Designing for Trust: Power, Policing and AI with Professor Martin Holbraad

Welcome back to part two of my conversation with Professor Martin Holbred, a leading anthropologist at University College London and the director of the ethnographic insights lab. In the last episode, we explored Martin's ontological approach to anthropology, not just as a method for studying the world, but as a way to reimagine its conceptual foundations. We talked through his recent collaboration with us on AI and police incident reporting and how ethnographic insight can reshape how we understand and design public safety systems. In part two, we take that conversation further. We pick up where we left off and dive into actor network theory. A foundational framework in anthropology that treats humans and machines not as opposites, but as part of the same distributed networks of actions and meanings. We discuss what it really means to say AI has agency. How systems build trust and how ethnographic methods can help us track the shifting dynamics between people and technology over time. Let's get right into it. So Professor Holbred, nice to have you back again for episode two of our discussion and I think where we left off last time, I was, we jumped right into this notion of actor network theory and we were trying to contrast in a bit with animism and where that conversation ended. And so maybe, maybe if you can take a step back here and you can continue your introductory 101 actor network theory and maybe lay out a bit of the few of the principles behind actor network theory before we dig into it a bit more. - Sure, you call me Professor Holbred. I don't know what to call you back now, my Hessian. I'm gonna stick with my Hessian. - Please call him Martin. - Sure. - So okay, 101, the first thing to say is that I'm not the most competent person to talk about actor network theory since it came out of a particular field of study that's called Science and Technology Studies STS, which is not my, I'm an anthropologist, I'm not an STS scholar, having said that, of course, Bruno Latour, who is the figure who came up with actor network theory is a French polymath, philosopher, sociologist, founder in many ways of at least a particular kind of STS theory that he represented and developed, but also anthropologists, he did field work. He was one of the first people to do field work with scientists. So basically you imagine anthropologists in a kind of colonial image that we've been trying to problematize in our discipline for a number of years now, going to some far off place, far off from the metropolis that is Europe or the United States. I'm studying with indigenous peoples and so on, that's the kind of stereotype, which has been quite harmful, actually, as well as producing fascinating research, of course. Due to its colonial legacy, harmful Bruno Latour went into science labs and studied people in white coasts as the indigenous tribe in inverted commas, right? That he was studying, and that was really innovative in the 1980s and 1990s and 2000s, as he was developing it. So it's really, really interesting. And in some ways, really, that comparison is not a bad one to convey also the center of his theoretical departure, which is to say that he was very, very keen on what he called symmetry, treating the modern, the contemporary, the Western, whatever word you want to use for it, post enlightenment and so on. In the same terms and with the same kind of frameworks as you would treat any other society, that anthropologist treats, which is sometimes identified as non-modern, non-Western, non-literal, all those uses of non-very problematic, of course, in our discipline, right? So that symmetry, and really, the substance of his point was not just methodological, it was epistemological and substantive, in saying that modernity is a narrative and an operation that turns on an activity that he called purification, which is to purify the world into different ontological caps, right? So on the one hand, you have nature, on the other hand, you have culture or society. And it's after splitting the universe into two, in this way things versus people, for example, things would come under nature, people would come under culture and society, et cetera. Then a lot of the kind of epistemological infrastructure of modernity of the post enlightenment world and so on is about making connections between the two, right? So the way that epistemologists, philosophers who deal with science will often imagine scientific endeavor is as an way to align our cultural representations, if you like, our mental cultural representations, with the way the world really is. So if you're like a correspondence theory of truth, right? I say the cat is on the mat, well, look at the mat is a cat on it, if it's on it, the cat's on the mat is a true statement, right? I'm being very simplistic here, right? Now, the tour is saying that is if you like a fiction, it's a sociocultural operation that modernity got. Very good at, it generates lots of kind of artificial problems. It's only if you've separated culture from nature that you ask questions about what counts as truth in my representation of nature as a cultural being, right? If you don't make that separation in the first place, as by the way, lots of our followers would say, people who are not modern, Western, capitalist, and so on are very prone to do, right? So for example, you mention animism, that's a classic situation in which nature is imbued with human characteristics. There's no separation between nature and culture. What Bruno Latour is saying in his symmetrical argument is that actually if you look at the doing of science, it is actually a series of implications between what the epistemologies will purify as a nature versus culture. So what you have in the lab, what you have in the papers that get written on the back of it, what you have in the conferences where those papers are presented and so on, is a multiplicity, a kind of end-dimensional network of hybrids between the things that the purifies would want to ontologically distinguish. They're all bound up together in this massive tangled mess that is the act of networks, really, right? And they're all interacting with each other, right? And scientists are in the business of connecting traces between those. He talks about the circulating references that connect the original earth that you took from a Amazonian rainforest to the little sample box in which you put it. A first kind of intervention that binds together a cultural bit with a natural bit, right? Then you carried it into the lab. You did certain operations with it. You mixed it with other substances and procedures and so on, right? Each of these little tiny infinitesimal or almost steps produces a line of reference from that thing that you simplistically call nature out there to this thing here that is society and culture in which that makes you get to process and present it, right? And the job of the anthropologists is to track these trackings, if you like, and that scientists themselves get up to and to show the multiple implications between categories that we like to imagine ideologically as distinct, right? So basically, act and network theory is exactly that. It's a theory in which all of these intricate constituents of reality are interacting with each other and a crucial point that he's making, and maybe I'll stop my 101 there, and I don't know how my STS colleagues will tell me if I did a good job or not, is that because you've got this intra-acting network of things, right? It's very difficult and, in fact, wrong to purify when it comes to the question of agency, right? So habitually, in the modern kind of constitution, you would identify agency with the society and culture side of the divide, right? Things in themselves don't have agency. It's what we do as humans that, you know, we're the ones who