Human Centered

High-tech Modernism

Episode Summary

CASBS research affiliate Henry Farrell and former CASBS fellow Marion Fourcade engage in a roundtable discussion with danah boyd, William Janeway, Charlton McIlwain, and Zeynep Tufekci on creating a moral political economy of high-tech governance.

Episode Notes

Episode Transcription

Narrator: From the Center for Advanced Study in the Behavioral Sciences at Stanford University, this is Human Centered. What are alternative ways of thinking about and constructing a moral-political economy of technology particularly machine learning, artificial intelligence, and algorithmic decision-making. Today on Human-Centered, another episode in our Social Science for a World in Crisis series. This episode, which originally webcast November 8th, 2021, is titled High-Tech Modernism, and it shares the focus with CASBS's ongoing program on creating a new moral-political economy. The featured panelists are Dana Boyd, partner researcher at Microsoft Research, as well as founder and president of the Board of Data and Society. Marion Fourcade, professor of sociology and director of the Social Science Matrix, both at UC Berkeley, and she was also a CASBS fellow from 2008 to 2009. William Janeway, special limited partner at Warburg Pincus and affiliated with the Faculty of Economics at Cambridge University. Charlton McIlwain, vice provost for faculty engagement and development, as well as professor of media, culture, and communications at NYU. Zeynep Tufekci, the McCall Term Associate Professor in the School of Information at the University of North Carolina and a Faculty Associate at the Berkman Klein Center for Internet and Society at Harvard. Kicking off the conversation is Henry Farrell, the SNF Agora Institute Professor at Johns Hopkins University and a 2021-22 CASBS Research Affiliate. The starting point for the group's conversation is a working paper co-authored by Farrell and Foucault which was inspired by the classic James C. Scott book Seeing Like a State. The two take Scott's conception of bureaucracies as classification systems and apply that framework to machine learning and algorithms. They argue that thinking of them as engines of classification may help us create a new moral-political economy of technology. Their goal was to stimulate debate and constructive disagreement among the panelists, as together they imagined alternatives to the current regime of surveillance capitalism. A quick heads up before we jump in: if you want to explore the topics in this episode further, you can find links to relevant materials and bios in the episode notes. Without further ado, join Human-Centered as we listen to the CASBS event, High-Tech Modernism.

Henry Farrell: Hi, I'm Henry Farrell, and I'm a professor at Johns Hopkins University, where I'm jointly affiliated to the Stavros Niarchos Foundation Institute Agora Institute, and to Johns Hopkins School for Advanced International Studies. And I'm here today to host the 18th episode of CASBS's webcast series, Social Science for a World of Crisis. And I want to talk about the topic for today's, for today's webcast, which is high-tech modernism. But before I want to do that, I want to do two things. First of all, I want to acknowledge the partners for this episode who are Data and Society, the Ethics Society and Technology Hub at Stanford, the Institute for New Economic Thinking and the Social Science Matrix at University of California at Berkeley. And secondly, I want to talk a little bit about the panelists. We have some wonderful panelists today, and you ought, of course, to have read the detailed biographies which were in the event promo. You may or may not have done that, depending. If you haven't done that yet, you should go to the chat box where we will link to the event promo again so that you can see in great detail what all of these panelists have done and it is a varied and wonderful list— because I'm not going to list that. I'm going to just give you a brief sense of who the people are and why it is that they have important things to say about the topic of today's discussion. So first, I'd like to begin with Dana Boyd. Dana is at Microsoft Research, and she is also the founder of Data Society, which she is still involved in via the board. And she has been thinking about data and the algorithmic society for a long, long time and has been doing wonderful and extraordinary work in this and is also in the throes of writing a book on the US Census, on the politics and social questions surrounding the census, which turn out to be incredibly intricate and important, which many of us are waiting for with bated breath. Marion Fourcade— I should say myself particularly among these. Marion Fourcade is a professor of sociology and also the director of the Social Science Matrix at the University of California at Berkeley. And as I see, the core of much of her work is the— I see it as being the complex relationship between, on the one hand, economic categories, and on the other hand, the way in which markets actually function as opposed to the idealized ways in which we often think that they do, and how it is that categorization processes and markets intersect in order to shape different classes and to shape the real-life opportunities that are open to individuals. So I think that this is something that comes through in a lot of her work and which clearly connects in very important ways to the discussions and arguments that we're going to have today. Bill Janeway is at Warburg Pincus and is also at the Faculty of Economics at Cambridge University, where he has recently been involved in the launch of the Janeway Institute. I welcome you to go and to look at their webpage. They are going to be doing many wonderful and interesting things studying the economy. And he also has been thinking on his, on his own behalf for a very long time about the intersection of three forces, effectively, of political, economic and financial forces and how they shape the long-term evolution of capitalism. Charlton McIlwain is a vice provost at New York University and is also the professor of media, culture, and communications there. And his recent book, Black Software, does two important things. First of all, it identifies an important vanguard of Black entrepreneurs, Black thinkers, Black doers who sought to try and take the new technologies of the internet and to shape them in ways which sort of were more reflective of their perspectives and of the broader cultural perspectives of America's Black community. And on the other hand, he also looks at how it is that we saw the advent of new algorithmic techniques, new forms of data specification, of sort of data analysis, which on the other hand sort of had crucial consequences, for example, for policing and also for the racial politics of the United States very often in a quite negative way for Black communities. Zeynep Tufekci is in the School of Information at the University of North Carolina. Her book, Twitter and Tear Gas, was one of the first really systematic efforts to understand how it is that social media and politics intersect with each other. It's a book that all of us have read and have learned enormously from. Those of you who have been following her over the last 18 months or so will know very well that she's been engaged in a very different set of questions, more or less she found herself propelled into a new and perhaps somewhat unexpected role as somebody who was effectively building an interface between academic scientific and social scientific debates about the coronavirus pandemic and the public conversations and arguments that were being waged, and so very often in less informed ways. And she has been somebody who has been at the forefront of trying to bring the two closer together. Although I think from reading her Twitter feed over the last few few weeks that she is very much looking forward to getting back to her core research after a long period of, you know, sort of what has to be only described as heroic and indeed Herculean public service. So these are a wonderful set of panelists who I think are going to come at these questions from a broad set of perspectives and sort of with a different, with very different criticisms, very different understandings of the way forward. And I hope that the conversation that we have reflects this multiplicity of viewpoints. And so what we're talking about today here is high-tech modernism. And those of you who've read the paper that is again linked from the event promotional page will have a pretty good idea of what this involves. For those of you who haven't, the 30-second version is as follows. It builds on an idea from James Scott who wrote a famous among social scientists book a number of years ago called Seeing Like a State, which looked at the ways in which bureaucracies in the 19th and 20th century constructed categories that they sought to use to try try and understand the world, these abstract categories that strained out a lot of real-life knowledge, and in turn sought to impose these categories in their policies and in ways that they sought to try and make the world more visible and more understandable to them. And this resulted very often, according to Scott, in pretty enormous policy tragedies. And so what the idea of high-tech modernism is trying to do is trying to understand the relationship between this world of 19th and 20th century bureaucracies and the 21st century world that we are entering into where algorithms and data play a crucial role in governance decisions, both in market and in the state. And the rough intuition here is that these processes actually have a lot in common with the processes that Scott identified when he wrote Seeing Like a State under the label of high modernism, but they also have some very, very crucial differences. And so one of the topics we'll be discussing today is the relationship between high modernism and high-tech modernism and what kinds of things we can possibly learn from this comparison. So with this, I'm going to pass over to the speakers in just a moment. Each of the speakers is going to talk for a few minutes about some major points that they want to develop with respect to high-tech modernism. After each of the speakers has finished, we're then going to move to a much more conversational format in which I will try as moderator to draw out as best as I can the conversation conversation between the different perspectives that have been advanced. And I will try, as I am doing this, to weave in as best as I can comments and questions from people in the audience for this event. And so if you want to participate, you should be able to find in your Zoom, you can use the questions and answers function in order to ask questions. What I would say is try to keep the questions as succinct and precise as possible. I will try to weave them in as best as I can to the debate and to the discussion, asking the panelists. I will say that depending on the volume of questions, I may not be able to get to all of them. But if I don't get your question, please do feel assured that the panelists will have an opportunity to read it and reflect upon it afterwards. And so with that, I am going to turn over to Marion, after which we are going to go in alphabetical order through the participants in the panel. And I hope you enjoy this. This is going to be a great conversation.

