Towards Critical Computational Humanities

In this blog post, I would like to argue for critical computational humanities. Computational humanities has many great advantages and promises. One of them is definitely the possibility of unlocking historical sources, using dazzling, cutting-edge new methods to look at really old sources and uncover what’s hidden within them. That’s certainly one of the fascinations the field has always held for me. Computational methods help us see things that would otherwise remain invisible. They allow us to analyse material at scales that would otherwise not be feasible, or so our arguments typically go.

Cite as: Sarah Lang, Towards Critical Computational Humanities, in: LaTeX Ninja’ing and the Digital Humanities (blog), October 8th 2026. https://latex-ninja.com/2026/10/08/towards-critical-computational-humanities/

But computation comes with strings attached. And not just computation itself, but also the methods associated with computational humanities and the whole tradition that has developed around them. Recently, in the introduction to our special issue on Computational Methods for Histories of Alchemy and Chemistry, we argued that these methods open up new epistemic horizons, but also bring new epistemic responsibilities. They give us exciting new possibilities, but they also come with problems: environmental costs, labour conditions, the opacity of models, a lack of explainability and accountability, and the corporate infrastructures we inadvertently support by using these technologies.

Meanwhile, I feel that computational humanities has largely been receiving technological developments rather than stepping forward to help shape the conversation. We have skills that are highly relevant to everything going on in the world today with AI and related technologies. So why are we just reacting? Why are we passively receiving this onslaught of new technology, trying to keep up with it rather than leading?

This is where my idea of critical computational humanities comes in. I think this could provide a paradigm that helps us do just that.

From Critical Digital Humanities to Critical Computational Humanities

The genealogy of this idea lies, of course, in critical digital humanities, which is by now quite well established. We’ve even recently renamed our DHd working group Critical Digital Humanities. So, by calling this Critical Computational Humanities, my goal is certainly not to denigrate critical digital humanities in any way. As I’ve argued elsewhere (in that aforementioned special issue), computational humanities cannot exist without digital humanities. And it will be important that critical computational humanities does not somehow claim a superior position to the “merely digital” either. In fact, I think that critical could help us bridge the divide between the digital and computational movements.

Critical digital humanities was strongly advanced, even pioneered, by David Berry. And what’s important here is that critical, in the sense of critical theory, does not simply mean refusing something or identifying problems without offering solutions. Critique is about diagnosing problems, but also about developing practical ways to address them. That’s really at the heart of what I have been trying to do with much of my recent work (see for example Why Looking More Closely at Our Data Is the Way to Better Research Ethics, Data Feminism as a Challenge for Digital Humanities? etc). I want to put a strong emphasis not just on criticising problems, but also on developing practical strategies through which everyday practitioners of computational methods can make a tangible difference. Because we can have all the great ideals but if they’re not practicable, we’re not going to see much progress. And I think it’s particularly important that we make it possible for as many people as possible to participate because that is what is going to cause the most progress.

What Does Good Computational Research Actually Mean?

So what would such a field do? For one, it would ask what it means to do good computational research. Of course, one aspect is scholarly rigour. And that’s something I feel is already quite strongly emphasised in computational humanities. As I’ve argued elsewhere, it is a rather masculinised field, with particular ideas about rigour, quantification, and what counts as “good scholarship”. But rigour doesn’t just mean being technically sophisticated, explainable, or reproducible. Rigour also means ethics. And this is where I think computational humanities is still lacking.

We all know the E in the CARE principles. But I sometimes have the feeling that nobody really knows what to do with it because it isn’t concrete enough. Compare this with the FAIR principles. They’re straightforward, technical, and practical, which I think has contributed significantly to their successful implementation. CARE, on the other hand, leaves many people confused. Having one of its principles simply be Ethics almost makes it sound as though ethics were something easy to implement. Yet it isn’t. But that’s no reason not to do it. In fact, that’s precisely why I think we should be focusing on it.

Ethics in research cannot simply mean adding an ethics statement at the end of a project or engaging in what we might call ethics washing. It needs to become part of how we understand good computational research in the first place.

Let me give an example. A lot of critical AI scholarship has identified datasets as one of the main sources of harm in AI systems. Of course, there are other problems too, but data is a particularly important one. The problem isn’t necessarily the machine learning model itself. Often, it’s the underlying data. Explainable AI (XAI) can help us understand why a model behaves the way it does, which is certainly useful. But explainability doesn’t fix the underlying problems with the data, nor the work practices, systems of valuation, prestige, and power that surround its production.

In a recently submitted paper, I discuss what I call Schrödinger’s data and how it leads to Schrödinger’s results. What I mean is that, with generative AI, we’re getting increasingly plausible outputs while often struggling to establish the quality of the inputs and the processes behind them.

But that leads us to another question: What is good enough data?

Computational humanities, like computational methods more broadly, depends on data. And historical data is especially challenging. It’s fragmented, has gone through numerous processes of selection and re-selection, and is often heavily skewed. So what can good enough mean in this context? I argue that we need to start by looking much more closely at our datasets: auditing and documenting them, measuring missingness, and charting what’s there and what’s missing. These are practical examples of what critical computational humanities work could look like.

I argue we should approach a critical computational humanities through the lens of three different kinds of diversity.