have agency, right? If you adopt an act and network theory approach to things, agencies distributed across the chain, right? And for example, in your intervention in urban planning, when you want to slow cars down from going from one neighborhood to another, we talk about the sleeping policeman, I don't know if you have that expression in the United States, right? It's the kind of hump that the car has to go over in order to keep its speed low, right? That is the distribution of the agency of a planner in material form. And the fact that we use this kind of anthropomorphizing language, at least in the UK version of English, to call it a sleeping policeman, it's rather a nice metaphor, is a kind of cipher of the fact that we take this thing to have agency. It's stopping us from going as fast as in an ill-advised moment we might think we should be going without a car, right? So we're interacting with a piece of concrete or asphalt or whatever it's made of, but we're interacting it with it in terms that distribute agency also to it. And to say that, oh, no, no, it only has agency by virtue of the fact that some planner decided to, but that's a kind of summary of the situation that is inaccurate to the way the world actually is as far as we know it's always concerned. I find it very fascinating in part because when we refer to things like personal assistance, AI assistance specifically. In our minds, I were visualizing an animate entity, almost something that is human-like and increasingly with generative AI that interaction tends to be very human-like. But the reality is that it is not human and the definition of what an agent is and the agency that that agent possesses tends to be very different depending upon who you talk to and what they're trying to do with it. An actor network theory, at least in that concept, doesn't require you to have this homogenous definition of what actors look like in this network. And that was truly fascinating for me. And so now, if we almost think of these AI agents and without trying to impose some sort of human like qualities on top of that agent, we are now talking about a combination of an agent and AI agent plus a human that is perhaps doing a set of things. Perhaps it's a network of humans and a network of agents that are interacting. And so in this notion of distributed agency, as you called it, does actor network theory actually allow us to think about combinations of these tools plus humans, almost as an entity? I think there's a term for it, "actant." Yes. And how is the notion of an actant different from a typical actor in anthropological studies? That is a brilliant question. If I can just start with the last thing to clarify because I made these distinctions just in my previous response, I would say the simplest way to say it is that actant, the neologism of actant, is necessary when you make this flattening symmetrical move that Latura makes of admitting that agency can be distributing, right? That's why you need a new word for it. Because once you talk about actors or you need agents, you're immediately setting up the connotations of a human as opposed to an inanimate object or whatever it might be, right? So from the point of view of, if you like the ontology of modernity, to blur that distinction is the signature of the non-modern, the animus, the superstition. Racism and colonialism come back in, right? When you start using these terms. So we need a new language in order to be able to overcome these ontological prejudices, if you'd like. One point that, so that's regarding the terminology. One point that I was going to make in relation to the earlier part of your question is how, I mean, Bruno Latura, I met him a few times, he was a very, very interesting person to talk to, I had to say, he sadly died a few years ago of cancer. He was really a leading light of the late 20th century, early 21st century, in global, you know, really up there. Had he lived, I think, I mean, what do I know? But I do think that he would be so, so excited and interested about the advent of AI, because in a way, the world in its concreteness has caught up with Bruno Latura. You guys, and I'm putting you in the category of people who are coming up with these tools and these ways of thinking and so on, right? The engineers, the computer scientists, the people who have precipitated this revolution, right? Are people who are conservatively messing with a human non-human distinction, right? You've created what in many ways could be described as honorary humans, some people would call it meta-humans and so on, right? It is absolutely fascinating for those of us who are kind of dying the world traditional anthropologists like myself to find out as, you know, colleagues of mine have been telling me recently that in many parts of the world, AI tools now are used as divinetry or shamanic technologies, right? So spirits in a place like Mongolia or, you know, different parts of South America or, you know, different parts of Russia. And so spirits that are traditionally used as consultants to give you advice as to whom you should marry or whether you should engage in a particular business venture or whatever, you know, many parts of the world have these practices, right? Hard press to think of any parts of the world that don't have them, right? AI is now grist to the mill of divination. It is the new way of doing divination and it adds Qdos to it because it's not just tradition now, it's tradition plus technology. So exactly the kind of stuff that Bruno Latour was transposing onto the study of science, these hybrids of nature culture with a little hyphen in between, right? So it's absolutely fascinating how in a way that the planet has caught up with the mind. I don't want a big Latour that much up. He wasn't a prophet or maybe in some sense he was, but certainly he was creating the ontological and epistemological frameworks that are most acutely suited to the AI revolution that we're living in today, right? Because precisely the distribution of agency and the hybrid correlation between person plus machine which goes all the way back now evolutionary past to the ape-like being, you know, taking a stick to catch, you know, a piece of food and bring it closer to them, right? This kind of marrying together of the human mind and body with an object that lies beyond it to create an open circuit, but a circuit, nevertheless, which involves technology and humanity imbricated with each other, right? This is the fundamental unit of analysis of the acta-network theory and that is the unit that AI is creating, right? So how to think through the distribution of agency, the distribution of responsibility, the distribution of creativity, the idea of intelligence, of course, it's in the name, artificial intelligence and so on, is basically a passing out in a Latourian universe. And I think a lot, if I may add one more point, a lot of the kind of heat and if not light of the kind of public debate around this is probably, I'm not an expert in this, but I'm kind of hazarding a kind of opinion here, is probably a function of the fact that our epistemological framework hasn't caught up with the objects that it's now called upon to deal with. I'm not saying Latour is the only answer, but certainly Latour was staking out a different way of thinking that is much, much better attuned to these problems than, you know, I don't know if I can refer to our previous conversation, but I was talking about the kind of gun debate in the United States, you know, is that the person is at the gun and so on. It's a kind of, there's no way out of this dilemma because it's badly formulated. And that's why you're endlessly kind of in this circle, you know, gun lobby or gun supporters versus gun critics and so on. And they're endlessly misunderstanding each other because they're framing the problem in the wrong terms. That would be