Skip to main content

Causal Revolutions: From Judea Pearl to Edmund Burke, via Catherine Gallagher and Thomas Bayes

Paper delivered at the ASECS 2022 “Eighteenth-Century Counterfactualism” panel

Published onApr 03, 2022
Causal Revolutions: From Judea Pearl to Edmund Burke, via Catherine Gallagher and Thomas Bayes

Lately I’ve been wearing several hats at my university: teaching in the English Department and pursuing my research interests in literary history, while also helping develop a research program in the ethics of artificial intelligence, which I was thought qualified to help with because I’ve had a part in building our college’s digital humanities program. At first, I worked on ethical AI mainly in an administrative or service capacity, but as I’ve become more conversant in the problems and possibilities of computer science research, I’ve begun to see exciting contiguities between research in that field and in my own home field of British literary studies. This paper attempts to explore one such contiguity, the one named by my title theme of “causal revolutions.” As my subtitle suggests, I’ll be telling the story of causal revolutions across statistics, computer science, literary history, and critical theory, by talking about the contributions a set of contemporary and eighteenth-century thinkers have made towards understanding a resurgence of counterfactual imagination. Because I’ll be telling this as a story about thinkers more than about thoughts, I may overpersonalize this story, reducing it to a series of “discoveries” rather than of movements. I hope you’ll forgive this reductive approach, however, when you see the cross-disciplinary breadth of the picture it allows me to sketch. I also hope you’ll forgive me my further disciplinary tresspasses in this talk. I must stipulate at the outset that not only am I not a computer scientist, I am also neither a mathematician, a philosopher, nor a historian of philosophy, as some of you may be. And while I am something of an historian of eighteenth-century thought, in that area, as you all know well, a little knowledge is indubitably a dangerous thing, and I may be even more at risk. So I sincerely hope those of you with such relevant expertise will gently correct my errors, either in the Q&A or after the talk, so that in due time I can revise this sketch into a more proper panorama.

Moving now beyond disclaimers, let’s begin where I will also end my trajectory today, with one of the most famous eighteenth century works on revolution, (2) Edmund Burke’s Reflections on the Revolution in France. Here is a passage most of you will recognize, (3) where Burke brings his whole counter-revolutionary argument to a point, declaring that “The science of constructing a commonwealth, or renovating it, or reforming it is, like every other experimental science, not to be taught a priori.” Politics is a science for Burke, but not a theoretical one. It ought to be a practical, empirical pursuit. The rights that concern political philosophers are, in Burke’s famous metaphor (4), refracted and reflected in common life “like rays of light which pierce into a dense medium.” We thus analyze such rights only with great difficulty, (5) since in practice human rights are “incapable of definition.” Revolutionary arguments based on the rights of man may proffer clear models, they may even be metaphysically true: but exactly for that reason, they will likely be morally and politically false, because they will be fundamentally impractical. (6) This argument of Burke’s we know, but I want to highlight an aspect of it that I haven’t seen emphasized. When Burke cautions his reader against using untested theories in moral philosophy, he does so by mounting an argument about the perils and possibilities of causal inference. (7)

If we look again at the outset of this passage, we find that Burke is explicitly talking about the difficulty of parsing “the real effects of moral causes,” in both their “immediate” and “remoter operation,”, and of tracing how “some obscure and almost latent causes” in states can be of the greatest consequence. And to be clear, Burke believes we should do this work, so long as we do so a posteriori. Burke is not saying we should have no political theories, he is saying we should revise such theories retrospectively—if you will, reflectively. (9) When we look again at the second part of Burke’s argument, we find that while “the rights of men” may be “incapable of definition,” they are “not impossible to be discerned,” but can be analyzed by “political reason” as “a computing principle; adding, subtracting, multiplying, and dividing, morally and not metaphysically or mathematically, true moral denominations.” (10) A computing principle: Burke is arguing here for the application of a computational causal analysis, so long as it is one that works with moral, not economic, denominations—with names, we might say, not with numbers. Everyone who has read the Reflections remembers what opprobrium Burke heaps, elsewhere in the text, on “the age of sophisters, oeconomists, and calculators” that “has succeeded after,” in his most famous phrase, “the age of chivalry is gone.” But Burke does seem to ascribe some value to a computational ethos, so long as it deals with the complexity of true moral causes, and seeks to understand the effects of what, like chivalry, has come before. In modern terms, we might say that Burke calls for a mixed mode of analysis, in which quantitative judgments are carefully grounded in qualitative knowledge.