Marion Fourcade: Thank you very much, Henry. And as Henry just mentioned, he and I wrote one of the think pieces that's posted on the website and that actually provides the title for today's event. So we are very grateful for everyone's engagement today. What I'll do is next, I'll sketch out the core of our argument for, for the audience. So algorithms, and especially machine learning algorithms, are among the most important technologies governing our current social order. To use Mary Douglas's phrase, they do the classifying. That is, they sort and they slot. What are the social implications of this simple fact? Well, many people, particularly technologists themselves, want a prescriptive recipe. How can we make sure that these tools will do no harm? And so there's a lively literature in philosophy, computer science, and law about how we can regulate and audit algorithms, or how we can bend their design to our ethical or political will. But the question from the social sciences is often different. They will ask, how do algorithms concretely govern? How do they compare to previous modes of government? And importantly, how does their mediation shape our moral intuitions and our political possibilities? So our piece, as Henry mentioned, is inspired by James Scott's book Seeing Like a State, a critique of what Scott calls high modernist bureaucracies. High modernist institutions and expert systems organize the world in ways that make it legible and prepare it for intervention. In the process, Scott says, they abstract away a lot of useful information. They function in an autocratic, top-down manner, trampling on local knowledge and lived experience. But the very visibility of their actions means that those so governed often respond to the way they are seen. Sometimes they conform to these ways, these categories, sometimes they grow into them, and sometimes they loudly contest them. Now algorithms have the same fundamental goal: to organize the world so it can be governed. In other words, they are a form of bureaucracy which we call high-tech modernism. But they govern differently. Instead of stripping away valuable information, they swallow up everything they can, that is, everything that can be made available through digital means. More information, if we follow Scott, ought to be a good thing. The problem is the categories that result from the new optimization processes are different. You know, they are emergent, they are bespoke, they are often multidimensional and dynamic. They change over time. You know, they are constantly changed through feedback processes. For the governed, of course, this is politically baffling. They cannot penetrate the mechanisms that manage their lives. Not because they are secret, which they obviously are, but because they cannot be made intelligible. Nonetheless, the governed must orient themselves to these mechanisms because their life chances depend on them. You know, you can think of the ability to cross a border, to obtain credit, to be visible on social media. This produces a very particular kind of moral economy. People are incited to make themselves fit for various kinds of sorting processes even though they don't really know how to do so. So in fact the divination of the algorithm has replaced the arbitrariness of the bureaucrat. Now by making people visible and comparable as individuals algorithms also make it difficult to see in algorithmic outcomes the operation of various structural forces such as the force of the class, gender, or racial structure. They also make it difficult for those who are sorted to coordinate around a common fate. Instead, high-tech modernism institutionalizes the notion that processes of social sorting are nothing but the outcome of a healthy, even meritocratic form of competition. Of course, social scientists know better. Individuals cannot be so easily disentangled from the social structure. And in practice, what we find, you know, is that fine-grained splitting, for instance, along some continuous risk scale, tends to lump back to reveal the operation of well-entrenched inequalities and prejudices. Now, that of course does not mean that nothing has changed in this transition. To realize its technical vision, high-tech modernism must double down on ingesting ever greater quantities of individual-level data, intruding ever deeper into people's lives, and habituating everyone to mundane surveillance and transparency. Second, it must implement the suppression of anything that stands behind— between the algorithm and the individual. Nowhere has this libertarian fantasy played out more vividly than in the public sphere. There, everyone must search for themselves. Anyone can offer content, anyone can be their own expert, and it is again up to the algorithm, which has been conveniently optimized for profit, to sort it out. The mediation of organized knowledge and high modernist elites like the professions is downgraded as inefficient, inconsistent and undemocratic. Not, of course, that the critique is entirely unwarranted, but, you know, the new government regime obscures the fact that it too depends on organized mediations— things like echo chambers, viral trends, or click farms. And perhaps, you know, the ultimate irony of the moral economy of high-tech modernism is that in fact it sanctifies a form of market fairness and the judgment of crowds for everyone, but in fact it is loath to subject itself to their sanction. And so maybe others will have more to say on this. I'll just stop here. Thank you.

Henry Farrell: Dana.