Latour's argument. And I think a lot of that can be transposed over to what he said about AI today. And this kind of inordinate to somewhat kind of social science fiction, kind of dystopia, stroke utopia of, are these machines more intelligent than us, will they run the world better than us and so on? I think a lot of that stuff is just conducted on the wrong premise, I am so certain. - And, you know, one of the elements of actor network theories is notion of symmetry. - Yes. - And last time around we talked about trust, trust between AI and humans in their interaction. How does the symmetry, especially when symmetry conditioned on the fact that you do not need this homogenous capability set within each actor in this network? How does the notion of trust perhaps change in actor network theory? - I can only be completely speculative in this, but certainly, the obvious point to me is that if trust is imagined as a peculiarly or distinctively human concern, that is whereby it's a category mistake to ask of a table or a computer or a pacemaker, whether it's trustworthy or not, then you're doing classic constitution of modernity in LaTour's term thinking, right? So in that way, the reason why LaTour and anthropologists have some affinities because we're all symmetrical in this way, right? Is to say, no, no, no, no, no, no. Like, if you're seeing things from the point of view of the people involved rather than from the point of view of the epistemological legislators, trust can entirely be a characteristic of what, a kind of purifying epistemology would call a mere animate object, or indeed a piece of code, or indeed a piece of legislation for that matter. People are not precious and ontologically, self-disciplining in this particular way, right? So there's another famous example of an anthropologist who's quite kind of confident with what LaTour talks about, a guy called Alfred Gell, who talks about animistic response that anyone has towards their car when it's not working, where you kick the car, it's like, what's wrong with you? Can you work, right? So, you know, in our day to day lives, from that kind of very frivolous example, to very elaborate cosmologically and politically charged, we're ascribing agency responsibility and, indeed, trust, or, you know, being the object of trust of mischief, trust, two things that are inanimate all the time, right? So the job of the anthropologist and the STS kind of scholar or act of network theorist is to take that seriously and look at the detail. I mean, one of the lessons that I hope, I mean, I wouldn't be presumptuous enough to dictate what people take from our previous conversations, but one of the kind of things that we're really big on in anthropologies work from the specifics. Don't work from first principles, because the social life is not conducted according to first principles, and if you think that you will arrive at first principles, well, there's another category mistake. Now, that's not how social and cultural life operate. They don't operate from a set of principles that then get applied to instances. It's a word that I think is really important in this context, which might be useful, and I think it has some currency also beyond social scientists abduction, right? So you work neither deductively nor just inductively, we tend to work abductively in our daily lives, right? So we come up with the best understanding, given the facts that we're presented with, right? And we're constantly making these reciprocal adjustments, which are abducted in their nature. So we constantly abduce, if that's a word, agency, a responsibility, and trust in situations which we ought not to, from the point of view, of the ontological, or it's logical legislators, right? I think that companies, such as Motorola Solutions, and people who are coming up with these products, already understand this because I see it all the time in their comms campaigns and in their advertising. The fact that almost all of the AI icons that I've come across have some magical quality to them, got star dust, they're all anthropomorphizing in different ways, right? Your copilot, your this, your that, right? These are all things that are obviously created to create a sense of trust, a sense of relationship, a sense, all of the social things that we talk about are very actively being constructed and created in the way that these companies are developing their products. And it's no surprise because that's exactly what a relationship with any tool looks like, right? People have feelings for their tools, you know? Ask a carpenter famously, you know, that famous philosophical example, for how they feel about their hammers and their tools, and so on, or a snooker player, or a baseball player, how they feel about their first bat, you know, all of these things are charged, socially, emotionally, culturally and so on. So it's no surprise that that should be happening in these technological tools as well. - It's quite fascinating to think about this in terms of how AI systems today, almost like tools, change the way we behave and be interact. It changes the way we perhaps interact with others because we have access to this tool. Are there ways in which the ethnographic approach that sits behind user experience research for us needs to change in the context of viewing these symmetric actors, AI being on the other side of that symmetry, is changing the way we behave as opposed to the user, the individual, but should we really be thinking about this in the context of a network that is responding or doing something? - Yes, I think that makes absolutely perfect sense. Again, I will repeat that good ethnography in a way has always been like that because you always, I mean, one of the signatures of ethnographic research is its wholism. You're always looking at the connections between things. You never, there's no use taking your side and say, "Mahesh, how do you feel about using XAI products?" It's been good for you, it's just so superficial. I've got to spend time with you doing it and I've got to see you in your family context in your co-workers and so on. See how this is playing out in your office, like what's happening, and what the machines are doing to use that kind of old language is just as important as what the people are doing, right? And everything else that is going on, the food, everything else that is going on around it, right? It's relationship to time, it's relationship to space, and so all of this needs to be taken into account, right? We have to lean into ethnography in this context, rather than imagine that because we're dealing with technological devices and so on, we should somehow be compromising on the ethnographic and bringing more computational, more, you know, not that I have anything against computational methods and it'd be absolutely fascinating to marry these things, right? But I think that's done by leaning in, by going more intensely into the ethnographic sensibility rather than stepping back from it. - So if maybe if I take us back to the research we did together on report writing, and we talked about this notion of the ghost writer approach versus a slightly different take on that, that is more reliant on the agency of the individual who's performing the action and helping them enter this virtuous cycle of self-improvement. In the context of actor network theory, are there concepts there that apply to how we think or frame that problem, where this notion of a ghost writer is in fact something that may be, the framework indicates that there's something wrong with it. - That's a really interesting question, a really interesting question because it strikes me, I mean, I hadn't thought about it before until you just said it now, but it strikes me that the ghost writer model is actually an example of, I was talking about failure of imagination last time, and I think I was kind of writing it