Williamson On Overfitting: A Note on Metaphysics, Logic and Epistemology in Philosophy of Education (13)

Timothy Williamson has recently been talking about what he calls a methodological pathology in natural and social science. The pathology is overfitting. The term is borrowed from statistical learning, but the analogy is not superficial. In that setting, a model is made increasingly complex in order to capture a set of data points. Each additional parameter improves the fit to the observed cases. Yet beyond a certain point the model ceases to track the underlying structure and instead begins to track noise, artefact, and accidental correlation. It looks more accurate, more comprehensive, more responsive, yet it has in fact lost its grip on what it is supposed to be modelling.

Williamson argues that something closely analogous happens in philosophical work when a concept, especially a morally or politically charged one, becomes the site of progressive theoretical accumulation. I think there are signs of this within philosophy of education. A term such as hope, inclusion, agency, or decolonisation begins with a relatively tractable role, often heuristic, often corrective. It highlights something missing or misdescribed. But then successive layers are added. One theorist introduces a critique of instrumental rationality, another brings in affect theory, another situates the concept within a genealogy of colonial epistemology, another contrasts it with a supposed Western metaphysics, another aligns it with a set of non-Western traditions, another adds psychoanalytic depth, another invokes ecological relationality, another reframes it through critical race theory, another through disability studies, and so on. Each move has some local plausibility. Each seems to capture something relevant. The concept becomes richer, more sophisticated, more attuned to complexity. Yet there is a danger that what is being produced is not depth but overfitting.

Williamson’s framework gives a precise way of diagnosing what has gone wrong. The philosopher is no longer tracking a well-disciplined object but is instead responding to a shifting field of relevance cues. Each added theoretical layer feels justified because it connects to something that seems pertinent to the concept’s normative horizon. If hope is about the future, then surely political theory is relevant. If it concerns oppressed groups, then postcolonial critique is relevant. If it concerns subjectivity, then psychoanalysis is relevant. If it concerns relationality, then ecological or non-Western traditions are relevant. The mind, and the discourse, expands the concept by absorbing everything that appears relevant to answering the question, what is hope in education. But the very ease with which these connections are made is what should raise suspicion. Maybe the concept is not being sharpened. Maybe it is being stretched.

I went to a conference of philosophy of education recently and felt that the invocation of “Western philosophy” versus “non-Western philosophy” was functioning exactly like a heuristic in Williamson’s sense. It did not emerge from a careful analysis of kinds, essences, or even coherent traditions. It operated instead as a broad sorting device. As such, “Western philosophy” becomes a placeholder for whatever is taken to be individualistic, rationalist, abstract, oppressive, or insufficiently relational. “Non-Western philosophy” becomes a placeholder for whatever is taken to be communal, embodied, holistic, or ethically superior in the relevant respect. These are not carefully delimited categories. They are relevance-driven constructs. They gather together disparate elements under morally and rhetorically charged headings. Once these placeholders are in play, any concept under discussion, let's say "hope", becomes overfitted to them. Hope is no longer analysed in terms of what it is, what kinds of educational states or practices it might correspond to, what grounds it, how it differs from expectation, aspiration, or resilience. Instead, it is redescribed through an ever-expanding set of associations. Hope becomes anti-Western, decolonial, relational, embodied, resistant, affective, communal. Each addition makes the concept appear richer. But from a Williamsonian perspective, the additions may simply reflect the operation of a broad and unexamined relevance filter. The concept is being tuned to fit a complex ideological dataset rather than to track a stable educational phenomenon.

I think this helps explain a pervasive distinction in philosophy. The analytic continental divide persists not because it maps a clear and stable philosophical distinction, but because it has become a convenient heuristic for organising discourse, identities, conferences, journals, and intellectual lineages. It bundles together styles, topics, methods, and historical affiliations in ways that are often inconsistent and overlapping. Yet it feels real because it has been repeatedly invoked and socially reinforced. The same risk attends newer educational buzzwords. Once a term becomes embedded in funding calls, policy frameworks, conference themes, and institutional missions, its apparent reality is sustained by practice rather than by conceptual discipline.