Danah Boyd: Thank you, Henry. Thank you, Marion. Like, the ability to give us a framework to talk about high-tech modernism and for us to have this conversation and debate is such a delight, and I'm honored to be here. Thanks to everyone for joining us. As I've been thinking about what, you know, Henry and Maryann are trying to play with, I keep coming back to one of the challenges that I think are missing from their conversation, and that has to do with networks and the role of networks in this particular configuration. You know, as Maryann just told us, right, we think of machine learning as tools that sort and slot, but this is not neutral and this is not objective because those algorithms are socially constructed in order to create certain kinds of abstractions that are dependent on logics that are rooted within compute computing. They are, after all, tools serving different kinds of purposes. Machine learning does not inherently produce categories, but it often reifies them. And the reason it reifies them is because many ML systems are designed to identify the minimum cuts within networks of data and splice that information into discrete categories. Now what that means is that machine learning systems see structure. They see structure within a network, how information relates to other forms of information, but without any sense of content, without any sense of meaning. So as a result, machine learning algorithms often identify structures that have been socially constructed, the very categories that were produced by high modernism, and then reifies and cements and further entrenches them. And for example, this is how a search engine learns race. It's not that this is a neutral category that has been, you know, that the machine learning system is looking for. It is one that was socially structured and one that affects every aspect of our networks and our information systems. Information. And so when machine learning is starting to cut information across networks, it sees those categories. It's also why we see autocomplete mechanisms starting to quote unquote know gender. It's not again that these are intrinsic, they're part of the structure. So this is where we have to ask ourselves of what it means to make it difficult to organize around categories, because in many ways it's not just that, you know, technology operates relationally by navigating networks, so do people. people have been put into categories by the state, by bureaucracy, by high modernism. But people have also wrapped their identities around those categories, embracing them in ways that calcify the categories while trying to resist the structures and the powers of the state. People live relationally. They live relationally to one another and to the social structures that we create all around them, including the machine learning algorithms we're talking about today. Now, of course, the power and the danger of high-tech systems stems from its tendency to structure those networks in particular ways particular ways for particular forms of power. Those who control the networks control how the structures operate. And just as machine learning cuts networks not into dense and discrete categories, so too does high-tech structures try to segment people into separate worlds. And that separation has social costs. Just as though the bureaucracy was able keeping people in and out of the institution, network structures allow people to be cut across different kinds of network graphs. And that means that we start to see things like polarization and extremism, because polarization and extremism actually come within the structure of the network. When networks are cut so that people do not have the relational formation that allows them to bond together. And so when those cuts are at their weakest, when they happen and create and reify structural holes, those structural holes become caverns that would normally be bridged through social processes. But they are not bridged here because there is no incentive to bridge them within such a structure. And that's where we have to look at the role of power in these systems. Because this is not hierarchical power. Power within networks forms differently. One way to think of it is a centrality power, who has the ability to control a broader sense of relations. And of course, big tech companies who control these structures have tremendous power, but not through traditional modes. Which is why they don't see themselves as powerful, because they're not looking at it with regard to— or they're not actually functioning in a hierarchical sense. Their power stems from their ability to make and remake and structure the networks, to control the flow of information, to structure the graph as they see fit. This is what Manuel Castells has long argued is called network power. And so it's not just the moral economy that's being remade here, it's also the financial economy. Hierarchical power is about boundary work. It's about restricting access to information. It's about creating bridges who can broker information. But economically oriented network power in the conversation that we're talking about, about high-tech modernism, is about maximizing the flow of information and profiting off of the flow of that, being able to sit as an arbiter of the structure itself. So it's a form of scale, but not the scale of a message as we would think of in high modernism. but the scale of the transactions, the scale of the intermediate flows. And that is why surveillance becomes normalized and it becomes part of the system, because surveillance is critical to the structuring of the networks and it is critical for being able to maximize one's position in relationship to the flow. And that of course gets us back to where the McMorral economy fits in. When network power is the core orientation within a capitalist society, this means scale is the currency. We have seen this through venture capital. We have seen this through the idea that we have a handful of players trying to maximize their position. It's not to simply assert hierarchical control, although often that's a byproduct of it, but to ensure that no other actor can achieve meaningful centrality. It is about maintaining control over the network, not just maintaining capital control. Now, of course, unchecked, the problem with this structure is it will lead to greater inequity than even traditional hierarchical power. Power, because when you can control the network and you control the flow of information, what ends up happening is not just about who's in or who's out, but who is in proximity in different ways to different opportunities in ways that end up controlling things. So this is where, as we think about the rest of this conversation, I invite all of you to think about where networks fit into all of this. Thanks.

Henry Farrell: Fantastic.

Charlton McIlwain: Bill.

William Janeway: Thank you very much, Henry, and truly delighted to be part of this conversation with such, such thoughtful and stimulating fellow discussants. I'm actually going to stand back a little bit from immediately addressing high-tech modernism because of its place in the context of the term that we use of a moral political economy. And that's because in critical ways that moral political economy is in process of transformation. And it may sound in the context of this conversation a little unusual to say that I'm bringing good news. The constructs of high-tech modernism, the algorithms, the networks, the central power of big tech, have emerged out of a long generation of a particular moral political economy, one defined conventionally as, quote, neoliberal, but rooted in a set of economic propositions, propositions developed and propagated by economists to define the appropriate role of market and state and indeed maximizing the domain of the market and minimizing and delegitimizing the role of the state. One particular expression of that, of course, was Section 230 of the 1996 Telecommunications Act, which liberates those who are the controlling central forces of the networks of which which they had talked and are the sources of the algorithms that animate this high-tech modernism, to fundamentally exempt them from the kind of oversight, the kind of political state oversight and regulation that had evolved previously over the course of the 20th century from the introduction of broadcast radio in the 1920s '60s right through till roughly the mid-late 1980s and '90s. Well, the fundamental foundations of that political economy are in process of being radically overhauled from within the discipline of economics. In fact, Before the global financial crisis, which represented the mega visible inescapable shock to neoliberal economics and the economic policies or the lack thereof that it had sponsored, it was an emerging body of microeconomic work of significance and of recognized significance. The Nobel Prizes of Economics 2001, 2002, and 2006 are worth taking a moment to consider because the work that was done there, all of which was available, as I say, was recognized in the highest way possible within the profession, stands available today for a contributor to the construction of this new moral political economy within which it becomes possible actively to address the, the high-tech modernism that's the subject of this, of this conversation. And in a way, it's just bad luck for Big Tech that its power has been amassed at such a time as when the context in which that power is observed and may be addressed has, has changed. 2001 was the Nobel Prize for Danny Kahneman for behavioral economics. 2002 was for George Akerlof, Mike Spence, and Joe Stiglitz on the economics of innovation, of the manner in which markets fail, the different ways markets fail, because information is not neutral and is not commonly shared amongst all market participants. And in 2006 to Ned Phelps for the— for the analysis of how expectations can be inconsistent, in fact can be assumed to be inconsistent as the default, in turn leading to market failure. With these sources of market failure all combining to effectively liquidate finally the late unlamented rational representative agent as the touchstone of a caricature of economics that unfortunately was carried over to too many of the other social sciences by economic economist imperialists, not all from the University of Chicago by any means, So the point here is that the stage is being set for an intellectual reconstruction of the intellectual frame with which it becomes possible to consider alternative ways to contest, to constrain, to regulate the sources and the consequences of high-tech modernism in a way that was simply not available 15, 10 years ago. And I think that this can inform our discussion of, if you like, the pragmatics of dealing with high-tech modernism in an open and constructive way. As I say, that has a reach and a relevance. Most recently, I mean, right today, demonstrated by the contest of the Biden administration's economic proposals, both for physical and human infrastructure investment. Again, programs that were simply not conceivable as practical politics, even if they're contested and divisive to an extent today, they weren't even conceivable a decade ago. This is a very encouraging environment. I believe. And with that, I'll stop.

Henry Farrell: Carlton.

Charlton McIlwain: Thank you. And thanks as well for this opportunity to weigh in on this conversation and to think about very thoughtful piece that you all produced, Henry and Marion, to help us stimulate our thinking in this way. My opening comments really, number one, they extend a bit in the same direction as Dana's and focus on the aspect of the category and categorization in your formulation. And for me, there was a quote, a historical quote that kept coming to my mind that I think animates and summarizes in a lot of ways much of my thoughts as I read through and engaged with the ideas in the paper. So I'll read that and share a little bit about why I think that sort of reflects some of my response and thinking here.

