320 if I had a crystal ball, there might be an elaboration of what developers 669 00:48:15.360 --> 00:48:19.599 already do today. Right, So as an example, we found we found 670 00:48:19.639 --> 00:48:23.800 cases where developers would slow walk dodging work. It would find other projects to 671 00:48:23.880 --> 00:48:29.920 prioritize because there's always something else to prioritize, right, so we might see 672 00:48:29.960 --> 00:48:35.599 more activity activity there. I mean certainly as regulation gets more serious, right, 673 00:48:36.079 --> 00:48:38.960 the game changes again, and that that seems set to happen, so 674 00:48:39.440 --> 00:48:42.440 it's not clear, and then you know, we'll also have to see what 675 00:48:42.519 --> 00:48:45.679 the next flavor of the day is. Also true, yes, then that 676 00:48:45.760 --> 00:48:51.159 is right yet, but it always inflects old issues, right, I mean, 677 00:48:52.519 --> 00:48:55.480 you know we've always had you know, supposed AGI right, like that's 678 00:48:55.480 --> 00:48:59.719 been a long term thing, So it'll surface things that are old instance, 679 00:48:59.760 --> 00:49:02.199 to just to jump down, that's exactly. And this is where it's even 680 00:49:02.239 --> 00:49:06.519 like AI feels new, but it's making us rethink the existing way that we 681 00:49:06.599 --> 00:49:08.079 make software. Right, So it wasn't like oh, this is new how 682 00:49:08.159 --> 00:49:10.840 we make software. It's more like, wait, let's think about why we're 683 00:49:10.880 --> 00:49:14.360 making it the way that we're making it because there's new implications for us. 684 00:49:14.360 --> 00:49:15.440 So I think that that's that's a great that's a great way to think about 685 00:49:15.480 --> 00:49:19.800 that. Since Don and I did this work, I've done some follow up 686 00:49:19.800 --> 00:49:23.159 pieces that ask a question like, you know, okay, so say we 687 00:49:23.280 --> 00:49:29.000 have developers who want to ask the down the supply chain questions who ask like, 688 00:49:29.480 --> 00:49:32.480 well, hang on, I'm uncomfortable that my software is being used down 689 00:49:32.559 --> 00:49:37.000 the chain by the US military. What we did in that in the sort 690 00:49:37.039 --> 00:49:40.719 of follow on work is survey a bunch of software developers with ethical concerns, 691 00:49:42.280 --> 00:49:45.360 and when they try and raise those concerns or they try and act on those 692 00:49:45.400 --> 00:49:51.159 concerns, I mean, they don't really have power to power is modularized itself. 693 00:49:51.199 --> 00:49:55.920 People have power over particular part of the supply chain. So I think 694 00:49:57.000 --> 00:49:59.920 that if I had to go from that to sort of thinking about the future, 695 00:50:00.039 --> 00:50:02.920 we're going to think about different ways that we can build power, either 696 00:50:04.039 --> 00:50:07.480 in part different parts of the supply chain, through collective action, through sort 697 00:50:07.519 --> 00:50:12.480 of organizing efforts. We're going to think about ways that we can build power, 698 00:50:12.519 --> 00:50:14.519 I mean, and part of that will be building power sort of by 699 00:50:14.559 --> 00:50:17.880 connecting different parts of the supply chain, like having more relationships outside of the 700 00:50:19.000 --> 00:50:24.159 chain that allow people to control more of it or to understand at least the 701 00:50:24.599 --> 00:50:30.559 impact of what they're creating down the chain. Now, also, I'm thinking 702 00:50:30.639 --> 00:50:35.239 about again how we conceive of what our job is, how we conceive of 703 00:50:35.360 --> 00:50:38.960 what we're allowed to raise as concerns, and sort of the scope of those 704 00:50:39.079 --> 00:50:45.199 so I'm in some very recent following work is thinking about different ways that we 705 00:50:45.320 --> 00:50:51.760 can give people sort of social permission to think or thinking hypotheticals that don't speak 706 00:50:51.800 --> 00:50:58.079 to what they're currently building, and allow people to talk about concerns or ethical 707 00:50:58.119 --> 00:51:01.519 concerns that they have that in ways that aren't a part of their ordinary day 708 00:51:01.559 --> 00:51:07.199 to day assigned duties or job roles. I mean, if I had to 709 00:51:07.599 --> 00:51:09.559 sort of outline where I think we should go next, it's really starting to 710 00:51:09.639 --> 00:51:15.760 interrogate questions of power and asking whether this is a you know, and a 711 00:51:15.840 --> 00:51:20.519 lot of framings of a ethics, just a question of individual developers building software 712 00:51:20.559 --> 00:51:22.400 better, or is there's more collective action and collective power that needs to be 713 00:51:22.480 --> 00:51:27.360 built, and then also thinking about opportunities for action that we have within our 714 00:51:27.360 --> 00:51:32.920 own armed rules to say and think about things that aren't necessarily within our particular 715 00:51:34.360 --> 00:51:37.920 job mondual job role. And I'll just very briefly add if the future involves 716 00:51:38.079 --> 00:51:45.239 more computer scientists using phrases like partial knowledges in a sentence, I will also 717 00:51:45.639 --> 00:51:52.320 consider that to be a better future, more computer scientists talking about the how 718 00:51:52.400 --> 