Let’s flash forward now from 1790 to 2020, and from the French Revolution to what the computer scientist and philosopher Judea Pearl has announced as “a causal revolution” in scientific analysis across all fields of inquiry, a revolution Pearl sees as having been made possible by a resurgence in counterfactual thinking catalyzed by twentieth-century developments in modal logic. (11) One of the most prominent thinkers in computer science, Pearl has been working for decades on techniques for causal analysis that he hopes will make it possible to write algorithms that will help robots and other AIs make judgments about what effects stem from what causes. (In so doing, it should be said, Pearl has participated in a general movement whereby twenty-first century theories about causation have emerged along the disciplinary border between computer science and philosophy.) Pearl considers the analyses he has discovered “revolutionary” because they overthrow a centuries-long consensus that philosophers and scientists can’t say much that’s meaningful about causation, but can only amass trivially materialistic observations about causal mechanisms. Pearl complicates, this is to say, the default idea of causation that most of us still work with in theory, if not in practice: the notion that when we think we see causation, what we see may just be correlation, and we can’t say much about even causation we think we observe without actually explaining the causal mechanism at hand. So while Pearl, from his perspective, calls restoring the dignity of causal analysis “revolutionary,” from the point of view of the dix-huitièmiste, this restoration is arguably counter-revolutionary. It seeks to depose the very mode of statistical thinking founded on correlation, that, according to our baseline Foucauldian paradigm, emerged around 1800 with the emergence of biopolitical governmentality. From the point of view of partisans of thst still-dominant mode of statistical analysis, causal analysis has little analytical utility. Still, Pearl, like Burke, is a counter-revolutionary with a counter-theory of how to proceed, and discovering how to say more about causation is the computational mode he embraces. (12)

To be sure, many statisticians consider Pearl’s claims about the revolutionary character of his method to be overblown hype coming from the author of a popular science bestseller, The Book of Why. And twenty-first century causal inference such as Pearl teaches may be counter-revolutionary, but it hardly seems counter-hegemonic. It is more than capable of doing its part to sustain what Gilles Deleuze memorably christened our society of control. Yet there are methodological contiguities between Pearl’s causal revolution and the basic orientation of avowedly counter-hegemonic poststructuralist critique, especially as such critique defines itself, exactly, in contrast to structuralism. And when Pearl lays out his method in The Book of Why, it is, more precisely, a counterfactual turn in statistical reasoning that he highlights as catalyzing his causal revolution. For Pearl, when we re-theorize causation, we reveal characteristic forms of causal relations, patterns that constitute a logical language which can be used to glean truths about causation in instances where we can’t yet know the causal mechanism itself. So far, so structuralist: yet amongst these forms, it is those that employ counterfactual thinking which Pearl highlights as the most important tools for a new, or restored, science of causation. Counterfactual thought, for Pearl, opens analytic space for what is contingent, not structurally necessary, and for what is true on the scale of the individual, not on that of the population. Such counterfactual thought depends on a faculty of imagination that mediates artfully, not mechanically, between the particular and the general, and that is conversely equipped to analyze mediation itself as a nexus of causes and effects. It is, to switch poststructuralist idioms, something of an analytic of the trace. (13)

When, in The Book of Why, Pearl invites his reader to join his “causal revolution” by ascending “the ladder of causation,” literary critics might understand Pearl to be inviting us to our own party. (14) At the bottom rung of Pearl’s ladder, we understand causation by making associations. We could call this the level of a cartoon Humeanism, of constant conjunction that never rises to an analytically actionable level. (15) We ascend to the middle rung when we use interventions to explore causation; this is the data-gathering approach of a materialist empiricism, of random-controlled trials and such. The true revolution arrives, however, and we reach the top rung, (16) when we reason counterfactually, and juxtapose incommensurate, qualitatively developed models in which causality’s full contingency can be on display. The stress Pearl places on “imagination” as a vector of “revolution” seems all the more remarkable, and romantic, when one remembers that the Romantics fashioned such an idea of imaginative capacity very much in response to the emergence of the machine age, the era which an AI scientist like Pearl aims at once to extend and to transvalue. Artificial intelligence, this transhistorical chiasmus suggests, may have a longue durée just as the Anthropocene does, and be likewise entwined with a reflexive critical consciousness from its start.

So whereas one might have imagined that a calculating logician like Pearl would resemble, not Burke the qualitative materialist, but instead Burke’s foes the quantitative revolutionaries, I want to propose instead that Pearl’s mode of analysis, in its openness to the counterfactual, and moreover and relatedly, in its embrace of the particular and of the mediatory, has a strong affinity with a Burkean historicist empiricism. The common denominator of these thinkers might be said to be a certain imaginative--one might even say, Romantic--historicism. Yet if this is so, I would want to quickly stipulate a few qualifications. This causally-oriented Romantic historicism only represents one facet of Burke’s thought. It also resonates, meanwhile, with the thought of other eighteenth-century thinkers whom we might consider inimical to causal thought, but who actually may not have been--even with the thought of David Hume, who partisans of the causal revolution have tried to reclaim for their camp. Finally, if a causal Romantic historicism exists, it remains notoriously unstable through its subsequent iterations, and its versions too often echo Burke’s own misogyny and rancid elitism, especially when they move from grasping the particular to making claims about the essential and reasserting, on that basis, unjust hierarchies and cruel effects.