Zaynep Tufekci: Here.

Charlton McIlwain: It's a quote by former Harlem, New York Congressman Adam Clayton Powell Jr. It's from 1960, and it was from a crowd address that he was engaging with of the NAACP. So activists who were gathered in part to hear him talk about computer automation and its intersection and implications on the labor market at the time. and he opened his comments saying this. He said, I shall not quote statistics. To do so would be a waste of your time and that of my staff. We know that the Afro-American is the last hired, first fired. We know that he pays a tax on being Black, which makes him the lowest wage earner in the nation. We know that he is quarantined regardless of ability and education, so that his highest achievement can be the attainment of only creature comforts. We know that he composes the largest number of unemployed in this nation today, and we know that the new era of automation does not include him. And so to me, when we think about the role of categories in the formulation or distinctions between high and high-tech modernism, uh, they made me constantly come back to this, uh, quote. And I think something that lies at the center of Powell's points. And so I'll articulate these into two points that really have to do with the— some of the premises of the works and the author's arguments, and not, not so much about the overall argument, which I think we are certainly in agreement with, but a couple of the premises, and really premises having to do with the category. And so the two points I would make are, first, that the category and their respective hierarchy— hierarchies precedes the algorithm. That rather than high-tech modernism producing new algorithms that spawn new categories that become new potential sites for visibility, recognition, contention, and debate, that they continue to latch on to the most strident, the persistent and salient of those categories that have come to define our society, both within and beyond the context of market economy. As an example that perhaps we'll get into later, yesterday's overt redlining and today's algorithmically derived reverse redlining result in the same categories of people being locked out of opportunity to buy into the American dream, which is perhaps a way to say, in a different way, that when it comes to some categories, there is no magic in or reason to put faith in the liberatory power of algorithms. and all right, I'll leave it at that for the first. The second point I will make is that not all categories are treated equally. And so particularly with respect to thinking about human agency, there is a quote in the paper that says, "Conscious agency is only possible where people know about the classifications." And yet I would contend that when it comes to racial classifications in particular, whether that be of Black people as Black or African American or otherwise, Indigenous people, those with CDIB cards or those with only folk ancestral histories, Latinos as Hispanic or white or Latinx or other ways of categorizing, we all have become quite aware of our imposed and lived classifications for some time. And the salience of these categories while they may take a different form, are in my view no less visible or salient translated through algorithms. And so it's their persistence or their persistent salience, either through the bureaucracy or through algorithm, has not yielded much difference in terms of the agency derived to resist the categories or their implications. And so all of this is to say that some categories remain salient salient in the same way between the bureaucratic hierarchies of high modernism and the so-called flat landscape created by high-tech modernism's algorithms. And I'll leave it there for now.

Henry Farrell: Zeynep.

Zaynep Tufekci: Hello, hello. So thank you again for inviting me here. As you said, so happy to be talking about something else. I don't know if I'm able to share share my screen. I have something on my screen. If I am— it might be the administrator is the only person who can do this, but I'm going to try, see if they can. Let me— I think I cannot do it, but I wanted to sort of go back to one thing that I know we have talked about. I think this might be work. So I think people can see this now. This is a figure I'm talking— I talked about in our emails. This is from the Seeing Like a State, and on the, I guess, the left, if you're looking at a screen, is the managed forest in Tuscany, and on the right is like a regular, normal, non-managed natural forest that has been cut for potentially for timber or other things. But has been left to grow on its own. So back to this thing. So the one of the key themes on Scott's book, one of my favorites, is the administrative ordering and structuring of nature and society in order to make it legible to those structures of governance, which also end up classifying and then discriminating and all the other things we talk about. And Scott gives examples ranging from obviously public health and things like like that. So it's not, it's not presented as purely as an oppressive regime, but clearly the project of modernity to make things readable and legible to those in power and governance for reshape it. So, and a lot of people have talked about some of the implications of this and what happens when we reify some of those categories and all of those things. So I think those super fascinating, but I'm really interested in this particular twist in that with machine learning and also the corresponding and preceding digitization of so many things around us, not just the internet, but, you know, carrying phones, gyroscopes in our phones, everything we do online, there has been this incredible production of data. Just, um, points, information, bits, everything that can be fed into these machine learning algorithms that can do the process of making legible for the purpose of optimization without having to plant those trees in those nice little corridors, which changes, I think, both the scale and the scope and the power and the operation of the project of modernity in many ways. So if you think about, like, from a tech example, one of the things that a database might need to do is ask you for your gender, ask you for your race, ask you for those things so that, you know, you get fed into a vaccination statistic at the end of the day or something like that. These are very much within the line of Scott's That's sort of ideal with or without technology. It's what we understand from this. But with machine learning, you can have a classification that is based on not necessarily even the categories you think are relevant, but just an optimization project. Who should we give money to as credit and who we should not, or a million other examples which MRI is the high-risk one, which person is— which piece of information should we show, or which post should we show a person to keep them engaged in our screen? And do that work without the plowing of the land. You can just let it grow in the natural form that Scott has and still use that incredible amount of data to optimize. Now, in fact, if anything, if you look at the history of machine learning as as a technical project. It's there since— right after World War II, people start trying to use it to do classification as a system, just using the basic linear algebra and all the things that go into it. For a long time, it flops. It doesn't work well to optimize or categorize. The breakthrough comes not that long ago, around 2012. In fact, so recent that I remember seeing one of the first papers that broke it open for people, which was the, I think, Dean and Eng, the Google paper. I always call it the cat paper where they took about a million little pixel images from YouTube and let the machine learning figure out you know, classify things, and it came out with a cat. It— we knew it was a cat because we could recognize that cat, but it isolated an image of a cat from these fairly low-res, like 100 by 100 pixels, and nobody had taught it anything of the sort. And all of a sudden, people are like, wait, we can do all these things with machine learning, which wasn't seen as that promising. There were some, you know, uh, technical twists and turns, but whatever else, you know, the backpropagation, everything else, the crucial development was the amount of data. Right, this incredible amount of data that allowed these algorithms to work. Now, I think this is important because many of the resistance, many of the forms of regulating, and many of the things of shaping the project of high modernity, either as trying to make it function better for society or just to resist resist it are based on the idea that you trip up these ordering, the administrative ordering, or you hide from it. Scott's book has people hiding in the highlands of Vietnam to escape the sort of the colonization or the bureaucratic state. Or do you have all these laws about what you can categorize people in, which categories you can discriminate against, and which are in law protected categories that you're you're not supposed to, and et cetera. With machine learning, all of a sudden, you can do things like take people's Instagram posts and use that in hiring decisions. There's nothing— there's no law that I know that stops anyone from doing that, even though that sort of kind of processing, optimizing, will discriminate in some ways and not in others. And some of them will be familiar to us, if it's gender or if it's race, we'll be able to post hoc go back and see it might have been doing this or that. But there's so many things it could be doing that we don't even totally understand, that we don't even know how to regulate because we don't understand. One of the things that I— examples I keep going back to is there's a lot of data on, um, the predictive power of these algorithms in what we clinically measure as, say, depression or mania. You can look at people's Instagram posts and the machine learning algorithm has fairly similar clinically validated predictive power for depression in the next 3 to 6 months, also for mania. And you don't have to like— whatever else you think about depression or mania, I'm comparing it to what we use at the moment, which is the clinical questionnaires. The algorithm, just looking at the post, is doing something at similar levels of prediction. But if you show the same photographs to people, they don't have the— it's not something you can eyeball. So I can imagine all sorts of ways of discriminating in, say, employment that our employment laws have no idea how to deal with, and people have no idea that that's even happening. Because if somebody asks you about— if somebody, like, if you do an interview and you sort of— somebody asks you very pointed questions about a protected category, that's a hint. If you go and measure things and you see women qualified according to some standard, you can objectively decide aren't being hired as much, you can make a decision. There's things if you can, like if you can see, if they have to administratively order it, you can also see into it in ways that our legal edifice has tried, or people in opposition and social movements have tried to put on the agenda as something to pay attention to. Whereas with machine learning I feel like it can be done without the administrative ordering that it requires. And sometimes just our laws have no— or our cultural resistance has no way of talking about it, let alone trying to resist that. And I think that kind of the technical power here is a bit to me like photography versus eyewitness. Like if you were trying to find people based on eyewitness or description, and you had no photography or digital or film imaging or anything like that, you could have ways of describing people. You could have ways of identifying if that was person. It would be subject to all sorts of things, but it's what it would be. Whereas once you have a technology like photography, identifying people becomes a different thing. And then the next step, you have facial recognition from, you know, building images. It becomes a different thing. So I would love for everything else people are saying to also kind of go back to the— it changes the scale and scope of the high-tech modernism— sorry, the high modernism as Scott envisioned it. And one of my favorite books on that, which is also not a perfect book, is James Benninger's Control Revolution, where he dealt with the information technology as a development to control the burgeoning production and consumption markets, as a way to try to get a handle on things that were becoming too big to try to organize by traditional methods of analog computing, or analog— not computing, but analog filing and counting. And then you have the transition. So this is— I, I see it both as a part of that, in that tradition, and also that it can do things in different ways and optimize in things that aren't in our cultural, social, political vocabulary that I think need to be brought in if we are to sort of wrap our minds around all of this. And I will leave that there.

