The Epistemology of the Digital: Rethinking Evidence and Argument in the Computational Age

The advent of computational methods in the humanities is not merely a technical augmentation of existing practices but a fundamental reconfiguration of the epistemological foundations upon which humanistic knowledge has been built. For centuries, the humanities have operated within a hermeneutic paradigm that privileges close reading, contextual interpretation, and the sovereign subject of understanding. The digital turn, however, introduces a new regime of evidence, causality, and argumentation—one that is shaped by datafication, algorithmic processing, and statistical reasoning. This chapter examines this transformative shift, analyzing how computational approaches challenge and reconfigure core concepts of humanistic inquiry. We will explore the tensions between distant viewing and close reading, interrogate the status of probabilistic models as forms of argument, and articulate the new rhetorics required to validate insights derived from computational methods. Engaging critically with science and technology studies (STS), we frame these developments not as neutral progress but as contested terrain where values, power, and epistemic authority are at stake. The ensuing technical chapters will build upon this epistemological groundwork, demonstrating both the possibilities and the perils of algorithmic humanities.

1.1 The Hermeneutic Foundation: Humanistic Epistemology Before Computation

The Western humanistic tradition has long been anchored in a hermeneutic epistemology that views understanding as a dialogical, context-sensitive, and ultimately subjective process. From Friedrich Schleiermacher’s universal hermeneutics to Hans-Georg Gadamer’s philosophical hermeneutics, the core assumption has been that meaning emerges through a circular movement between part and whole, between the interpreter’s horizon and that of the text[^1]. This approach is inherently particularistic: it seeks to recover the uniqueness of a work, its historical embeddedness, and the intentionality (or at least the effect) of its production. Evidence in this paradigm consists of textual details, intertextual allusions, linguistic nuances, and historical documentation—all of which require a discerning, trained human eye to perceive and interpret. The scholar’s authority rests on their ability to perform such close reading, to construct a coherent narrative that accounts for the complexity and ambiguity of the humanistic object.

The twentieth century saw this hermeneutic tradition both solidified and challenged. The New Critical emphasis on the autonomy of the text and the organic unity of the literary work shifted attention away from authorial intention and historical context toward the internal dynamics of the text itself[^2]. Yet even New Criticism retained the primacy of close reading as its methodological core. Later, reader-response theory and poststructuralism further destabilized the notion of a fixed meaning, emphasizing the role of the reader and the endless play of signification[^3]. Nevertheless, across these theoretical shifts, the basic unit of analysis remained the individual text (or artifact), and the primary mode of engagement remained intensive, qualitative interpretation. The humanist’s toolkit consisted of exegesis, comparative analysis, and historical contextualization—all practices that assume a bounded, comprehensible object of study.

This hermeneutic epistemology carries with it specific conceptions of evidence, causality, and argument. Evidence is not raw data but carefully selected instances that illuminate a broader interpretive claim. Causality is often narrative and teleological: events are linked through motives, ideologies, or historical forces that require empathetic understanding. Argumentation is rhetorical and dialectical: it seeks to persuade through logical coherence, textual support, and the plausibility of its interpretive framework. Truth claims are provisional and contextual, always open to revision in light of new readings or historical discoveries. Objectivity, in this view, is not a matter of value-neutral measurement but of critical self-awareness, methodological rigor, and engagement with alternative interpretations[^4].

1.2 The Computational Turn: A New Epistemological Regime

The introduction of computational methods into humanistic research—whether through stylometry, topic modeling, network analysis, or geographic information systems—represents a departure from this hermeneutic foundation. At its core is the process of datafication: the transformation of humanistic phenomena (texts, images, artifacts, events) into quantifiable, machine-readable data[^5]. This act is not neutral; it involves decisions about what to measure, how to represent it, and what to exclude. A novel becomes a vector of word frequencies; a painting becomes a matrix of pixel values; a historical event becomes a node in a network. The richness, ambiguity, and materiality of the original object are inevitably reduced to the parameters of the computational model.

This datafication enables a shift from intensive to extensive analysis. Where close reading delves deeply into a small number of texts, computational methods can survey vast corpora—millions of books, thousands of paintings, decades of newspapers—identifying patterns that would be invisible at the micro level. This “distant viewing” or “macroanalysis,” as Franco Moretti famously advocated, promises a new kind of evidence: not the telling detail but the statistical trend; not the isolated motif but the long-term evolution of literary genres[^6]. Evidence thus becomes aggregate, probabilistic, and comparative. Causality, however, becomes more elusive: correlations abound, but the mechanisms linking variables remain opaque without additional historical or theoretical work. Argumentation shifts from the construction of a persuasive narrative to the validation of a model—its accuracy, robustness, and generalizability.

This new regime is underpinned by several epistemological commitments that often remain unstated. First, there is an implicit faith in the objectivity of numbers: the belief that quantified data offers a more transparent, less biased window onto reality than subjective interpretation. Second, there is a performative understanding of knowledge: insights are not discovered but enacted through the interaction of data, algorithm, and researcher[^7]. Third, there is a pragmatist orientation toward utility: the value of a computational finding lies in its ability to predict, classify, or reveal hidden structures, rather than in its capacity to capture meaning in a humanistic sense. These commitments align humanistic research more closely with the epistemic cultures of the natural and social sciences, particularly with data science and computational social science[^8].

Yet this alignment is fraught with tension. Humanistic questions often concern meaning, value, and representation—domains that resist straightforward quantification. The computational turn thus forces a re-examination of what counts as legitimate evidence and valid argument in the humanities. It also raises critical questions about the role of the humanist: Are we to become data scientists, or must we develop a new hybrid competence that retains critical distance from computational outputs? The following sections unpack these tensions in detail.

1.3 Evidence Transformed: From Textual Detail to Statistical Significance

The most immediate impact of computational methods is on the concept of evidence. In traditional humanistic research, evidence is selective and interpretive: a scholar points to a specific passage, image, or document to support a claim, and the strength of that evidence lies in its pertinence and richness. In computational research, evidence often consists of aggregated data patterns: a spike in word frequency, a cluster in a network, a correlation between variables. The unit of evidence shifts from th