How should social science think about AI?

Big structures and large processes are back, baby

Things have been happening since the New York Times published a comment piece by Alison Gopnik, James Evans and myself on the relationship between AI and social science last Sunday. Our piece talks about how the AI labs play an increasingly large role in shaping academic research.

In the interim, we’ve seen what happens when this relationship turns predatory. Not that OpenAI describes it in quite that way. When OpenAI’s Sébastien Bubeck offered help, collaboration and credit to an outside mathematician whose approach to solving a major problem he had gazumped by onboarding millions of dollars of compute over a few days, he says that he framed his offer in terms of “working together” to make “something good.” When the mathematician threatened to go public, Bubeck reportedly asked “[w]hy would you ruin your career?,” but he was most certainly asking in sorrow rather than anger. When Tom Hagen conveys the offer that you can’t refuse to the movie producer in Godfather I, he is, after all, presenting that rare treasure which is the friendship of Vito Corleone. It’s tragic and unfortunate for all concerned that his offer is rudely declined, necessitating the decapitation of a prized racehorse. With just a little more mutual understanding, all that unpleasantness could have been avoided.

Nor is that all that is happening. Media discussion about AI has rapidly shifted from ‘excitable’ to ‘febrile,’¹ following revelations of agentic AI driven cybersecurity problems, an Anthropic employee quitting because of existential risk concerns and other stuff too. Max Read has an extremely helpful summary here. Much of the popular debate (and some of the AI debate too) has been driven more by misleading metaphors than by actual understanding of the technology and what happened, let alone what its actual consequences for society might be.

All this unsurprisingly reinforces my belief that Alison, James and myself are right. We need a public interest oriented version of the social sciences and we need it now. But it is worth explaining why.

As we have already argued in a piece we wrote with Cosma Shalizi for Science, the best way to think about AI is as another social, political, cultural and economic shock in the series of shocks that have been underway over the long industrial revolution. The particular version of this argument laid out in the NYT rests heavily on the arguments of the late Charles Tilly, who was both a sociologist and political scientist, and arguably a historian as well. Tilly pointed out that the social sciences in their modern form were by-products of the upheavals of the nineteenth century industrial revolution and the desire to address and remedy them.

We would be better off if we could rely on actual social science, rather than the kind of bong-hit P(d00m)ing in much of this week’s media commentary to understand the upheavals that are happening around us.

Equally, although we don’t talk about it in the NYT article, Tilly scomplained that the social sciences were blinkered by the defunct anxieties of the period when they had originated. They needed, in his argument, to think much more systematically and ambitiously about big historical processes. That seems apposite as we wander amidst the miasma-producing fever swamp of ‘feel the AGI.’

The challenge is to figure out (a) how to fund the social sciences so that they don’t depend on AI labs, and (b) how to reorient the social sciences so that they approach AI in actually useful ways. We talk a lot about the first in the NYT piece: I summarize it below. We don’t talk about the second, which is what I am going to end up talking about here (and it is perfectly possible that Alison and James disagree with some or all of what I say - any errors are mine and mine alone).

******

The New York Times article explains how massive shifts in social science funding may damage our ability to understand the consequences of AI. We start from the contention that the social sciences’ original job was to map out the large scale disruptions of the industrial revolution, enabling collective response.

The fear that machines were wrecking society created political demand for reliable information on what was happening on the ground and how people were living. In Britain, where the Industrial Revolution began, the journalist Henry Mayhew calculated that hundreds of Londoners made their living gathering bones, rags and cigar ends from the street, as well as dog excrement for tanneries. … Social science research arose to study these issues in a more rigorous fashion. Economists studied markets, political scientists studied politics and the state, sociologists studied changing society, and psychologists studied minds. Though these investigations had flaws and were sometimes biased, together they rebuilt our public understanding of what was necessary and possible.

Here, we are partly riffing on Tilly who explains in his book, Big Structures, Large Processes and Huge Comparisons that “Out of … nineteenth-century reflections on capitalism, national states, and the consequences of their growth grew the disciplines of social science as we know them.”

The social sciences should now be gearing themselves up to deal with a new set of radical technological disruptions. However, they are poorly situated to do that.

The Trump administration has slashed federal funding for the social and behavioral sciences, while private funding for certain A.I.-related work has skyrocketed. The result is that many social scientists have left academia and traditional research institutions for A.I. firms that are effectively building out their own social science programs — which threaten to favor the ambitions of A.I.’s creators over the public’s interests. To navigate the shock waves of A.I., we must restore social science research as an enterprise that starts from the public interest.