Henry Farrell: Okay, so there are some fantastic questions, some great criticisms that have come up, and Let me jump in. And what I'd like to do is to jump in on the basis of a question from somebody in the audience, but to try and turn it so that it addresses, first of all, Charlton and Dana, and then maybe I get to some of the points that Zeynep has just raised. And so, as I understood both Charlton and Dana in different ways, they give us a sense of how it is that much of what we're seeing at the moment is effectively old forms of systematic prejudice poured into new bottles. That is, that what we are seeing is the reification or the, you know, in a new form of forms of structural and systematic prejudice which have been with us for a long period of time. And that this is something which, you know, which is really, it's a different version of an older problem. And so if you wanted to look at that, then one possible question which I think arises from that and maybe from the Powell quote that Charlton had in particular, is does this mean that the old kinds of ways in which people sought to organize against these forms of systematic prejudice still work in this new algorithmic environment, or do we need to look to new kinds of tools in order to push back, in order to create a better moral economy? And Getting to Zeynep's argument, which is about the broader ways in which machine learning perhaps redefines the space, there's a question from a member of the audience which points to Oscar Gandhi's notion of algorithmically defined groups who experience prejudice. And so on one interpretation, you might say that these algorithmically defined groups are very often going to be the kinds of groups of people who have experienced prejudice systematically in the past. On another, it could be that there are some forms of groups which are basically new groupings. They are groupings which are discovered perhaps by some unsupervised machine learning process which categorizes people in ways that aren't necessarily linked to some systematic, some obvious marker that helps these people to identify each other and to work together with each other perhaps to push back against prejudice. This might lead to a more pessimistic account of what the possibilities of counteraction might be. So I would love to hear, maybe starting with Charlton, then Dana, and then Zeynep, talking to this, and then Marion, perhaps, if you would like to respond to— respond in turn from the perspective of the paper.

Charlton McIlwain: Yeah, that's— I'll try to weave in a couple of things in the response here. I do think there's a sense in which we need different tools. I don't know that those tools need to be computational, algorithmic, or technological in any way. But I think there has to be something different, and that's where my point lies, which is really around agency. That is, if we premise agency on the ability to know and have salient for us the categories and the ways in which they negatively impact those who are categorized, that's the sense, and that's the point I'm trying to make, make for some group of people has always been the case, has always been visible, has always been salient, and therefore been able to be the subject of both agency and action, some of which has been effective, much of which has not been in certain respects. And so I think that we need a different tool. I don't know what that is, but one observation I would make, and Zeynep brought this this up, if we look at the distinction between algorithmic derivations where I may not know what's gone into the algorithm, but I have recognized and identified the outcome and noticed that there is a disparate outcome or disparate impact or effect on a particular group of people and can therefore say, and this is how we have done traditionally to say, look, there is some sense of implicit prejudice or stereotyping or discrimination that's going on. And that has been for many years taken up by the legal apparatus and the courts and been seen as a good model or justification for saying that discrimination has taken place. I think what we're starting to see is an erosion of disparate impact theory that is going hand in hand with the growth of new algorithmic systems that are producing discrimination and judges and so forth less amenable to these these as evidence for discrimination taking place. But yet, for those impacted, that is the same. The impact is the same. The outcome is the same. It is familiar. But I think we need different grounds on which to fight, if for no other reason than the algorithmic evidence no longer being in evidence for that type of discrimination. Hopefully that makes sense to some degree.

