Prediction does not appear, at first, to be something that a sociologist, or really any analyst of anything, can safely ascribe to those (or that) which they are studying without running afoul of about a thousand different stringent rules that define how probability can be used for the purposes of generating knowledge. If we follow the likes of Ian Hacking (1975) and Lorraine Daston (1988) (among others), then “modern fact-making” has a lot to do with ways of using probability, especially for the purposes of making predictions. To the degree that this transforms probability into prediction, as referring to the epistemic practices that analysts use to generate a knowledge claim, this usage actually places limits on what probability can mean, how prediction can be used and where we might find it. If we don’t have certain epistemic practices (e.g. a nice regression analysis) then we can’t say that prediction is occurring anywhere if we are not doing it ourselves.
As Hacking and Daston indicate, however, for probability to be limited almost entirely to epistemic practices in this sense would appear strange to those who can stake any sort of claim to having “discovered” probability, especially Blaise Pascal. He, for one, did not understand probability to be limited to efforts at making predictions for the purposes of knowledge. For Pascal, probability had direct analogues in lived experience (without calculation) in the form of senses of risk and high stakes, and the perceived fairness of outcome, particularly in games of chance. If this seems unusual to us now, given the strictures we place on probability and prediction, these points are far less unusual for what is fast appearing as a major paradigm in cognitive science, namely predictive processing (see Clark 2013; Friston 2009; Wiese and Metzinger 2017; Williams 2018; Hohwy 2020).
To put it simply, predictive processing makes prediction the primary function of the brain. The brain evolved to allow for the optimal form of engagement with a contingent and probabilistic environment that is never in a steady state. Given that our grey matter is locked away inside a thick layer of protective bone (e.g. the skull), it has no direct way of perceiving or “understanding” what is coming at it from the outside world. What it does have are the senses, which themselves evolved to gather information about that environment. Predictive processing says, in essence, that the brain can have “knowledge” of its environment by building the equivalent of a model and using it to constantly generating predictions about what the incoming sensory information could be. This works in a continuous way, both at the level of the neuron and synapse, and at the level of the whole organism. The brain does not “represent” what it is dealing with, then, but it uses associations, co-occurrences, tendencies and rhythms to predict what it is dealing with.
All of this is contingent on making the equivalent of constant, future-oriented but past-deriving, best guesses. When those guesses are wrong, this generates error, which forms the content of our perceptions. In other words, what we perceive and consciously attend to is the leftover error of our generative models and their predictions of our sensory input. When those guesses are right, by contrast, we don’t have perceptual content because there is no error. The generative models we build are themselves multi-tiered, and the predictions they make work at several different levels of composition. A full explanation of predictive processing far exceeds the limits of this post. But this, in particular, is worth mentioning because it means that a generative model is not static or unchanging. Quite the contrary, generative models constantly change (at some compositional level) in order to better ensure prediction error minimization.
Some of these points will probably not sound that unusual. The relationship between minimized perceptual content and action is commonly referred to in discussions of embodiment and moral intuition, for example. What probably sounds very unusual, however, is the central role given here to prediction.
As mentioned, prediction has been essentially cordoned off in the protected sphere of knowledge, to be used only by specialists wielding specialist tools and training. While it can be done by the folk, we (the analysts) love to point out how they do it poorly. On the off chance they happen to predict correctly (e.g. gambling on the long shot), this is celebrated as the exception that proves the rule. After all, the folk do not have our epistemic practices or training. All they have is their (subjective) experience and biases. In fact, brandishing those presumably bad at predicting by those with increasingly sophisticated techniques to make predictions on increasingly large datasets has become par for the course in the era of “analytics” (Hofman, Sharma and Watts 2017), and this particular symbolic power is now wielded quite overtly in a variety of fields (like baseball). Thus, to take prediction away from action could have, all along, been just another way of saying that because we (the analysts) predict and they (the folk) do not predict or do so poorly, they need us.