now, and now I know why I was writing it, because it's this anthropomorphite as if the best way to accept that now we've got these machines that do a lot of the stuff that humans can do, is to say that they're honorary humans, right? So the stuff we already do as humans just gets projected onto the machines, and that's really dangerous, right? For all the reasons that we said before, 'cause humans need to carry on doing the stuff that humans do, such as being good police officers, we don't want a world in which we don't at least, at this point in time, we don't have a world in which we can do away with humans being in charge of policing, right? And I don't frankly want to imagine one, I think George Orwell did a pretty good job of imagining what that world might look like, right? So the kind of ghost writer model is a one-to-one projection of the human onto the machine, right? Instead of you having to write those boring reports, goes the argument, okay? And AI told to do it for you, right? So the way that we ended up reimagining that relationship, which is to use the current language and enhancements rather than a substitution of the person, right? That's the kind of very current way of framing a virtuous kind of use of AI at the moment as an enhancement and so on, right? It's also a redistribution, if you like, between the human and the machine, right? It's not a straight substitution. It's an augmentation of aspects that the human can do, but the machine can help them do better. So for example, the cognitive interview idea, which I mentioned last time, which comes from psychology and has a history to it, right? That's the kind of thing that in the human situation that we call police work is not really feasible, but a machine can bring that in in a very kind of efficient cost-effective and embedded way. So the machine and the human now are not just playing a zero-sum game, they're making the pie bigger. There's more stuff that can be done, right? So I'm not sure if that's entirely, you know, following the act of network premises or if there's more to it than that, but it's certainly consistent with a kind of loosening up of the ontological distinctions that you're making, being an important step in imagining new possibilities for what AI could do. I mean, I imagine a virtuous imagination of AI as being a kind of apex between social science imagination or what is humanly possible and engineering imagination of what is machinically possible, right? And if those kind of series were to be made to converge and interact with each other in a kind of symmetrical way, as we were saying before, I think that's possibly the way forward for imagining AI tools that are truly operative in the way that we want them to be, rather than, which is a kind of dystopian image, this constant fear of replacement, that they're going to just take stuff that we cherish, offer us and do it better, and they're going to be a threat and so on, you know. - And I think one of the things you just mentioned, which I think is quite interesting is this notion of redistribution of responsibilities or agency across this network, you have the combination of a tool that's helping you write reports or helping you create better reports in a way where it's not a ghost writer, but there are also people downstream who are consuming those reports, whether they are your supervisors, prosecutors, defense attorneys, investigators, et cetera. There's a redistribution, it seems, of agency that happens in that whole network at this point, given what we are able to do or could do with the right imagination with AI. As you were thinking about, and as you spoke to many of the supervisors who were mentoring young officers, did you see a possibility there for perhaps even redistributing some of the responsibilities there? - That was not the brief of the project, so we didn't look out for it. I think certainly, I mean, I can say very superficially that there were two kinds of police officers that that is itself a kind of simplification, of course, but we were in our conversations with my colleagues when we were doing the project, we ended up asking about any chinstance, was it that? Was it one kind of, was it the other? There were the kind of tech and a foe, kind of somewhat skeptical approach that basically all of this stuff is really scary and dangerous and so on. And there were actually many, many people, including people of my own, that 51, my own age and above, so not kind of millennial people who are all o-fei or younger with technology, who are actually really, really positive and said, you know, we really need help in this area, right? So there was an openness to that. Of course, you know, as far as I know, none of these people are reading Bruno Latour, so I'm not sure if they are thinking through to the possibility that responsibility itself might be redistributed as opposed to leaving the coordinates of who it does and doesn't have responsibility intact and then deciding whether we're going to include machines or not and so on. I mean, I think that shifting of the coordinates of what can count as a responsible actant to use that word in this context is something that is work to be done. It's not, it's not something, I think, I think imagination hasn't got there yet because the problem hasn't been formulated in these terms. I think we really are talking, I mean, this conversation right now, at least to me and maybe it's my own ignorance, but it is at the forefront in terms of, you know, framing, you know, ontologically to use that word again, you know, what kind of entities there exist here and what flows and follows from that. I think that conversation is really added at its inception at the moment. I think, you know, if I can just make a comment, like I think the discussion which, you know, I'm not closely acquainted or don't follow very closely but nevertheless I'm very aware of about the question of intelligence and how we define intelligence. It's absolutely fascinating and of course, you know, the field they for philosophers and so on. But I think there's so much more to this and the question that you were asking now and the fact that you guys are developing tools in a space where, you know, responsibility, moral, peril, life and death situations are at stake, you know, shifts the, you know, the framework of the conversation away from just this kind of somewhat, sometimes sterile conversations, you know, not quite sure what really hangs from it of deciding whether these tools are actually intelligent or not and so on, right? I think there's a kind of shift when you're working with emergency services police that that happens, that makes this conversation rather more interesting, I think. Sorry, I fudged your question because I don't actually know what the answer is. I think all I can say is that it's an extremely important and interesting question to be asking, how might we redistribute and redefine, not just redistribute, redefine and redistribute, the question of responsibility and trust to go back to your earlier question, vis-à-vis this new configuration of machine people hybrid. - And partly, my question was motivated by the fact that you can think about a tool and a tool in its very simplistic form will do the same thing if you do perform the exact same action or give it the same input every single time. And as time goes on, maybe there are certain physical properties of that tool that might change, which may change its behavior in some form. But for the most part, it is rather deterministic. It obeys certain laws of physics or behavior. Software in many ways has been that for many, many, many, many years. And I almost think of intelligence in this context, perhaps naively so, intelligence in this context is something that is designed to adapt, that is designed to change where it is not as deterministic as software once was. And perhaps that