Danah Boyd: Thanks, Charlton. I'll pick up per Henry's ordering. So one of the things I would start with is the idea— and this is why I keep coming back to networks, right? Networks allow us to see that networks and groups don't fit comfortably together. And that's true mathematically, but it's also true culturally and analytically. And what we have done over the last 100 years in trying to build cases, as Charlton pointed out, to build cases for anti-discrimination, has started and centered on the idea of a group. And so we can demarcate who— groups, and then we do a lot of boundary work. Who's in that group? Who's not in the group? What happens when the group doesn't actually work within a society? And that boundary work of grouping then allows us to make cases, often statistically oriented, to say, you know, is this disparate impact? That is a framework, and I think that that framework is very much failing right now. And the framework is failing because one of the things that we quickly learned with systems like algorithmic systems, especially ML systems, is you can actually work around those requirements. You can work around those structures. You can ensure that you never tip over, for example, an equal employment opportunity requirement. You can make certain that you never tip over it. But does that mean that everybody is being treated equitably? Absolutely not. And I think this is where Charles and I completely agree. Because I think that for many people, it is a new form of the same thing, but now with different language and with different ways of coding. Coping. And so, because our systems actually bake in already existing structural inequities, many of the worst costs of this, you know, are borne amongst those who are already, you know, feeling the pain within a particular society on a particular issue. And that's one of the reasons why that is normally the challenge that people point out is like, look how these technologies reify those structures. And that is true, but that is also not going to always be true, which is why we do need different languages and new different frames, because what's going to happen is that the systems are going to get a lot smarter. And if we don't figure out how to talk about inequity in the network, we're going to end up this moment. We're not going to have ability to talk about linguistically. I don't think we're there yet. So I think part of it is to think, you know, in, you know, we're moving ourselves away from categories rather than always putting ourselves back into categories. We move our way back out of categories and say, given someone's positionality within a structure. Do they have an equitable opportunity for X? What are their access to their resources? Who do they have that they can turn to for, you know, access? Do they have information? Where are they within that broader position? And right now that is a way to look at it analytically, but we don't yet have the ability to structure that more holistically. It's one of my constant frustrations about the lack of statistics that allow us to understand networks, and the closest we get there is we get to start to have intersectional language, right? Intersectional language allows us to see categories as they relate. But even when we think about categories as they relate, it's not just about adding up all the categories to understand, you know, who's in a position of, you know, subservience to a structural system. It's very much about understanding how these systems are entwined. And that's one of the reasons why I do think that the key to this is going to be moving away from our group-based models, because I don't think our group-based models can get us in response to these these systems. But that doesn't mean that we are prepared for it, and I don't feel like we are. And I'll turn it to you, Zeynep.

Zaynep Tufekci: Well, so much there. I'm both— I'm definitely agreeing with we need to change the way we approach questions of discrimination equity. We don't have the conceptual or linguistic tools with it. And one of the questions I get most, I think, are from the policy people, has come from the finance side, partly because only there we have this particular rule that if you are turned down for credit, you must be given an explanation, right? Whatever level it is, you're supposed to be given an explanation, whereas of course with the machine learning systems, you can't— you're not given an explanation. And there are all these sort of from even with FICO, which is technically not known exactly what it is. It's kind of known what the weights are and what things are. And when you say somebody's FICO score is low, you can even sort of break it down and what's going on. So it fulfills the legal requirement, but there are all these schemes now to provide credit to people or access to people based on just sort of chewing through a lot of data in a non-traditional way of almost always machine learning and going through that. Now, the proponents of such systems will say, and it may even be correct, that in some ways they may be— you may have larger proportions of, say, Black people who do get selected to be given credit, or women, or the things that the law protects, sees as protected categories, and then they fail through that. But it might also be doing all sorts of things we do not understand. And if you are denied credit, it is really— there is no answer. What can I change to be creditworthy? Just by definition almost doesn't have an answer besides please the algorithm, but it's not really known if there's any particular thing you need to do to please the algorithm. And the questions for regulating such things often I get from legislators or lawmakers is, well, we'll make them show us the code, we'll make them, like, we'll force them into transparency, which with machine learning is not going to give you anything. You know, Google Translate right now I think is like 500 lines of code. You can turn it over, that's not where the classification is occurring, it's that interaction between that large corpus of translated things, you know, billions of billions of stuff, giant matrices and teeny little bit of code. So they want to see the code because they're almost thinking like there's a FICO-like process there. They're going to look up and see the Wizard of Oz behind it. And I can totally see the company saying, OK, you know, we'll put you into a room and show you the code. Like, we might not publish it. And they might even make a case, blah, blah, blah. And they'll show the code. And then, like, we haven't done a thing. Because the discriminated person, or the person who's been turned down from credit, their answer remains, you must please the algorithm. It's this new deity that accepts certain sacrifices and rejects others, but nobody's telling you what. It's this completely Kafkaesque thing. So in the name of rationality and further modernism, we are in the weirdest, almost sort of millenarian religious moment where things get decided with not the programmers, not the data people, not the law people, not the person— like, nobody can tell you tell you what on earth just happened or what you must do to improve or get the outcome you want. And then there's this sort of, uh, shaking of heads and shrugging, and then we move along. And I don't think this is sustainable or desirable, even if it looks better in some of the categories we're used to measuring, like race or gender, which we have some sense of it. It could— like, I feel like what can be measured is the tip of the iceberg of what might be going on in this big process we don't really understand. And please, the algorithm doesn't seem like a very good way to base equity or fairness.

Marion Fourcade: All right, this is all terrific. I want to add something about, you know, where sort of this impulse is coming from is partly from an ambition, however twisted and problematic, you know, to get rid of the, you know, the most blatant forms of discrimination. And, you know, there's sort of this idea that if you're going to use if you can use sort of individual behavioral data, somehow you're going to be able to sort of judge people underneath the problems of the social structure. And in some ways, it goes back to an old economic dream that you must individualize, individualize, individualize. But of course, any sociologist would know that all individual behavior is socially patterned, and for reasons that have to do with the way that it is being recorded, for the way that simply people act differently under very different kind of opportunity structures. And so it is those differences, of course, that get then, you know, that get translated into what looks on the face of it like, you know, this individual behavior. And this is where the question of disparate impact comes, you know, comes in, and this is where it becomes really problematic. But I think it's important to understand that there is this sort of moral economy behind all of this, that you'll be able somehow to do this, and that all you need to do is individualize more, you know, and you have to— and it is also what's pushing, I think, all of these firms and organizations to want to collect more and more data because there's this idea that somehow the answer is in more and more data. But of course it won't be, right? Of course it won't be because the social structure is still there, right? And it is there in everything that people do. And especially in the way in which it is measured and categorized and individualized and so on. So I think that's the sort of fundamental problem at the core of this, right? Now there is also the question of sort of the politics of this. So we've talked about the fact that the politics may be, it's very hard to orient yourself to sort of some, to the category of, I don't know, people who have a credit score between 520 and 580 or something, you know, that's not really a category. In fact, a lot of the stuff that, you know, comes out is not categories in sort of the traditional sense. You may have categories that are, you know, actually along some sort of continuum or scale, right? And so politically, because there's always this notion that, you know, things are always in movement, it becomes really hard to orient yourself to that. So you, of course, you have to fall back Black, you know, on the more traditional categories that we know, you know, feel the burden of this, you know, these disparities more. But this is not to say that the other categories are irrelevant. They are, but they're, you know, like the low credit scores or something, you know. this is something, it's happening. And if you have a low credit score, it might create all kinds of problems outside of finance, in the housing market, in the labor market, perhaps, in the insurance market and so on. It may percolate in a lot of ways. So it has some sort of— what Gandy calls these algorithmically defined different groups or categories do have also sort of an impact. So that's sort of where I want to go. I mean, I think we're creating this sort of new form of meritocracies, but we don't know how to orient ourselves because we don't know what the, you know, what counts really as merit. I mean, I think as Zeynep really explained very well.