But is this the case? Predictive processing poses a serious question to this assumption and, with it, the role that prediction plays in making sociological knowledge different from folk knowledge. There is also a bit of history worth mentioning. The assumption that prediction plays only a negligible part in action, while other things like values and beliefs play a big part, comes from Talcott Parsons, who explicitly set out to marginalize prediction (1937: 64). Sociologists are rightfully in the mood of poo-pooing Parsons and have been for quite some time; but any proposal to put prediction into action remains just as heretical today as it did to Parsons in the 1930s. As one of his major points about action, the presumption that prediction can play no direct or significant role in action has still not been revisited let alone revised.
The purpose of this post is simply to sketch out the suggestion that we can even do this (e.g. put prediction into action) without falling over our feet and retreating sheepishly to the safety of the domain the Parsons carved out for us should we ever wish to talk about “action” again. Far be it from me to attempt to do this on my own. So for the purposes of illustration, a few pages from John Dewey’s Logic: The Theory of Inquiry (1938: 101-116) (and few from Human Nature and Conduct [1922]) will be enlisted for the task. I will argue that, in these pages, which are themselves famous because in them Dewey gives specific proposals about the process or stages of inquiry, Dewey does put prediction into action, and he does so in a way that does not seem that controversial; though, for any legitimate contemporary meaning of “prediction,” these are heretical claims.
For Dewey, in contrast to Parsons, the action situation is not neatly parsed into the “objective state of affairs” that could be described with scientific precision by an external observer (and for which prediction is appropriate) and the “subjective point of view” of the actor (for which, by implication, prediction does not apply, lest we “squeeze out” the creative, voluntaristic element). Instead, the “state of affairs” is, according to Dewey (1922, p. 100ff), irreducibly composed of an entanglement of both objective and subjective elements. The very act of perception of a given state of affairs on the part of the actor introduces such a subjective element (for Parsons perception was not necessarily part of the subjective element of the action schema).
Perception is not just purely spectatorial or contemplationist, then, but serves as the “initial stage” in a dynamic action cycle. Perception is for something, and this something is anticipation and prediction. Thus, “the terminal outcome when anticipated (as it is when a moving cause of affairs is perceived) becomes an end-in-view, an aim, purpose, a prediction usable as a plan in shaping the course of events” (Dewey 1922:101, italics added). In a stronger sense, for Dewey perceptions are predictions, which in their turn are ends-in-view. Perceptions are “projections of possible consequences; they are ends-in-view. The in-viewness of ends is as much conditioning by antecedent natural conditions as is perception of contemporary objets external to the organism, trees and stones or whatever” (102).
For Dewey (1938), this can extend even further into what arguably remains his most influential contribution to pragmatist thought: the process of inquiry, as it “enters into every aspect of every area” of life (101). Inquiry, as Dewey defines it, is the “controlled or directed transformation of an indeterminate situation into one that is so determinate in its constituent distinctions and relations as to convert the elements of the original situation into a unified whole” (104-105). This filters into all subsequent understandings of pragmatist problem-solving.
The “indeterminate situation” (105) that provides antecedent conditions for inquiry is constituted by doubt, but this is not a purely subjective state (“in us”). Doubt refers to our placement in a situation that is doubtful because we cannot respond to it as we are accustomed: “the particular quality of what pervades the given materials, constituting them a situation … is a unique doubtfulness which makes that situation to be just and only the situation it is” (105). Specifically this means that we cannot form ends-in-view with respect to the situation, though we can “[respond] to it … [in] blind and wild overt activities.” As Dewey stresses, “it is the situation that has these traits,” which means that we are simply a part of the situation in being doubtful; one part of the total configuration. To simply “change our mind” with respect to the doubtful situation is hardly enough to change it, though with any indeterminate situation, we might respond by carrying through a “withdrawal from reality.” The only thing that will really be effective, however, is what Dewey calls a “restoration of integration” in which the situation changes as our situation within it changes (e.g. as we change) (106).