What is particularly damaging about this form of overfitting is that it gives the illusion of depth while actually blocking analysis. The more a concept is loaded with theoretical references, the harder it becomes to ask simple but crucial questions. What is the minimal structure required for hope in an educational context? Is it a disposition of a learner, a feature of a curriculum, a property of an institutional arrangement, or a relation between agents and possible futures? What distinguishes hope from optimism, expectation, or desire? Under what conditions is it present or absent? What grounds it? What would count as its failure? These are the kinds of questions that Fine’s machinery would encourage us to ask. But in an overfitted discourse, they can seem almost naive, because the concept has already been surrounded by a cloud of associations that makes it appear too rich to be pinned down.

There is also a political cost. When a concept becomes overfitted in this way, disagreement becomes difficult to articulate. To question any part of the conceptual bundle can be taken as rejecting the entire moral orientation with which it is associated. To question whether “Western philosophy” is a coherent category can be heard as defending oppression. To question whether “non-Western philosophy” forms a unified alternative can be heard as dismissing marginalised traditions. (If this example doesn't work for you, try asking similar questions about trans people.) The heuristic categories thus acquire a protective function. They shield the concept from the kind of analytic scrutiny that would reveal its internal incoherence.

Williamson’s approach does not deny that there are real differences among philosophical traditions, nor that there are important critiques of dominant frameworks. What it insists on is that we do not take the felt relevance of these differences as evidence that they can be bundled into a single explanatory contrast. The fact that certain features of what is called Western philosophy are relevant to certain educational problems does not entail that “Western philosophy” is a unified object. Nor does the fact that certain non-Western traditions offer resources for thinking about relationality entail that “non-Western philosophy” forms a coherent alternative. These are precisely the kinds of moves where a relevance heuristic masquerades as a metaphysical or conceptual distinction.

The same point can be made in a more Finean register. If we ask what kind of entity “hope in education” is supposed to be, and what grounds it, and what generated forms instantiate it, the overfitted version of the concept becomes very hard to sustain. The more elements are added, the less clear it becomes what would count as the underlying structure. Is hope grounded in temporal orientation toward future goods? In affective states? In institutional arrangements that make certain futures possible? In narratives that sustain agency? In cultural frameworks that shape expectation? In political conditions that enable or block possibility? All of these may be relevant, but unless we can specify how they relate, whether one grounds another, whether they jointly generate a determinate educational state, or whether they simply co-occur in some valued cases, the concept remains diffuse.

A disciplined analysis might begin by distinguishing several different but related notions. One might separate hope as a learner’s disposition from hope as a property of an educational environment. One might distinguish epistemic hope, the sense that understanding is possible, from practical hope, the sense that one’s efforts can lead somewhere, from social hope, the sense that one’s place in a community is secure, from political hope, the sense that structures can change. One might then ask what grounds each, what conditions are necessary, what structures generate them, and whether they are compossible. That would be a Finean reconstruction. It would likely produce a more complex map, but one in which the complexity is articulated rather than accumulated.

What overfitting does instead is to collapse these distinctions into a single thickened term and then attach to it a series of theoretical affiliations that reinforce its normative appeal. The result is a concept that is rhetorically powerful but analytically weak. It can do work in motivating reform or signalling alignment, but it struggles to support precise judgement or explanation.The broader lesson is that philosophy of education, perhaps more than some other areas, is vulnerable to this dynamic because of its dual commitment to critique and to normative vision. It must engage with large moral and political questions, and it must often respond quickly to changing social concerns. Under those pressures, heuristic concepts are indispensable. But the Williamson Fine pivot shows that without a countervailing commitment to ontological and conceptual discipline, these heuristics can harden into pseudo-objects. They begin to look like deep structures when they are in fact products of accumulated relevance.

To resist this, one does not need to abandon ambition or sensitivity to complexity. One needs to separate two activities that are often run together. One is the expansion of a concept to capture a wide range of concerns. The other is the analysis of whether those concerns actually belong to a single thing. The first is often necessary for critique and mobilisation. The second is necessary for understanding. When they are conflated, overfitting becomes almost inevitable. Seen in this light, the proliferation of theorists and references around a concept is not in itself a sign of depth. It may be a sign that the concept is being asked to do too much. The presence of many frameworks can indicate that no single structure has been identified. The concept is being stabilised by intellectual accumulation rather than by analytic clarity. Williamson’s caution is that we should not mistake that accumulation for evidence that there is a single underlying phenomenon to which all these frameworks are converging. It may instead be that they are all responding to different aspects of a loose and overextended term.