00:51:55.719 power actually works. Better off we are signed me up to that that sounds 719 00:51:55.800 --> 00:51:59.320 right on, you know, it's like, let's get some meta modules going 720 00:51:59.639 --> 00:52:02.039 aside to the other modules or something. But I want to say Don and 721 00:52:02.239 --> 00:52:05.679 David, thank you so much for joining me on the pot today. This 722 00:52:05.719 --> 00:52:08.800 has been absolutely fascinating conversation and I'm really excited to dive in with our listeners 723 00:52:08.840 --> 00:52:12.039 too and just get a sense of like where folks are at with this, 724 00:52:12.119 --> 00:52:15.320 because I mean, AI is loud as we know, right, it's everywhere 725 00:52:15.360 --> 00:52:17.400 in the news, pulling us away from our desks constantly to go look at 726 00:52:17.480 --> 00:52:21.639 some other random story. And I think that the work that you're doing here 727 00:52:21.719 --> 00:52:24.320 is really important of helping us actually take a moment to shine the light. 728 00:52:24.400 --> 00:52:28.599 Let's actually reflect on the process of how this gets made in the real world 729 00:52:28.760 --> 00:52:32.280 with real developers and ask is that process, this, this epistemological technical process 730 00:52:32.599 --> 00:52:37.599 the best way forward if we're thinking about a broader scope of of you know, 731 00:52:37.679 --> 00:52:39.920 well being in a bigger sense of users, a bigger sense of implications 732 00:52:39.960 --> 00:52:44.440 and ethics and roles. So it's a big order, but y'all squeeze it 733 00:52:44.480 --> 00:52:47.280 into like a sixteen page paper. So that's an excellent start, And excited 734 00:52:47.280 --> 00:52:50.079 to see to see where we can build next. So thanks, thanks so 735 00:52:50.159 --> 00:52:52.039 much for joining me on the on the pot today. Thank you wonderful, 736 00:52:52.119 --> 00:52:57.119 be with you, Adam. And that's a wrap for this episode of This 737 00:52:57.280 --> 00:53:00.920 Anthro Life. We've had a fascinating conversation with our today about responsible AI development 738 00:53:00.960 --> 00:53:06.679 and the role of developers in creating ethical and beneficial software products. I want 739 00:53:06.679 --> 00:53:09.320 to extend a huge thank you to our guest today David gray Witter and don 740 00:53:09.440 --> 00:53:14.280 Nefis. Thanks again for sharing your insights and expertise for this. Now let's 741 00:53:14.320 --> 00:53:19.880 recap our top three takeaways from today's episode. First, mondularity in software development 742 00:53:20.000 --> 00:53:22.639 is a double edged sword. It allows for a more efficient creation and deployment 743 00:53:22.679 --> 00:53:30.239 of technical systems, but potentially limits user experience and data privacy. Second, 744 00:53:30.519 --> 00:53:36.039 user research and design must consider the broader stakeholders and people implicated in the deployment 745 00:53:36.159 --> 00:53:40.599 and development of systems, beyond just the paying customer or knowingly engaged users. 746 00:53:42.400 --> 00:53:47.960 And Third, dislocated accountabilities require the reassembly of partial knowledge using a broader supply 747 00:53:49.119 --> 00:53:52.440 chain or value chain framework. And as always, I want to hear from 748 00:53:52.519 --> 00:53:55.960 you. Do you think it's possible to give developers more power to act on 749 00:53:57.039 --> 00:54:01.519 ethical concerns within their organizations and how could collective action and connection between different parts 750 00:54:01.760 --> 00:54:06.920 of a supply chain play a role in this or at What steps could we 751 00:54:07.039 --> 00:54:10.800 take to ensure that the broader stakeholders and people implicated in the deployment development of 752 00:54:10.840 --> 00:54:16.360 algorithmic systems are considered in the research, design development efforts rather than just paying 753 00:54:16.400 --> 00:54:21.679 customers or known users and as individuals, what actions can we take to ensure 754 00:54:21.719 --> 00:54:25.159 responsible AI development? Can we also work collectively to address these challenges around AI 755 00:54:25.239 --> 00:54:30.280 bias and accountability. Remember, you can get in touch with me over at 756 00:54:30.480 --> 00:54:34.480 this anthrolife dot org on the contact page or reply to emails if you're on 757 00:54:34.519 --> 00:54:37.480 a newsletter list, and if you don't get emails from us, I highly 758 00:54:37.519 --> 00:54:43.360 recommend you subscribe over at this anthro life substack to get newsletters, blogs and 759 00:54:43.480 --> 00:54:45.119 to be up to date with all the happenings in the anthro curious community. 760 00:54:45.599 --> 00:54:49.719 And lastly, I want to ask you, my dear listeners, to help 761 00:54:50.000 --> 00:54:52.559 spread the word about this anthro Life. If you find something of value in 762 00:54:52.639 --> 00:54:57.079 this podcast or this episode, please share it with someone who needs to hear 763 00:54:57.119 --> 00:55:00.880 it. Your support goes a long way and growing our community once again, 764 00:55:00.920 --> 00:55:02.960 thank you for sharing your time and your energy with me and our guests. 765 00:55:04.320 --> 00:55:07.320 I'm your host Adam Gamwell, and you're listening to this anthrow life. We'll 766 00:55:07.320 --> 00:55:07.039 see you next time.