Henry Farrell: So I know that Bill would like to respond to some of these questions as well. And I also want to ask Kim a more specific question, which is about the role of economic logic. Because of course, when we're talking about the politics of this, very often we think about resistance from beneath, but also the role of the state is pretty important. And antitrust is very often the way in which the state looks to get involved in these markets or market-like processes. So I guess the question I have is, when we're thinking about these major questions from the perspective of economic theory and antitrust. Clearly economic theory, George Stigler and people in the 1960s helped shape the world that we're in. And I'd be interested, in addition to whatever other comments you have about what's happened so far, if you could talk a little bit about what kinds of possibilities there might be for antitrust, and then other people can jump in on top of that as they see fit.

William Janeway: Thanks, Henry. Um, first I just wanted to pick up on, um, the manner in which Zeynep correctly, um, discussed how machine learning algorithms specifically play such a powerful role in the construction of the algorithms that, as Dana said, reify, uh, categories. They don't just reify them, they legitimize them. Machine learning in the social world takes correlations and presents them as objective truth and presents them with a kind of implicit presumption that if you just study them closely enough, you will discover the causal relationship behind and underlying the correlations. And of course, the first law of big data is the more, the more data you have, the more the certainty that false correlations, meaningless correlations, will rise exponentially, and that the challenge of extracting causality from behind that sea of correlations is all the more difficult. And this, in a sense, to me, is kind the reductio absurdum is this incredible exponential increase in the number of parameters applied to natural language processing systems like GPT-3, which are still in the position of simply trying to predict out of the vast amounts of data they've consumed what will be most likely to be the next word that a human being would when the software itself has no idea what the words mean, either individually or jointly or strung together. Gary Marcus and Rodney Brooks provide excellent critique of, of this sort of pretentiousness of machine learning world. But that also gets me around to where Dana has been talking about networks. As it happens, over the last really 10 to 15 years within economics, again, in this rich field of microeconomics that's been evolving, uh, behind the caricature of the rational representative agent, an enormous amount of work has been going on to understand first how the economic system itself is a network of networks. Networks, how within those networks, those networks of production and distribution, centrality and bottlenecks create the potential for concentrated market power. And concentrated market power then can be seen in the trade-off— as a result of a trade-off between efficiency and resilience. And we have spent precisely in that long generation from when Stigler and Posner liquidated the intellectual basis for antitrust policy, we've seen the triumph of efficiency in the construction of networks. That is least cost as the only criterion of merit, leading to radically reduced resilience, as demonstrated most recently, of course, in the supply chain crises that the pandemic generated and revealed. So it's in that context, again, I say that the libertarian bros of Silicon Valley have had terrible timing. It is the case that antitrust law The law is finding a new basis for legitimacy and antitrust practice is quite likely to follow. And that can be at the level of whether Facebook can own, can have bought up and then sought to integrate the data sources from what would have been what would have been potential competitors to the more behavioral abuses that were for many, many years the legitimate subject of an active Federal Trade Commission, which is now perhaps again likely to become active. So the antitrust question is very well embedded within in this transformation of economics, which also, as I say, increasingly takes seriously the role of networks as to how the world works and analyzing networks in order to identify sources of power that in turn may generate market failure on the one hand, social failure on the other hand, and legitimize political response.

Henry Farrell: Okay, I want to bring in a question from the audience, and we're going We're coming towards the end, so if people can be as quick in their answers as possible, it would be wonderful. But somebody from the audience points out that a lot of the problem that we see today is of a kind of an environment, a physical environment of the internet which is being homogenized rather like the plantations of trees that Scott describes and seeing like the stages of you have these normalen baum or whatever he calls them. But now we have this huge infrastructures which are being run directly by Facebook, by Google, and by others, and which make it far more difficult for actors to come from beneath and to change stuff and to disrupt things in the way that, at least in theory, they were supposed to do. So how do we deal with that kind of problem and with the other sources of power that people identify as being major, major, major major constraints that are perhaps sort of making it more difficult to deal with this broader set of political problems? Dana.

Danah Boyd: The thing that I would say about the tech platforms and the centrality that they've had in this whole conversation is that this has everything to do with a particular arrangement of capitalism. Venture capitalists want a return on investment. There is no— like 20 years ago, there were countless technology companies that were starting without major VC. Now you can't get a restaurant started without VC. VC. Like, that is the state that many of this is in, such that the expectation is not just that you will become a profitable organization, but that you will continue to produce return on investment for your stockholders in many ways, or funders in different ways, until you collapse. And that kind of hockey stick growth— like, I always joke that, you know, nothing grows at that speed other than cancer, right? And that's where I think we are at right now, is that we've created a cancerous structure because we expect return on investment at that scale.

Zaynep Tufekci: So if I could tie that to one more thing, is that if all of this had happened in a different political capitalism environment, of course, we would have had different things. But one of the things that Dana's just talked about that's really crucial to this is this cash that's sloshing around the global finance system without a place to go is a key driver of like everything we're talking about in many ways. And that at some point needs to come into the thing because you have that much venture capital money because yeah, you have all this cash sloshing around. And what has happened is, you know, what the startup ecology is, they're just waiting to be purchased by Google or Facebook or one of those, like it's the echo higher thing. And as she said, that's, I think that's a pretty good way of saying it. You need VC to start a restaurant and hopefully somebody will buy you out and that's your ticket. So that's part of the distorting process of, um, so it's going to sound weird, but in some ways it is anti-innovative. It is genuinely anti-competitive in my view, because whenever I I talk to sort of my more edgy friends, including in these companies who have all sorts of ideas of things to do that might even at least on the margins and sometimes fundamentally transform some of these things or look into some of the black box problems or this or that. Like, why bother? Like, there's nobody who seems even slightly motivated to bother because the game seems to be you tell a story to venture capitalists and you get money from them, and then some big company buys you, and then everything is the same way it is. And this can function to the degree there is this much excess lopsided cash in the hands of few people that's looking for a place to go, is distorting this whole thing. I do want to say one thing though, to something that was said in terms of the causality link. I think I'm gonna disagree a little bit here. So we can spice up this discussion. I think one of the inflections we're seeing is that trying to look for causality in the mechanisms of intervention is very much a 20th century thing that is being increasingly superseded by the optimization thing where you don't feel like you need to understand to intervene for good or bad and you just are presented with some of these things. So we're, we're not having the correlation-causation being conflated problem. We're having this, we don't even care and we don't really understand, kind of moving away from it, which I can't— like, as with all things, I can't say, oh, it's always terrible because, you know, these things can read pap smears better than people and we don't know. Like, I'm not— this is an interesting— we don't know why people read pap smears the way they do either. It's this weirdo art sciencey things and the machine learning can do it. So it's not a 100% always terrible thing, but in some things, of course, the whole legitimization of society is based on a theory of how society works and causality and what we're doing and what we're not doing and why, so that people can have a political voice in the process. And if you remove causality from the discussion, you remove political voice as well. And I think that's a major problem with, like, with Oscar Gandhi's The Big Sort, kind of the book which is prescient and amazing and excellent, and I can still read it and understand it, which is amazing for that kind of book, right? Because usually books get outdated, and his book is excellent even though it was written before Google was even founded. But my feeling is we've fallen behind it in terms of the politics of this discussion because of this move away from causality.