Underlying Dewey’s proposals, then, is a kind of cognitive mechanism, which he does not label outright, but which, likewise, rests on prediction, and on which the stages of inquiry itself appear to rest. For Dewey (107-108), it is possible to remain in the doubtful situation forever, particularly should you find an effective means of “withdrawing from reality.” The next stage in the process of inquiry will only occur through a change in “cognitive operations,” specifically what Dewey labels “the institution of the problem … The first result of evocation of inquiry is that the situation is taken, adjudged, to be problematic. To see that a situation requires inquiry is the initial step in inquiry” (107). But to take this step, as Dewey implies, requires a change in the manner of prediction, and in a not dissimilar sense as a roughly equivalent mechanism identified by predictive processing.
If the indeterminate situation does not allow for perceptions as “ends-in-view,” then in the problematic situation the actor (e.g. “the interpretant”) changes because, in the situation, she is now characterized by an explicit representation: “without a problem, there is blind grasping in the dark.” This representation is needed as a change in cognition, but only as a mediating and not a permanent state. But the constant in this process, that allows representation to appear now explicitly and then only to disappear later on, can only be successive forms of prediction that, in Dewey’s terms, is trying to obtain an end-in-view. In other words, the explicit representation of “problem” itself presupposes a prediction about error. More generally, we are part of a problematic situation because we predict that it should go one way and it does not, and then we anticipate what would be required to minimize that error, which then forms the basis for future action. In almost a directly analogous sense, predictive processing refers to this as “active inference.”
Hence, what follows this (“the determination of a problem-situation”) is subsequently characterized by the generation of “ideas” as part of the inherently progressive nature of inquiry along the lines of continuous prediction or forward-searching (e.g. guessing): “The statement of a problematic situation in terms of a problem has no meaning save as the problem instituted has, in the very terms of its statement, reference to a possible solution” (108). Put differently, the one (problem) never occurs without the other (solution); we actively infer solutions because we have problems. Dewey (110-111) uses this to critique all prior conceptions of “ideas” in a western philosophical tradition (empiricists, rationalists and Kantians) for not seeing how perceptions and ideas function correlatively rather than separately:
Observations of facts and suggested meanings or ideas arise and develop in correspondence with each other. The more the facts of the case come to light in consequence of their being subjected to observation, the clearer and more pertinent become the conceptions of the way the problem constituted by these facts is to be dealt with. On the other side, the clearer the idea, the more definite … become the operations of observation and of execution that must be performed in order to resolve the situation (109).
Ideas are not removed from the situation, or entirely defined by the situation. Rather, the most important thing about them is that they have a direction in relation to the situation. But this only works if they suggest a forward-facing (temporally speaking) cognitive mechanism, which again seems perfectly analogous to a predictive function that is trying (slowly) to minimize error. Dewey seeks to redeem the role of “suggestions” (which have “received scant courtesy in logical theory”) by giving them not the diminished importance of half-completed ideas, but elevating them to “the primary stuff of logical ideas.” In this sense, suggestions demonstrate how “perceptual and conceptual materials are instituted in functional correlativity to each other in such a manner that the former locates and describes the problem which the latter represents a method of solution” (111; emphasis added).
To “reason,” then, means to examine the meaning of ideas according to their simultaneous statement of problem and solution (e.g. “relationally”). For Dewey, this process involves “operating with symbols (constituting propositions)… in the sense of ratiocination and rational discourse.” If a suggested meaning is “immediately accepted,” then the inquiry will end prematurely. Full reasoning consists of a kind of “check upon immediate acceptance [as] the examination of … the meaning in question” according to what it “implies in relation to other meanings in the system of which it is a member” (111). By “meaning” Dewey refers to symbols in a semiotic sense or the connection of sign and object in a non-problematic or habitual way. This therefore opens those habitual associations up to transformation as the situation becomes more determinate. Dewey also emphasizes how symbols perform the semiotic function of “fact-meanings.” The process of inquiry subjects these connections to “ideas [as] operational in that they instigate and direct further operations of observation; they are proposals and plans for acting upon existing conditions to bring new facts to light and to organize all the selected facts into a coherent whole” (112-113). The process remains forward-facing, which means that there can be “trial facts” that can be taken on-board with a certain provisionality: “they are tested and ‘proved’ with respect to their evidential function.” Ideas and facts, then, become “operative” in the process of inquiry (problem-solving) “to the degree in which they are connected with experiment” (114). Again, all of this presupposes that forward momentum, or searching, appears to be fueled by advancing and constant prediction.