At exactly the same time as the Trump administration is zeroing out National Science Foundation funding for political science, economics and sociology, the two dominant AI “labs” - OpenAI and Anthropic - are promising enormous amounts of money to fund social scientists interested in working on the transition to an AI economy, and how best to manage the disruption.

We suggest that this is not going to work out well. Even if you believe that the AI labs are doing this with the public interest in mind (there are publicly minded people working in them), their understanding of the general interest is highly idiosyncratic. Specifically, as we note, “They are more interested in smoothing the A.I. transition than questioning it.” I personally think you could go a bit further (I don’t know if James or Alison agree). I suspect that some of the labs’ willingness to put money into the social sciences stems from the realization of senior people that the practical build out of AI is highly unpopular in the U.S and their urgent desire to figure out how to sugarcoat the pill so that it will get swallowed.

******

What we don’t talk much about in the article is social science’s relationship to AI itself. Here, a different aspect of Tilly’s argument can help. Even if the social sciences manage to get the funding they need, they may end up approaching the problems that they are supposed to study in unhelpful ways. When Tilly says that the social sciences are the products of nineteenth century traumas, he also suggests this makes them worse suited to deal with more recent problems. In his account, they remain trapped behind the bars of nineteenth century cultural neuroses.

The beginning of Big Structures is simply fantastic. I’m going to quote it in extenso, not just to convey its insights, but because social scientists just don’t write like that any more.

We bear the nineteenth century like an incubus. Inspect the map of almost any American city. Notice the telltale marks: rail lines slicing one section from another; a speculator’s grid, with its numbered rectilinear streets and avenues repeating themselves to the horizon; clustered residential areas, once serving as suburban middle-class refuges from the city but now absorbed into the urban mass. Stroll through and see it up close. Notice the characteristic artifacts: department store, office building, warehouse, factory, chimney, boiler, electric pole, street mixing people with machines. For all the twentieth century’s new technologies and stylistic experiments, the apparatus of everyday life still bears strong markings of the nineteenth century .

The same is true for many of our ideas and institutional arrangements. … In these waning years of the twentieth century, the nineteenth century also keeps its hold on many ideas about social organization. In the analysis of social change, we cling loyally to ideas built up by nineteenth-century intellectuals. Intellectuals formed those ideas in their astonished reaction to what they saw going on around them: unprecedented concentrations of population, production, capital, coercive force, and organizational power. They formed ideas treating increasing differentiation as the master process of social change, ideas of societies as coherent but delicate structures vulnerable to imbalances between differentiation and integration, and other ideas connected to them. The nineteenth-century incubus weighs us down. I hope this little book will serve as a lever to lift some of the burden. It addresses one big question: How can we improve our understanding of the large-scale structures and processes that were transforming the world of the nineteenth century and those that are transforming our world today?

If you’re like me (in fairness: you probably aren’t and shouldn’t be), this is a thrilling call to arms, which still resonates as the twenty-first century begins to wax. The call resonates more because we are obliged to think about large scale changes, whether we want to or not. New structures and new processes are emerging, and we are poorly situated to understand, let alone guide them.

Tilly, writing in 1984, suggests that the social sciences ought to be better suited to think about such structures and processes, but that they often ,tend to on “timeless, placeless models of social organization and social change,” which are often reanimated by the dream of a return to order in lieu of “historically grounded analyses,” which recognize the mess and disorder of actually functioning human societies.

What does Tilly want such historical analysis to do? First, he argues that it should not get stuck at the level of the nation state. There are many important relationships that cross national boundaries. Indeed, Tilly argues we should replace talk of “society” and “societies,” which smack of some imaginary lost age of cohesion, with “multiple social relationships,” which vary over space and time. Nor should it be methodogically individualist: individuals are important, but social relations involve two or more of them. We ought treat the notion that history passes through abstracted “stages” of development with the contempt that it merits, discarding it as well as nostrums about modernization, fears of inevitable social atomization, and the belief that there is an innate distinction between legitimate and illegitimate violence. Instead, we should focus on the larger processes that produce enduring patterns: world systems that may in turn produce “networks of coercion,” some of which cluster into states; processes of proletarianization, urbanization, capital accumulation, state-making and bureaucratization, which vary and combine with each other. We should compare across different situations, but not get hung up on the “classical logic” of searching for precisely comparable cases. Our methodology should fit “like a sweatshirt and not like a straitjacket.”