William Janeway: I have to say that I still care about causality, partly because I care about political action in response to what's going on in markets. But speaking as a venture capitalist, I want to make two quick points. First, um, 20 or 30 years ago, no significant computer company was started without a venture capitalist. You have to go back to Hewlett-Packard, uh, to find one. Um, what is interesting today is a moment in capitalism which I think is arguably, um, unprecedented as a consequence of the limited range of tools available to states in response to to two successive mega shocks, the global financial crisis and the pandemic. The central banks of the world have done what they have never before done in peacetime. They've driven real interest rates, ex-inflation, to negative levels and generated incentives for exactly that flood of money that has come into the world of tech has overwhelmed, has almost drowned and forced some to come and increase, some venture capital firms, far beyond their capacity to actually function as professional venture capitalists. And they have funded in turn a business model which has been known as blitzscaling. Do not worry about your customers giving you enough money to develop and deliver something of value, a service, a good, something of economic value, of commercial value. Just rely on the ability to sell securities to investors. This is producing, of course, a flood of companies whose business models make those of the late 1990s of the dot-com seem almost rational, almost plausible, because these are completely out of control. There's an electric vehicle, electric truck company going public to raise $10 billion at a $50 billion valuation this week with zero revenues and no particular reason to believe it will ever be able to pay its bills. And of course, even at a trillion-dollar valuation, Tesla does not generate positive cash flow on an actual cash-to-cash versus a manipulated accounting basis. So this is a very unusual moment in the history of capitalism. And it does bear on the state of tech, the state of high tech, and the state of high-tech modernism.

Henry Farrell: So Charlton, in your book, you try— you talked about how it is that Black entrepreneurs back in the 1990s looked to use AOL and similar types of companies to jumpstart a broader sense of Black community. So given these problems and given perhaps also what Dana said about how it is that structural holes between different communities are— have become caverns in a sense, what kinds of possibilities are there out there for bringing together community and business models today, or are these becoming more antithetical things?

Charlton McIlwain: I think in a way they are becoming a little bit more antithetical, and I think it goes back to this question around antitrust and this ties the platforms and so forth, where when we think about scale and the scale of potential change, it's almost as if there is no other ground to effectively play on unless it's the, the ground of these these major platforms, meaning it's hard for me to understand and see the possibility of change happening from the ground up or in communities wherever they spawn, be that online or quote unquote offline, that has the possibility of the kinds of scale, scale of change needed to really transform this. And yet, you know, I think the irony and the tragedy is that in a moment post-George Floyd and so and so forth, these are also platforms offering to step in and fulfill that or fill that gap and asking questions about what role can we play, the platforms, in making these things more just and so forth. And I think it is a tragedy that that becomes the site at which we have to play and engage and almost only one if we think about a kind change at a scale that needs to be done to effectively fight the scale of the problem as it's laid out.

Henry Farrell: So we're coming towards the end of the session, but I was wondering, we have a couple of minutes left. Marion, would you like to jump in with any broad thoughts to tie this all together?

Marion Fourcade: I don't know if I can tie this all together. I actually wanted to go back to the question of VC and the availability of cash, which I think really stands at the core of what we are seeing today. And Bill mentioned the business model of blitzscaling, but how do you blitzscale? One of the things that I think is underappreciated is the ability of these firms, precisely because they are backed up by so much cash, to actually blitzscale without selling anything, without really selling. That is blitzscaling by what I've called in other words, gift, essentially. That is offering free service and so on, and enrolling people in this sort of very, in this manner, but making it essentially too much of a good deal and too irresistible and so on. And so forth. And that I think is, you know, continues to be very important in Silicon Valley and in this, you know, in the structuring of digital capitalism. We see it, we saw it actually happening during the pandemic, you know, with sort of this, you know, dozens of hundreds of educational apps that, you know, teachers can download and kids can download. And so on for free. And then once you have established yourself, this is when you move to a monetized model. And so I think there's a dynamic here, which is a network dynamic that Dana was talking about very much at the beginning of this, that is, that is at the core, that is actually relying on a very human kind of sociality and sort of this human impulse to reciprocity and to take part collectively in things. And so this is where sort of the economy for me and the economy where the economic and the social actually meet in this sort of irresistible dynamic of the gift and the social structure that stands behind it.

Henry Farrell: Well, this has just been a fantastic conversation. Sorry, just making sure that I'm not muted. And I just would like to point in passing, people who are interested in the importance ready access to money as a constraint or not as a constraint. This is a major theme of Bill's book, which went into its second edition a few years ago, and how it shapes the economy. But I think one of the things that has really come across to me from this wonderful conversation is the crucial importance of bringing together, on the one hand, the kinds of conversations that people are having about new media, On the other hand, social sciences, because there really has not been nearly as much of that conversation as you might expect. Social scientists, with some important exceptions, have not paid nearly as much attention to these dynamics as they ought to. The result has been that I think that there has been an impoverishment of social science dialogue and also a difficulty in confronting the really big structural questions that are raised by this new world of algorithms, by this new world of business models, and how they are reshaping the world. So this has been a wonderful conversation, but I feel like it's kind of a taster for a much bigger and much broader conversation that ought to be happening around us. I'm just delighted that we've had such wonderful participation from people who are really engaged in this debate at an extremely high level, and I look forward— I really hope that we see more of work coming out of this conversation and also returning to these themes from a variety of different perspectives, arguing out, uh, in a broader sense, uh, these are very, very important questions. So thank you all for participating, and I also want to thank again the sponsors of this event, uh, the co-sponsors Data and Society, the Essex Society and Technology Hub at Stanford, the Institute for New Economic Thinking, uh, and the Social Science Matrix at University of California at Berkeley. And finally, I want to give you a heads up, which is that this is, of course, one in a series of talks. You can go back and you can see many other talks in this series on YouTube. There will be more coming up. And in just a few seconds, you're going to see a screen coming up, if you're in the audience, describing the Social Science for a World in Crisis series and giving you a sense of the other things that you can find in this set of debates that has been happening. So thank you, everybody. Thank you to the panelists. Analysts, and thank you all for joining us today.

Narrator: That was Dana Boyd, Henry Farrell, Marion Foucault, William Janeway, Charlton McIlwain, and Zeynep Tufekci discussing high-tech modernism. If you enjoyed this episode and crave more intellectual conversations, be sure to follow us online or in your podcast app of choice. There's a nice catalog of interviews and discussions exploring a wide spectrum of social science topics. And if you're interested in learning more about the center's people, projects, and rich history, you can visit us at our website at casbs.stanford.edu. Or you can join in the conversation on Twitter. We're @CASBSStanford. Until next time, from everyone at CASBS and the Human Centered team, thanks for listening.