change is one where it personalizes itself to you. It tries to understand you, your preferences, your tone, how you speak, what type of language would indicate sarcasm versus what would indicate seriousness and fact. But ultimately, you're dealing with an actor in this network who is changing, who is not static. And by virtue of that, the relationship between two or more actors in this network are also going to change. And by the way, we as humans do this all the time where I adapt to what you're saying. My response to you tomorrow may be a bit different from what it is today based upon other contexts that I have. And so this actor network, if we continue with this framework, this actor network is constantly evolving. It is constantly changing. And to a certain extent, as engineers and designers, as us adding elements into this network, this actor network, specifically to make the humans in this network more effective, more capable, to be able to do their jobs better, it is to say as this AI actor or this agent helps this human and continues to adapt, what is the equilibrium state, or what is even that temporary equilibrium state where we can make this network do the right sets of things or to do something that is better. And so this is in part a question and in part a statement. But what I've been struggling with and really also trying to understand here is, as we think of these AI agents as almost perturbing this actor network, perhaps the base network is just a network of humans who are using some deterministic tools, very predictable. But at the end of the day, perhaps even humans who are trained to mechanically do certain things, but really why it's a human centric task is because they're able to apply a certain level of intelligence, human intelligence in this context. These AI elements now affect the equilibrium of that human network, hopefully positively. And how should we be thinking about behavior, specifically the change in behavior, so that the direction of that change can be positive, the direction of that change could be something that yields the right sort of efficacy, efficiency. The notion of actor network theory seems like, at least qualitatively, it allows you to think about those concepts. Yeah. But qualitatively, how to sort of frame that concept is something that I'm still trying to think about. Yeah. So one thing that I would-- I mean, there's so many things in the question and how long you took to formulate it is just the function of the richness of all of the issues that you're thinking about in here. So it would probably need another hour to talk about all of the aspects of the things that you raised. But one point that occurred to me would be relevant is let's take the generativity of AI. So the fact that it's non-reproducible, it gives you a different result every time you're asking and so on, right? That's what I take on in a non-engineer but a very crude way. Let's say that's the generativity, right? Now, funnily enough, a long time ago with a couple of colleagues I wrote a paper called, "Technologies of the Imagination," way before AI was even a dream in anyone's mind or whatever, anyway. My point is that the way we define imagination here was precisely that was the-- or "Technologies of the Imagination" are technologies which precipitate results which are under-determined by their causes, which sounds like a contradiction. But of course, that is exactly the contradiction this genus sequa that we associate with imagination is that even though you can pinpoint the conditions that brought it about, you cannot tell an exhaustive story of how those conditions led to the result that they led to. So in that sense, the human creativity, human imagination, is generative in a sense that is at least analogous, if not identical to the sense in which we use that word when we talk about AI, right? And I think I mentioned that in our conversation previously when I talked about anthropologists being attentive, not only to the conventions of social and cultural life, but also the invention of social and cultural life. And the great figure there would be a Royal Wagner and American anthropologist who wrote this great book, Prophetic Book, in the early '70s called "The Invention of Culture," which is all about that, right? Imagine culture as a jazz solo. You don't know what John Coltrane's going to come up with tonight, right? But you know, it's going to be good because it's John Coltrane, but it's every night, it's going to be different, right? So imagine if human culture was to be thought of in these terms, rather than, you know, kinship systems and traditions and rules and, you know, blah, blah, blah, right? No, we're in the process of doing jazz solos all the time, even though we don't always admit it. Also, in that sense, there's a symmetry here. AI is doing that bit, which was so quintessentially human, wasn't it? Like, we thought, we're the imaginative ones, right? Homosapiensapiens is the one that does imagination. It's not the other species, right? And now we've got these machines. Now, let me add that to that thought, the point that I made before about the specificity of the ethnographic method, right? Ethnography does not promise you the deduction of, you know, the underlying framework and principle that explains everything in human behavior, right? That's a misplaced aspiration for ethnography, right? Ethnography treats each situation as it comes, tries to work out the assumptions that underlie this situation and shows the way in which this situation is surprising, visibly, the assumptions that you made about it. So you made an assumption about what AI is and what office reporting is. We went into the situation, we spoke to the police officers, we showed that their assumptions were actually in some distance from the assumptions that you were making and that precipitated a shift in your own thinking, right? It was through an engagement with the specificity of police work that that was possible, right? Now, if we're saying that because human beings are inventive, imaginative, creative, never produced the same answer to the same question and so on, they require a methodology for their study which is embraces the specific and the contingent and doesn't try to explain it away. Well, hey presto, the same holds for our study of AI, or a generative AI, right? We need to have an ethnographically minded engagement with these tools. We need to orient ourselves towards the contingency of the results that they produce, the specificities, right? Rather than lament the fact that these specificities are black boxed and we can't unpick them, embrace it and say, well, that's precisely what we do with humans. So let's do it with the machines and indeed, let's do it with the network, the act of network of human plus machine or human hyphen machine. You've got that kind of compounded potential for creativity there. So let's engage with it in this ethnographically spirited method of embracing the specific and tracking it in its development. So, I mean, if I can again be very speculative 'cause I think your conversations and your questions have been very kind of inviting of speculation on my part maybe because of my sheer kind of poverty of the my knowledge base. But it seems to me that we may end up in a situation where we're leaning more and more towards, if not in, but towards. Navigating this situation in terms of a constant protocol of monitoring and constantly a kind of dynamic system of monitoring, or I'll say, protocol rather than system because system has a very kind of static connotation to it. A dynamic, recursive protocol of looking at the generative aspects so much of the human and of the machine and of their combination as they develop and this, you know, towards this unknown horizon that we're all kind of venturing towards. As we always have with human phenomena, we've always been venturing