Three Diversities for Critical Computational Humanities

1. Diversity of Data: What’s Missing?

The first is the diversity of data. This means asking what is missing from our datasets, but also recognising that entire datasets may not exist in the first place. Which people, places, periods, cultures, and disciplines have computational resources available to them, and which do not? This is also an infrastructure question.

Take philosophy, for example. Interestingly, it’s one of the disciplines most actively engaged in conversations with computer science about ethics, yet it has never taken up digital humanities to quite the same extent as some other fields. Something similar can be said about my own discipline, the history of science and knowledge. Things might be a little better there, but there are still significant gaps. And when there are fewer well-prepared datasets, you simply have a much worse starting point for applying computational methods.

Datasets are not evenly distributed across disciplines, periods, cultures, languages, or topics. Yet computational research depends on these datasets existing. So we need to understand not only the gaps and biases within datasets, but also the uneven distribution of datasets themselves. This is why I argue for greater diversity of and in data.

2. Diversity of Methods: Looking Beyond the Streetlight

The second is the diversity of methods.

I think most of us have heard the streetlight metaphor. Imagine you’ve lost your keys at night. There’s a streetlight nearby, so you start looking for them under the light. Not because you have any particular reason to believe that’s where you dropped them, but simply because that’s where you can see. I think something similar is happening with large language models right now. Everybody is using them, and we’re becoming increasingly blind to all the other great methods out there. And there are so many!

There are plenty of computational methods that are extremely useful for humanities research, and many more that have not yet been fully explored. For example, I’m currently beginning to explore the use of computational chemistry algorithms for studying alchemical texts. As far as I’m aware, this simply hasn’t been done before.

There could be a gold mine of untapped methodological potential here. Yet we’re so focused on using the same tools over and over again that we risk overlooking it. It’s a bit like the old saying: when the only tool you have is a hammer, everything looks like a nail. Of course, large language models are quite versatile. Maybe they’re more like the WD-40 of computational methods because they really can help fix things sometimes. But we need to be careful not to lose our methodological diversity.

Rather than simply following whichever methods are fashionable, we should be looking creatively for other approaches, including methods from other disciplines that we can bring into productive dialogue with the humanities. Such combinations might allow us to do something genuinely new and answer questions that are genuinely interesting for humanities research. And isn’t that what we should be aiming for? Rather than applying whichever method is currently trending to a wide range of research questions, regardless of whether it truly fits?

3. Diversity of Knowledges: What Becomes Computable?

The third diversity is something of a combination of the previous two. I call it the diversity of knowledges. It asks: What becomes computable?

I’ve already hinted at this problem before. Not all fields have the same knowledge infrastructures or digital resources. Some disciplines developed computational resources much earlier than others.

But there’s another issue: many computational methods work considerably less well for certain historical periods and materials. If I’m allowed to generalise a little, this applies to basically everything that isn’t English, Western, and from the eighteenth century onwards. Of course, there are exceptions. But generally speaking, if you’re working with historical material that isn’t relatively modern, English, or part of Western (elite) culture, computational methods are likely to work considerably less well.

This is partly because of the unequal availability of datasets and partly because of differences in model performance. Together, these inequalities determine which kinds of knowledge are easier to study and produce. Modern, Western, English-language, well-digitised materials are computationally convenient. Other languages, periods, cultures, and source types are much harder to work with and therefore risk being overlooked even more. Yet I think this is precisely where some of the greatest potential for new insights lies.

We’ve probably all heard the term knowledge collapse used in discussions of AI. And while it sounds rather dramatic, I think we risk something similar in computational humanities. If we increasingly apply computational methods only to the materials for which they work most easily, we end up studying an ever narrower range of historical sources.

We focus on the data that already exists and is well prepared, rather than undertaking the substantial work required to prepare new material. And in doing so, we risk reinforcing precisely the inequalities and gaps that already characterise our historical record. The diversity of knowledges is about resisting this tendency and asking how computational humanities can help expand, rather than narrow, the range of histories and forms of knowledge we study.

From Curiosity to Care

I’d like to end by suggesting two values that I think should guide critical computational humanities: curiosity and care.

At the moment, I imagine computational humanities as something of a one-way street. We import computational methods into the humanities. We learn new technologies, acquire technical skills, and explore what these methods can do for our research. And that’s wonderful! I would describe this through the value of curiosity.

Curiosity represents the technical literacies through which we bring computational knowledge and methods into the humanities.

But I think we need a second aspect. And that’s what I would summarise under the value of care.

Care means bringing our humanities expertise back into computational technologies, where that expertise is often lacking and urgently needed. It represents the critical literacies that we can contribute to computational methods and to AI communities more broadly.

In other words, I think we need to turn that one-way street into a two-way exchange.

To do that, (critical) computational humanities needs to move centre stage.

We shouldn’t just use computational technologies. We should challenge them, test them, and help shape them.

We have expertise that matters. We have methods and critical traditions that can contribute to some of the most important technological debates of our time. We deserve a place at the table where these decisions are being made. But to get there, we need to stop communicating only among ourselves. We need to move onto a broader stage and bring our expertise into conversations beyond our own disciplinary communities.

That’s my argument for critical computational humanities. Not just more computational methods in the humanities, but more humanities in computational methods. Not just curiosity, but also, crucially, care.

That is my manifesto for Critical Computational Humanities.

So long, and thanks for all the fish!