UC Press Blog

What can literary and cultural criticism contribute to our understanding of generative AI? Quite a lot, argues Lauren M. E. Goodlad, Distinguished Professor of English and Comparative Literature at Rutgers University and contributor to a new Representations special cluster on AI. In the Q&A below, Goodlad discusses why nineteenth-century scholarship offers a surprisingly insightful vantage point on today’s AI boom, what literary critics can contribute to interdisciplinary debates about large language models, and why putting the “humanities in the loop” means questioning the idea that Big Tech gets to determine the future of writing and education.

To begin, why is a professor of English with a specialty in 19th-century and Victorian literature studying AI?

With apologies for some deliberate cheekiness, can I simply reply, why not? Put differently, given the prevalence of technodeterministic hype, the topic of “AI” needs some serious scrutiny from academic historians and comparatists. They’re well situated to analyze the politico-economic, sociotechnical, and moral coordinates of this aggressive paradigm and to do so from the stance of earlier epochs of technological will-to-power. Doing this work requires serious openness to cross-disciplinary research which has been a hallmark of Victorian studies for many decades.

Consider that “AI” today is not the name of a specific technology. It is Big Tech’s preferred rubric for the marketing of massively resource-intensive statistical models that mine internet-size troves of data (mostly “scraped” without consent, compensation, or credit) and which—despite the near memorization of datasets that would require about 20,000 years for a human to read—rely on armies of human data workers (often exploited workers in the global south) to make them look smarter than they are. 

Now consider the nineteenth century, which gave rise to the statistical modeling of human characteristics (“anthropometrics”) including eugenicist pseudo-sciences for the supposed measure and ranking of intelligence. Nineteenth-centuryists have explored the speculative financial ventures surrounding cultures of human enslavement and slaughter (rooted in ideologies of racial supremacy and manifest destiny); the enclosure of the commons; the exploitation of factory labor (and the gradual rise of trade unions); capitalist extraction and the expansion of territorial empires on a planetary scale (alongside conservationist and anti-colonial movements); the Taylorization of industrial manufacture; booms and bust of every conceivable kind (along with the advance of regulatory and social welfare regimes); and the rise (and eventual legislative curtailment) of monopolies in railroads, oil, and steel. In the form of text-generating chatbot implementations that model and mimic human data, “AI” is projected as a new manifest destiny—this time powered by digital surveillance, the privatization of the online commons, and an unprecedented concentration of power and resources. 

Of course, the field of critical AI studies already benefits from a great deal of momentum from other humanities-adjacent disciplines—media studies, science and technology studies, and the field of human-computer interaction, for example. But I think literary critics are positioned to add some important new voices. 

I want to clarify that I’m not suggesting that no good can come of powerful pattern-finding in high-quality datasets curated in conversation with domain experts to address specific tasks (for example, better interpretation of data in climate or radiology, or the accelerated development of new drugs). But the push to implement “AI first and fast” in workplaces and classrooms primarily involves generative AI—a relatively new technology that has set off a trillion dollar speculative bubble. (And of course, as lovers of Dickens and Trollope know well, bubbles are a recurrent feature of Victorian fiction.) 

One of the themes running through the essays is that the biggest questions about AI aren't just technological; they're social, political, and cultural. What do you hope readers take away about the role of the humanities in shaping conversations about AI?

I should make clear that I wrote this essay with literary critics primarily in mind (though it is legible to any curious reader, whatever their background). Because few literary critics know gen AI-adjacent fields (such as Natural Language Processing [NLP] or computational linguistics), they can fall prey to technodeterministic hype. Yet, despite that limitation, most people in disciplines that center on reading and writing have good instincts about the limitations of chatbots as tools for research or writing. They know that reading summaries in place of novels or books is a travesty; that grammatical polish is not a substitute for hard-won thought, close analysis, the careful use of evidence, or an unusual voice. And if they’re keeping up with the news, they’ve also heard about the cognitive, emotional, and pedagogical harms of reliance on chatbots; the squandering of energy and water in a time of climate change; the determination to blindside creative workers through models that “train on” their work without permission; and the perilous concentration of power in a small handful of tech oligarchs—the most vocal of whom are outright authoritarians who seek to buy elections, forestall regulation, control the media, and dominate the world economy.

Even so, some literary critics may unthinkingly hew to the tenets of “legacy poststructuralism.” They’re accustomed to thinking of “language” in its textual form and to relegating the study of language use to scholars of communication, writing studies, or linguistics. This has at least two regrettable downsides. First, it hobbles the ability of literary critics to join interdisciplinary conversations about language models that are premised on active theories of language (harking back to the speech act theory of J.L. Austin). Second and consequently, they largely forget the unresolved tensions in poststructuralist theory even though, as it happens, those unanswered questions speak powerfully to the present conjuncture. 