Tilly was opinionated: his general methodological recommendations for the social sciences bear a remarkable resemblance to his own theoretical predilections. You don’t have to buy into all of his arguments to find the book valuable, though it would be really interesting to think about how his ideas might work today. We are able to do a lot more to study networks and flows than we were in the 1980s, when Tilly wrote. Equally, the social sciences now lay a much greater emphasis on identifying causal relationships, which tends to push away from the kinds of big secular shifts that he suggests we ought focus on.

Even if Tilly’s methodological hobbyhorses aren’t yours, I think that two elements remain important. First - that social scientists should not get caught up in simplifying narratives. In particular, they ought ignore both the soothing suggestions from modernizers that our history is an ineluctable path from past chaos to future order, and their intellectual adversaries, who hark back to some imagined past era of organic harmony before the steam engine, electricity and flush toilets ruined everything. Past, present, future - all are far messier and murkier than our abstractions suggest, with lots of different processes happening at once.

Second, that social scientists ought pay a lot of attention to struggle and power. People disagree and fight - the variable processes that drive change are not smooth and technocratic, but the consequence of their collective struggles with and against each other. And some have much greater power than others to shape their fates and the fates of those around them.

******

This gets us back to the question of how the social sciences ought best study AI. The lesson from Tilly - and from others too - is that a good research agenda ought emphasize complex processes and power. That would be very different from the public debates of the last week (and indeed, the dominant debates of the last couple of years). In his summary, Max distinguishes between “doomers” who believe AI is powerful and will kill us all, accelerationists who believe AI is powerful and will liberate us all, and skeptics who don’t think that AI is powerful. There isn’t much room in this three cornered debate for what seems to me to be the best starting point for the social sciences: a recognition that AI is powerful (and will reshape power relations among humans) but that the consequences will flow messily through multiple complex processes that we really need to start mapping and understanding.

There ought to be room, and there is at least one programmatic article that points in the direction of building it. That is a piece that Alondra Nelson wrote for Daedalus earlier this year.

Alondra does not start from Tilly’s work (indeed she doesn’t mention it), but she ends up in a roughly similar place. She argues that we need to focus on big secular changes as Weber did (Tilly too praises Weber for studying processes in an unromantic way). She says that we absolutely need to focus on power:

Weber’s focus on rationalization left crucial social dimensions unexamined. At the same moment, W. E. B. Du Bois insisted that technological change could not be understood apart from relations of power, particularly those structured by race and empire.

We ought understand not just how technology shapes who has bargaining power but what seems socially possible. Specifically:

Efficiency and productivity became unquestioned ends, narrowing thought and desire and foreclosing the imagination of alternatives. This dynamic is legible in today’s AI systems. Recommendation engines steer attention and desire, curating environments that make certain possibilities appear natural while obscuring others.

This then, seems to me to provide the other part of the argument. Alison, James and I argue that the social sciences need funding to map the consequences of AI. Alondra sketches out a map for how sociology, economics, political science, and science and technology studies should address the ways in which AI itself is becoming a structuring force, shaping a multiplicity of processes.

What is needed is not simply a catalog of [AI’s] effects across sectors, but integrated accounts of how algorithmic systems are transforming the fundamental coordinates of social existence: the nature of work and economic value, the character of public life and political authority, the production and circulation of knowledge, and the very boundaries of human agency. Such frameworks will grasp artificial intelligence not as a discrete technology but as an infrastructural layer that increasingly mediates how institutions operate, how individuals are seen and sorted, and how collective futures are imagined and foreclosed.

AI, from this perspective, is best understood as a social technology:

At a high level, a consistent finding emerges from this research: AI is not an external force acting upon society from outside but an internal phenomenon constituted through social processes, institutional choices, market dynamics, and political struggles. Its effects are neither uniform nor preordained. They vary with context, design, and the distribution of power that determines how these technologies are built, deployed, and contested.

Cosma and I might additionally point to the opportunity to bring computer science and the social sciences together to understand this, building on the work of Herbert Simon.

That, then, is why I think we don’t just need well funded and politically and economically independent social sciences to understand the consequences of AI, but also to reorient the social sciences so that they are again well suited to deal with big structures and large processes. We are in a world where it is urgently important to understand change at scale, since it is just such change that we are going through.² The social sciences need both solid funding, and a lot of internal work to reorient themselves towards the questions that matter. I’ll be writing more on this, both on my own and with others.

1. My mental image of “AI debate” bears an embarrassingly close resemblance to the dial on the forehead of 2000 AD’s Mean Machine only with ‘excitable,’ ‘febrile,’ ‘unmoored’ and ‘deranged’ as the four possible settings. My bad, for sure.

2. I don’t think it needs to be said that there are other sources of such change than AI, but I’m still saying it, just in case.