We might hope a better version of such Romantic historicism would be our own default poststructuralist, quasi-formalist new historicism. I hope a picture is starting to emerge of how the philosophical turn to counterfactuals on the part of computer scientists might give us purchase on how causation matters to such models in literary and historical studies, whether or not we have realized this is the case. (17) And for examples of such models, we now have a burgeoning “counterfactual turn” in literary studies to draw upon, with echt new historicist Catherine Gallagher’s Telling It Like It Wasn’t the flagship of the flotilla setting this course, and Damian Walford Davies’ collection Counterfactual Romanticism of special interest to its subfield. These works provide plenty of bracing arguments about the role of a whole variety of causal arguments in the cosmologies concerned in the Age of Revolution, arguments in which conceptions of providence, necessity, and contingency all get their due. Gallagher’s argument in this regard is especially exciting, as it pursues the development of what we might consider a vernacular counterfactual poststructuralism from its Enlightenment beginnings up into today’s flourishing genres of alternative reality fiction. Still, much room remains for more intimate engagements with counterfactuals at the level of subjective historical experience, engagements of a sort that might align with Andrew Miller’s explorations of the optative mode; and room also remains for more theoretical engagements with the range of concepts surrounding the causation problematic. (18) One source for the latter is Riyukta Raghunath’s patient engagement with the possible worlds theories of the narrative theorist Marie-Laure Ryan and of the philosopher David Lewis, who bridges, perhaps better than any other single figure from late twentieth century thought, the span between the logical methods of computer science and the imaginative methods of science fiction and of the counterfactual genres Gallagher explores. For the former, more intimate mode of exploring counterfactuals, finally, we have as an exemplar Emily Rohrbach’s patient unraveling, in a recent essay, of the psychological drama of Jane Austen’s Emma as a multilayered counterfactual causal self-analysis.

What does it tell us that the difference between Pearl’s approach and that of literary scholars may not be all that great, and that Pearl’s brand of causal analysis may be more tractable than one might expect for literary and cultural historiography? One thing it might suggest is that the development of arguments and books may be more contiguous with the development of computational artificial intelligence than we might otherwise have thought. Having sketched this contiguity in the criticism in our time, (19) I want to conclude by returning to trace it in eighteenth-century computational theory. Pearl expresses his theories of causal inference by means of a technique of directed acyclic graphs he developed from his work on Bayesian Networks. Now, thanks to Nate Silver’s popularization of Bayesianism, (20) we are familiar with this probabilistic method in which we begin from, and continually recast, the qualitative conjectural judgments called, in popular shorthand, our priors. What would it mean to recur once more to assimilate a longue duree of causal thought that this time includes the originating Bayes of Bayesianism, the eighteenth century dissenting minister Thomas Bayes, in our graph of intellectual-historical cause and effect?

I have waited until my conclusion to pose this question because I am quite conscious I may well not be the best person at this table to give an answer; my favorite pages on Bayes in historical context are those Jesse Molesworth devotes to him in Chance and the Eighteenth Century Novel. “From a cultural perspective--from a conceptual perspective,” Jesse writes, “Bayesian analysis seems the natural product of the mid-[eighteenth-]century mind’s fascination with inductive thinking.” But is this the sort of inductive thinking that validates causation, or the sort that seeks to bracket it? (21) To gain a broader perspective on this question, we can group Bayes with figures like David Hume, whose Enquiries are said to have inspired Bayes’s theorizing, and Richard Price, who rescued Bayes’s essay on probability from his papers after his death, finished it and published it through the Royal Society. Hume is often said to have cast causal analysis into disrepute with his concept of constant conjunction; Bayes’s theory of probability has often seemed like a scheme to bypass direct causal analysis in favor of a calculus of chances; Price, as an original moral intuitionist, foregoes the patient study of outcomes that Burke recommends as political science. Yet from another point of view, these thinkers all validate causal analysis, and in so doing cultivate an art of the counterfactual and of the particular that laid the foundation for the imaginative theater in which Pearl’s computational practice would have its beginnings: the literature of Romantic historicism. (22) One wishes one could consult Burke on the question, especially since Bayes’s friend Price, the moral philosopher who rescued his manuscript on probability, is of course the same Richard Price whose sermons on Revolution from “the Old Jewry” so excised Burke as to prompt him to issue his Reflections and to inspire him to predict, in a paranoid but uncannily prescient tableau, the fate of the French royal family. Read together, these thinkers present a monumental intellectual-historical tangle of revolution and counter-revolution, associationism, intuitionism, and causal analysis, a tangle one imagines could only be properly unraveled by an especially learned, and witty, work of counterfactual historical fiction. And [to echo Jesse Molesworth’s provocative observation in his introduction to this panel] not only is there no mention of Jacobite alternative history in Gallagher’s account of the eighteenth century, there is no mention in her entire book of either Walter Scott or Georg Lukacs. Thank you. (23)

Additional Bibliography to pursue:

Seb Posky

Amos Funkenstein

Lorraine Daston on C19 probablility (already recalled?)

No comments here
Why not start the discussion?