in an unknown horizon. Well, when we dealt with machines, we had this comfort zone that if we can go under the bonnet, we'll know what's gonna happen. Well, now we don't. Well, let's transpose them. All of these ways in which we've been very comfortably dealing with imagination, creativity, unpredictability and so on in the human domain, let's now transpose it to the machine as well. That's the lesson that I would learn. So I would learn, I would imagine that the way in which we're going to navigate the future in this environment is going to be an embrace of specificity in an iterative, recursive, always open to transformation way, which is not static and legislative in its aesthetic. It's kind of dynamic and operative, if you like, right? It's recursive in that kind of cybernetic way, right? The all cyberneticist had exactly this idea that you're constantly revising the algorithm in view of the output, right? This recursive circle is constantly operating and that I think has to operate in our moral, political, epistemological navigation of this space just as much as it operates in the technological domain. In our previous conversation, you drew a distinction between what you commonly think of as culture. And to me, there's a lot of structure to how you design an actor network or form an actor network, almost specifically, to answer a very specific problem, the specificity question. Is there a link there? Is there a link between the notion of structure and the notion of designing actors who can influence this actor network to create that structure the right way? So there are interactions that we design from a user experience standpoint. There are certain interactions that we design. There are certain sources of knowledge that we may feed into an AI, a databases that it can tap into. Certain people may have access to certain databases, certain people may not have access to certain types of data. It may be that something captures health care data and that health care data is not accessible by everyone who is perhaps helping an individual. So there is access to information as one element of structure. And so as I think about it as just like you talked about infrastructure, infrastructure, roadways, timings of buses, et cetera, helping one but not other. In this case, with specificity in mind, and we can even go back to the example of authoring officer narratives reports. There's structure there. There is access to certain information. There could be structure in terms of the process that the officer leverages, perhaps it's this notion of a better interview process, this cognitive interview process, et cetera. I'm trying to see whether there's a mapping between that notion of a structure but in this context of this network that we're talking about because that network has both an emergent structure and a base structure that exists, which is maybe the access to data. Yeah. Maybe a point that might be just worth making is so often in a kind of anthropological context, appeals to structure, connot, as I said, costability. And one has to always think about the ways in which structures are subverted or resisted and all of the things that so a famous anthropologist to use and not the same word, but in a reference to a completely different person turner again, but in this case, a Victor turner. He's spent his whole career in some ways developing theories of anti-structure. You may know the word liminal. That's Victor turners word, right? So the things that happen in the threshold of structure rather than embedded in it and how, what we were talking about before, invention, creativity, and so on, depends on this interplay between structure and anti-structure, that happens. And ritual for him was the space par excellence where anti-structures are able to be made visible and socially operative in different ways, right? One of the things that hung on this for turner was the association between structure and not only the static, the organizational, the thing that doesn't change when other things change, but also hierarchy. So when you talked about access to data, for example, I think one of the big worries there is, of course, the hierarchical distinctions that are getting inscribed by the decisions that you make as to who has access to something and who doesn't, right? I think the kind of open access movement, which doesn't apply, by the way, to medical data, right? But when we're talking about open source and so on, right? Often has a kind of almost millenarian concept in it that if only we all had access to everything, the world would be a more equal place, right? So there is a kind of correlation in the way that we imagine structures in our differential accesses to them and the way that we imagine social hierarchies. So I think that, I'm answering in a kind of general and abstract way, but I think that question of hierarchy is probably going to rear it. It's head very early on in that conversation about structure. And I think, in a way, I mean, here's another way of saying something slightly different, in a way being very tight and strict in the distinction that you make between structure and its opposite operates often as a power move, and I would kind of hazard the further thought on the top of that thought, which is that often in our interaction with technology, what we and our frustration sometimes when we're not happy with the tools that we're presented with by a particular company, is often with a feeling that they have imagined in a structure, they've kind of embedded in the structure of their product, a kind of set of assumptions and presumptions about what might matter to me as a user that I don't recognize and that I feel that I need to press against, so that in a way who gets to decide and what decisions are made with regard to structure in the way that a particular product is organized, is a way of inscribing the kind of hierarchical relationship between the designers and the users and so on. And I think there's a kind of tussle going on there. And in a way, you might want to say that your experience researchers and their, in alliance with designers who work with them, are mediating that tussle in some ways. I think often, I mean, I don't want to say anything that is respectful about engineers, because I know that. I think often engineers think that they know stuff, right? Because they imagine a solution to a problem of their own making. But what they forgot to ask is, is this problem, a problem for the people who are going to buy this product? And that's what the UX people and the designers and indeed the anthropologist, people like myself, are having to mediate. And the point I'm making is that I propose your point of structure, is that this is often about power. And on the concept of power, I think there's some very specific definitions of as to what power means in the framework of actor network theory. And as you were talking there, it occurred to me that there's an imbalance in power when you take a ghost rider approach to authoring a report versus somebody who is conducting a cognitive interview or a tool that's conducting an cognitive interview. So if you go back to the notion of power and you think about the pros and cons of different approaches to AI helping that report authoring process, is power something that is a worthwhile consideration in that process? Yes, I think that power is another idiom in which we can talk about the things that matter to police officers in this instance, and we're talking about this example, or uses more broadly. And I think it's always at stake. And of course, you know, in a kind of law enforcement context, power is wielded in a very significant way, right? We don't pretend to think of technology as very complicated digital stuff, but all everything. And there's a stick that the APU's that we were talking about, but that's technology too. So all of these kind