Digital humanities represents an emerging form of the humanities rather than an alternative to traditional humanistic scholarship. New technological approaches should certainly be used to address existing research questions, but more importantly, they can bring into view questions that traditional humanities scholarship has not previously recognized as problems at all. Such questions may have no established empirical precedents to draw upon and may even require a break with conventional assumptions. Digital humanities should therefore be understood as a fallible, incomplete, and updatable mode of inquiry—one better equipped to engage with the future.

Falsifiable assumption

The fundamental cyclical model of traditional humanistic scholarship is the hermeneutic circle—a process in which texts and interpretations continually deepen one another. As interpretations become increasingly coherent, disagreements are gradually resolved. Yet preserving the openness of knowledge requires allowing others to use the same data to falsify existing conclusions. The more sophisticated a model becomes, the more important it is to ask what evidence would cause it to fail; the more internally consistent a research framework is, the more actively counterexamples should be sought. This commitment to falsifiability is a prerequisite for maintaining the vitality of digital humanities. No conclusion derived from data and algorithms should therefore be regarded as an ultimate truth, but rather as a hypothesis with only provisional validity.

By transforming algorithmic black boxes into transparent structures open to questioning and refutation, digital humanities may render some of the questions posed by older paradigms irrelevant altogether. Humanistic scholarship should thus welcome falsification because it signals contact with the genuine boundaries of knowledge.

Incomplete exploration

At the methodological core of digital humanities lie formal systems. When algorithms are used to analyze algorithms themselves, or when formal methods are employed to investigate the limits of formalization, the inquiry enters a G?delian realm of self-reference.

Kurt G?del demonstrated that in any sufficiently powerful and consistent formal system, there necessarily exists a proposition that can be neither proved nor disproved. Any attempt to address the future through a single algorithm will eventually encounter the limits of computability. Meaning, emotion, aesthetic experience, and historical context are inherently resistant to complete formalization. The same text may generate radically different interpretations across historical contexts and readerships. Aesthetic judgment cannot be fully translated into computable feature vectors; formalization will inevitably leave something unresolved and encounter cases that cannot be decided algorithmically.

Measured against the full body of historical texts, any database is necessarily an incomplete sample. The very act of indexing a text already reflects the researcher’s assumptions and limitations. To acknowledge “incompleteness” is to recognize that humanistic understanding is always a work in progress. Such an acknowledgment rejects the idea that knowledge can be enclosed within a self-sufficient system. Instead, it calls for sustained attention to the boundaries of formalization and to the tension between the computable and the uncomputable, allowing digital humanities to develop as a genuinely open, future-oriented form of inquiry.

Updatable cycle

The basic logic of Bayes’ theorem is to begin with a hypothesis, allow the evidence to speak, and revise that hypothesis as new evidence emerges. The initial hypothesis may be wrong, and some degree of error at the outset is unavoidable. What matters is the speed and magnitude of revision: Each new piece of evidence requires the relative weights assigned to different possibilities to be recalibrated. Humanistic data often present the dual challenges of small sample sizes and high dimensionality. Some datasets may contain only a few dozen texts, while each one may involve thousands of feature dimensions.

The “priors” of humanistic scholars often derive from beliefs and experience, scholarly traditions, and historical contexts that are difficult to quantify. Digital humanities seeks to make such tacit knowledge explicit. Not all evidence is equivalent: Evidence that merely confirms existing preferences must be distinguished from evidence that genuinely challenges a hypothesis. At the same time, conclusions should be probabilistic, carry degrees of confidence, and remain open to revision. Once the evidence becomes sufficiently compelling, a fundamental change in belief is not a betrayal but a rational response. Updatability means that scholarly findings are no longer one-time deliverables, but objects subject to continuous revision and maintenance.

By introducing mechanisms for detecting errors, acknowledging limitations, and self-correction, digital humanities incorporates its own epistemological limits into its methodology. Rather than accepting the fixed boundaries implied by “this is how it has always been” or “this is where it ends,” digital humanities operates within an as-yet-unnamed territory between computation and interpretation, probability and narrative, and the decidable and the undecidable—one in which every line of code and every annotation remains open to revision.

Li Feiyue is a professor from the School of Humanities at Tsinghua University.