The task, of course, is not to purge philosophy of education of its rich theoretical resources, but to bring those resources to bear more selectively and more critically. Instead of asking how many frameworks can be attached to a concept, one asks which frameworks help identify its structure, its grounds, its limits, and its relations to other concepts. Instead of assuming that broad oppositions such as Western versus non-Western track deep differences, one asks what precise distinctions are being made, and whether they correspond to stable kinds or merely to shifting relevance judgments.

A further symptom of the same overfitting dynamic lies in the way names of major theorists are often deployed within philosophy of education. One encounters passages in which Wittgenstein, Dewey, Levinas, Heidegger, Derrida, Butler, Spivak, or others are invoked through a single phrase or compressed reference, as if the mere citation were sufficient to anchor an argument. A line is quoted, sometimes detached from its argumentative setting, and then made to carry a considerable conceptual load. The name begins to function less as a pointer to a worked-out position and more as a token within a shared discourse. It signals alignment, seriousness, or critical orientation. It also serves as a kind of intellectual shorthand, allowing complex and often contested bodies of thought to be treated as if they offered a unified and readily applicable insight. From a Williamsonian perspective, this is another instance of a relevance heuristic operating unchecked. The thinker’s name becomes associated with a cluster of themes, language games for Wittgenstein, experience and democracy for Dewey, alterity for Levinas, being and disclosure for Heidegger, différance for Derrida, performativity for Butler, subalternity for Spivak etc etc. When an educational problem appears to touch one of these themes, the corresponding name is brought in. The relevance feels immediate. Yet what is often missing is any specification of which claims from that thinker are in play, how they are to be interpreted, what their internal tensions are, and how exactly they connect to the educational case at hand. The invocation is doing classificatory work rather than analytic work. It helps situate the discussion within a recognisable intellectual landscape, but it does not yet establish what is being argued.

The result is a form of conceptual overfitting at the level of authority. Just as a concept like hope becomes overloaded by absorbing multiple frameworks, so too a philosophical reference becomes overloaded by standing in for an entire tradition or set of arguments. A single line from Levinas may be taken to ground a whole ethics of education. A gesture toward Derrida may be treated as sufficient to establish indeterminacy or critique of presence. Wittgenstein may be cited to dissolve problems without showing how the dissolution proceeds. In each case, the name functions as a kind of conceptual amplifier. It enlarges the apparent depth of the claim without necessarily increasing its precision.

Seen through the Finean lens, this practice also obscures the ontological and explanatory questions that should be asked. If one invokes Levinas in a discussion of responsibility in the classroom, what exactly is the educational entity or relation being posited. Is it a generated ethical relation between teacher and student? Is it grounded in asymmetry of authority? Is it a structural feature of pedagogical encounter? Or is the reference merely gesturing toward a moral orientation without specifying its form? Similarly, if one invokes Dewey in relation to experience, what is the status of experience in the educational ontology being proposed? Is it a constitutive element of learning, a condition for meaning, a process that generates understanding? Without such specification, the reference remains at the level of thematic resonance. 

Williamson reminds us that the fact that a thinker’s name feels relevant to a discussion is not evidence that their position has been integrated into the argument. Nor is it evidence that the educational concept under discussion has been clarified. It may simply show that the discourse has learned to associate certain names with certain evaluative or explanatory moves. The invocation then stabilises the concept socially rather than analytically. It becomes harder to question the concept, because it is now backed by canonical authority, even if the connection is thin.

This does not mean that engagement with these thinkers should be abandoned. On the contrary, a more disciplined use of them would deepen educational theory considerably. But that discipline would require moving from citation as signal to citation as argument. It would require specifying which claims are being used, how they bear on the educational case, and how they interact with other elements of the analysis. It would also require recognising that these thinkers themselves often offer multiple, sometimes incompatible, lines of thought, rather than a single unified doctrine that can be easily imported.

In this way, the critique returns to the same underlying point. Whether the issue is a buzzword like inclusion, a master concept like hope, or a canonical name like Heidegger or Dewey, the danger is that educational discourse substitutes accumulation for structure. Concepts and authorities are layered together because they seem relevant, because they carry normative weight, because they resonate with the concerns at hand. But without the kind of pressure exemplified by Williamson, and without the kind of structural articulation exemplified by Fine, that layering can produce the appearance of depth without its substance. The task is to recover a mode of thought in which relevance is tested, structure is specified, and authority is earned through argument rather than invoked through association.