of the way with all that the police officer is invested with through their training, through the equipment that they're given, it's not just the gun, but the report, the uniform. All of that is at a very, very fundamental level about power. And it's about, you know, it's an pertinence, if you like, of the state's power, right? Famous sociology sociologist Max Weber defined the state as the entity that is vested with the power of violence, of legitimate violence, right? That's a kind of central definition of what a state is, that it's a legitimately violent entity, right? And the police alongside other forces, including the armed forces are in the business of wielding that power, right? So when you're in the business as you guys are of producing tools and other enhancements to help in the process of police work, you're also in the business of producing instruments of power, right? So the level to talk about responsibility, the level of responsibility is increased exponentially when you're dealing with these kinds of tools, right? So I think that the question of power is always at stake and I know that Motorola Solutions is very, very cognizant of that because I've looked at the documentation that you sent me and so on. It's very, very front and central in your thinking about the tools that you're developing. And again, I would say that all of the things that we discussed before apply to this too, you know, the potential of the generativity to produce new and unexpected forms of power, which can be wielded in a positive as well as in an abusive way and so on, right? All of these frontiers are open. And to my mind, the only way to address these problems is to deal with them in their specificity and not to imagine that you can predict or cover the bases of the different problems that might emerge when you're dealing with it with a piece of generative technology as compounded by the fact that the people that use it are by the very nature generative to use that word, i.e. imaginative, creative, unpredictable and so on, right? So yeah, the stakes are extremely high in this context. I also think that there's a connection between power and trust in the sense that if I have two actors in this network and the AI actor, and perhaps connected to a human actor, the AI actor, if you attribute a fair amount of power to that AI actor, the level of trust that you have to require there is proportionately significantly greater. So a system where you don't attribute too much power to the AI, but it can still be helpful, perhaps requires a different calibration of power. Again, not to ignore the specificity of the problem that we're trying to solve, but even with between people, I think there is this balance between power and trust. I think that's an extremely, extremely important point that you make. That's exactly right. I mean, there's no way that that could be anything other than right. And I think that is, of course, in political science, political sociology and political anthropology, the question of how under what conditions and to what degree we come to trust political authority is absolutely central, right? Goes all the way back to Plato, and it's Republic, and, you know, forward, you know, what is the legitimate use of violence with which we have in-- what does the legitimacy of the violence in which we've invested-- of the power of violence with which we've invested the state consists, right? What are the conditions of possibility for legitimacy, for legitimate violence, right? Because if it's illegitimate, then, of course, trust breaks down. The question of legitimacy in this context is a question of trust. If I say that it's legitimate for the state to incarcerate people who've committed crimes and so on, right, I'm saying that I trust the state to make that choice. And, yeah, at the moment, I certainly don't trust, you know, Motorola solutions to make that choice, not because I don't trust Motorola solutions, but because you're not vested with that. You don't have that kind of legitimacy, right? So if you produce a tool that starts to make a, you know, to gain agency in the decision-making process as to who should go to prison, I am immediately on my guard because, unlike the democratically elected government of the day, you have no business making those decisions. I'm not talking about you specifically, I'm just, you know. So I think that is, of course, absolutely central to the kind of social, if you like, calculation that technology companies need to make is to understand how does the enormous degree of power that these tools can potentially provide tally up with a thresholds of trust and legitimacy that are socially current in any given situations and socially and politically current. And I think that's another kind of laturian conundrum, right? Because the technological, the physical, the social and the political, how do all of these things get implicated in one big circle of mutual interference, right? Is a kind of act-to-network way of setting up the problem if we have to use those terms, or an ethnographically rich and complex situation that you need to delve into in its contingency in specificity. If you have any chance of making any headway and thinking your way through to a solution to how to distribute these things, right? So in a way, you know, the message, I know we're kind of reaching the end of our conversation. I mean, the message is embrace the specificity. Put the time in and the resource, right? That there's no way of cutting corners into this because it's all better off as to where this is going to go, right? So we have to track it in the way that it's actually going because in its very nature, it's not telling us its own destination. So whether we trust it, whether we invest it with power, what degree of power, who, under what conditions do we configure that human machine relationship and so on, are all open questions that need to be addressed each in its specificity. And what is true for law enforcement may not be true for house maintenance or transport planning. There's no reason to assume that you will arrive at one set of principles that applies across the board. That's not to say that there won't be cross-fertilization that you can't learn lessons, of course you can. But you do it in this abductive, iterative, specific way, rather than a kind of deductive, you know, let's arrive at the fundamental principles that will explain everything kind of way, which I just don't think I just don't think they're fit for purpose for this phenomenon. Martin, this has been a fascinating conversation. So maybe before we end, we took a long journey from what is anthropology to how do we think about AI in the midst of this to a specific example of how do you design an AI-assisted report writing process to actor network theory, notions of bias fairness and trust in the middle of all that the balance between power and trust. So if you were to leave the audience perhaps largely composed of AI engineers, but also perhaps users of AI in various different ways, what are the few bits of advice, sources, resources that they should be thinking about as you go design AI capabilities, AI agents even in this new world where you have this network of people doing very important things but benefiting by these AI entities that may be helping you do it. Wow. I mean, there's going to be a lot that one could say to that. It's a very nice way to end. So something that really is that, to my mind, at the forefront of my own discipline, anthropology, a kind of development theoretically that I've been involved in myself and I kind of really believe is quite a radical kind of move that we need to make as social scientists working with people, which is to count and to the possibility of what we call ontological multiplicity. So there's a very familiar idea. And we talked about the notion of culture earlier in our conversation that