Thus, for Dewey, the transformation of the situation into “determinate” involves a change of “symbols” in the form of habitual associations (sign to object) which themselves always remain provisional and never fully determinate (114-115). This is what alters our “self” (interpretant) within the situation as no longer in a doubtful state, and replaces this with what we might call a “confident” state as signifying a kind of assurance of action in relation to the situation.
Thus, having passed through the stages of inquiry, and with new habitual associations, we are now predicting it well within the continuous flow of action. In Dewey’s terms, problem and solution effectively merge at the end of inquiry, and the forward-facing search ends. But we can translate the folk terms that Dewey uses here almost directly into the more technical terms that form the basis of predictive processing: the problem or trial-situation ends with the erasure of prediction error by a change in the generative model, such that the tiered coding of sensory input will generate the perceptions that the generative model expects. X is now Y in a non-problematic way, which for Dewey means that it becomes a “symbol” as a connection that is now habitual (see also Peirce CP 2: 234). Inquiry in “common sense” and inquiry in science are not different, according to Dewey, they simply involve differences in problems. For common sense, problems appear from symbols as the habitual culture of groups (115-116).
This can lead us to make an even more radical claim: prediction in action and prediction in sociology are also not different; they simply involve differences in problems between those that occur in the continuous course of action, and those that are deliberately manufactured for the purposes of staging trials and leveraging them in order to make knowledge claims. Shared generative models also appear among actually-existing groups that make similar predictions, perceive similar things based on similar error, make similar active inferences, and therefore “solve problems” in ways that have a family resemblance.
It seems then, without too much presumptuousness, we can take Dewey’s original definition of inquiry and retranslate it into its implied cognitive terms:
The controlled or directed transformation of an indeterminate situation into one that is so determinate in its constituent distinctions and relations as to convert the elements of the original situation into a unified whole (Dewey 1938: 104-105).
We can translate this into a general statement about problem-solving as follows
The higher order transformation of a situation with lots of prediction error into a generative model that is able to convert the elements of the original situation into a predictable whole.
A follow-up post will discuss the broader significance of this translation in relation to pragmatist theories of action.
References
Clark, Andy. 2013. “Whatever next? Predictive Brains, Situated Agents, and the Future of Cognitive Science.” The Behavioral and Brain Sciences 36(3):181–204.
Daston, Lorraine. 1988. Classical Probability in the Enlightenment. Princeton University Press.
Dewey, John. 1938. Logic: The Theory of Inquiry. New York: Holt, Reinhart and Winston.
Dewey, John. 1922. Human Nature and Conduct. New York: Henry Holt.
Friston, Karl. 2009. “The Free-Energy Principle: A Rough Guide to the Brain?” Trends in Cognitive Sciences 13:293–301.
Hacking, Ian. 1975. The Emergence of Probability: A Philosophical Study of Early Ideas about Probability, Induction and Statistical Inference. Cambridge University Press.
Hofman, Jake M., Amit Sharma, and Duncan J. Watts. 2017. “Prediction and Explanation in Social Systems.” Science 355(6324):486–88.
Hohwy, Jakob. 2020. “New directions in predictive processing.” Mind and Language 35: 209-223.
Parsons, Talcott. 1937. The Structure of Social Action. New York: Free Press.
Wiese, Wanja, and Thomas Metzinger. 2017. “Vanilla PP for Philosophers: A Primer on Predictive Processing.” in Philosophy and Predictive Processing.
Williams, Daniel. 2018. “Pragmatism and the Predictive Mind.” Phenomenology and the Cognitive Sciences 17:835–59.
Pingback: Habit as Prediction – Culture, Cognition, and Action (culturecog)
Pingback: What Is Predictive Processing? – Culture, Cognition, and Action (culturecog)
Pingback: Toward a Theory of Habitus Breakage – Culture, Cognition, and Action (culturecog)