one thing can look different from one person's perspective and another person's perspective, right? Cultural relativism is based on this idea. You can look at the world with different tinted glasses and see different things, right? How you can have moral relativism, extinological relativism, and so on. The move that I'm talking about is the move of saying that the plurality that is an issue when you're doing ethnographic work of the kind that we've been talking about in our conversation is the plurality of what things are and that is the domain of ontology, right? In a kind of philosophical context. So things are never one thing. Things are multiple. They change their ontological constitution across time, across space, across social situations, across practical settings in which they're put to use. And being aware of this possibility of an open horizon of ontological possibility is I think really, really important when you're designing products. I think we've seen a really good example of this in the study that we did for Motorola solutions and the incident reporting tools that Motorola's developing. So to give the example that we've already discussed, you go in with the assumption that an incident report is a record as faithful as possible of the officer's subjective recall of what happened in the incident that the report is about. So essentially a report here is taken to be a subjectively driven reflection of the persons or the author's memory. We then do ethnographic research with police officers who actually write these reports or supervisors who read these reports and correct them and call the officer who wrote them up for not asking the right question or whatever the case may be. And what we find is that from their point of view what the report is is actually quite differently configured. The ontological question, what is a report is given a different answer by the police officers who write it. They're saying, ultimately, this is a performance of my professional competence. This is a record of my capacity to convince a jury, a promotions panel, a complaints panel and so on that I do my job well. Yes, it presents the case or the incident that it's about. But as a document, what it is for me, what it is in this context, is not captured by saying that it's a record of my memory. So, and of course, this is not an either or it can be both, it can be ambiguous between the two, it can be more than two things, but this difference, the shift in thinking that is involved when you open yourself up to the possibility of ontological multiplicity is I think absolutely central to coming up with designs that are conceptualized in a way that can gain traction, right? A good example is a colleague of mine who actually was involved in the Motorola Solutions project that we've been talking about, Gemma Totea Proctor, who's a PhD student in our department at UCL in London. She did another project for a different client who, a company that makes laptops. The engineers of that company, of course, assume that laptops are machines for doing spreadsheets, for doing work processing, for doing gaming, for doing the kinds of uses that engineers imagine their laptops for, right? And the whole company is oriented towards providing solutions to those kinds of problems. Now Gemma did ethnographic field work with people using these laptops in their everyday lives, and she tells the story of a woman at home with a laptop turned on and balanced precariously on a kind of coffee table in the living room with a meeting going on at the camera and off and the microphone muted, and the user in this case preparing a meal with a view of picking up the kids from school in a little while and preparing a snack for them, getting some clothes ready, and so on. All the kind of stuff that you do, and this is a very gendered example, you know, this is a woman who's balancing effectively her role as a mother, as a homemaker, and as a professional. And she says to Gemma, this machine here that's precariously balancing on my coffee table is a machine that allows me to live the life that I want to live, right? My capacity to be in, it's a comment, of course, on remote working and all of those things, but it's also a comment on the machine itself, right? She doesn't see the laptop as just, you know, a machine that can run in particular software. She sees it as an integrated constituent of her ongoing moral, personal, social project of balancing different bits of her life against each other and making both of them work or many of them work. The bit of preparing a meal, the bit of attending a meeting, the bit of being on time for picking up the kids from school. These things are possible for her because of this piece of technology, right? So what the laptop is in this situation simply doesn't match up to what the engineer had imagined for it, right? So if you are a person who's working in designing and coming up in an ideation of, you know, possible new tools and applications for AI and so on, I think it's well worth counting and seeing the possibility of ontological multiplicity as I define it, I think that's, that should be a basic principle of design. Makes a lot of sense and that's a, that's a great place to end. Thank you, Martin. Thank you for spending the time. Thank you for spending a fair amount of time in this conversation. Thank you for helping us do the early research on I think what will end up being a great solution for our first responders. So it was a privilege to have a conversation with you. My house has been an absolute pleasure. It's really been fun talking with you and I hope it's of some use to someone. I'm sure it will be. After this conversation, what became clearer to me is that anthropology offers more than just cultural interpretation. It offers a radically different way of thinking about systems frameworks like actor network theory invite us to rethink agency, not as something humans own and delegate, but as something co-produced agency lives in the relationships between tools, practices, people and policies. That changes how we understand AI, not as a ghost in the machine, but as a participant in dynamic shifting networks, where meaning, power and responsibility are constantly negotiated. That's where ethnography becomes vital. Not as passive observation, but as a form of conceptual engineering. It reveals how systems are actually used. It shows how concepts like trust, fairness and memory take shape on the ground. And how those meanings vary across the many subcultures and roles that make up public safety. This isn't just useful, it's essential. Because in the age of AI, we're not just designing for a generic user, we're designing for the multiple ways of being a user. For the detective in the field, the officer behind the desk, the prosecutor preparing a case, even within one role needs an artifact shift moment to moment. A report isn't just a record, it's a memory aid, a legal artifact, a shield, a performance. What a tool is depends on how, when and by whom it's used. This then is Martin's ontological multiplicity in action, overlapping realities within the same system. And those realities are never neutral, they're shaped by power, pressure and purpose. So to design AI that truly serves public safety, we need to design not for abstractions, but for the layered, specific lived experiences of the people inside those networks. Finally, and perhaps most importantly for those of us who seek to be on the cutting edge of innovation, anthropology reminds us that imagination is a method that every system encodes assumptions about the world. And that those assumptions can be questioned, rethought and re-imagined. My thanks again to Martin for such a far-reaching conversation and to you for joining us in exploring this complex space where culture, technology and public safety needs. Hope you enjoyed it, thanks again. [BLANK_AUDIO]