My essay introduces theories of language use (borrowed from writing studies, sociolinguistics, psycholinguistics, and the interdisciplinary authors of “Stochastic Parrots,” an influential 2021 essay by Emily M. Bender, Timnit Gebru and colleagues). And I look again at the work of Jacques Derrida, the Russian formalists, and Fredric Jameson as they challenged the langue-centered legacy of Ferdinand de Saussure. My argument is that literary critics have in fact, “always been action theorists.” In Saussurean terms, language models train on the signifiers of language but not the signs (which, by definition, combine “signifier” and “signified”). I urge a shift away from langue and toward parole (enacted language use)—a shift that Derrida, Mikhail Bakhtin, and Jameson sought in their own post-Saussurean engagements decades ago but which has been forgotten. I hope to revive these questions and the substrate of action theory they rest on. 

This way of theorizing the advent of text-generating chatbots should help critics to back away from over-fastidiousness about the obvious fact that people usually speak, write, or gesture because they have something they want to say. Whatever the medium or the mode of technical mediation, people turn to language when they want addressees to know what they mean. Theorists often use intention to describe this impulse to enact meaning through language use. Derrida himself writes of intention. But I prefer a different Derridean usage: vouloir-dire—an idiomatic formation for meaning or intention that literally translates into “the desire to say.” 

I think that years of reflexive deferral to poststructuralist prescriptions that most haven’t seriously engaged since grad school have left a surprising number of literary critics squeamish about the “desire to say” even though their own work and life obviously rely on it. They wouldn’t try so hard to encourage students to focus on the process of writing as distinct from the product handed in for a grade if they didn’t believe they were strengthening the student’s “desire to say.” 

All that said, I’m not suggesting that literary critics should turn away from textual analysis—I love the work of literary critics and have participated in it for most of my career. I simply hope that literary critics can loosen the grip of a vaguely recalled set of taboos in order to recognize their commitments to enacted language—that is, to social, embodied, and relational practices of meaning-making. 

AI technologies are evolving incredibly quickly, but the questions raised in this issue seem likely to endure. Looking ahead, what conversations about language, writing, or education do you hope this cluster will inspire?

I’m now at work on a book called The Lifecycle of Writing Subjects: On the Futures of Human Poiesis.  Perhaps that title (which riffs on a Ted Chiang novella) provides some sense of the kinds of conversation I hope to spark. The essay—and several of the responses to it—encapsulate the recurring points that I learned from linguists and from my own return to poststructuralist theory: language is not a dataset; disembodied statistical models that train on the signifiers of language but not the signs do not understand the language the way the person reading this interview does. That is why gen AI relies so heavily on human data workers. 

But I also hope it’s clear that some of the most economically and politically powerful companies in human history are doing their best to create the perception that “AI technologies are evolving incredibly quickly.” It’s important not to overstate their case. A few years ago I teamed up with Matthew Stone (a computer scientist who works in NLP) to edit and introduce a two-part special issue on large language models (LLMs), generative AI, and the rise of chatbots which was subtitled “Beyond Chatbot-K” and came out in April and October 2024. We chose that subtitle because we know that tech companies mythologize version numbers to create an aura of inexorable progress. We expected GPT-5 to come out soon after the issue appeared. As it happened that was one of OpenAI’s least impressive launches. Since then, the hype has shifted to so-called “reasoning” models (even more computationally intensive models that slow down the delivery of statistical inference to compare potential outputs) and “agentic AI” (which harnesses a variety of programs to LLM-based systems).  Anthropic’s coding tool has grabbed many headlines and accelerated the implications for how human software developers will likely work (as well as for the quality of code). One of the most important stories right now is the increasing awareness of how expensive it is to run models that deliver statistical inference—often considerably more expensive than employing the human workers the models supposedly can replace, but rarely can replace no matter how large the budget for automation. 

This opens a whole new set of social and political questions. But the bottom line for me is that there is hardly a sentence in the entire 2024 special issue that I would not firmly stand behind. Big Tech companies want us to believe that their innovations move too quickly for researchers to keep up because their air of inevitability depends on an uninterrupted monologue. They may not get their wish! As you say, the big questions are enduring: people feel them viscerally. Increasingly, Americans—and especially young people—do not trust or want “AI.” That is especially true regarding the build out of data centers; but it also true about the pervasive harms of addictive algorithms, “dynamic” pricing, faulty facial recognition, biased hiring models, and the arrogant Big Tech presumption that chatbots are the future of education or healthcare.

Law schools are leading the way toward a sensible focus on the cultivation and assessment of active learning, resistance is on the rise, and the goal of expanding critical AI literacies is becoming common sense from K-12 through graduate school. That does not mean that gen AI chatbots may not find a place in subjects for which a large degree of automation is appropriate (e.g., in certain auditing practices or in computer science learning once the core skills are in place). Nor does it mean that people who enjoy engaging with chatbots should be shamed or mocked. What it does mean is that “AI first and fast” is almost always a losing proposition. 

If “humanities in the loop” means anything it is that researchers, educators, students, and the public at large need to actively participate in the forging of their own futures. Tech leaders and their investors may believe that they alone get to make decisions about what counts as progress. But billions of people have only begun to recognize that they have something very important to say. 

AI has become one of the defining public conversations of our time. Why is Representations the right venue for bringing literary and cultural criticism into that conversation?

Representations’ editors have been wonderful to work with! I can’t imagine a better space for this conversation: the editors have assembled a group of theorists that include a linguistic anthropologist, experts in psychoanalysis and the history of literary criticism, and a digital humanities dream team that has collaborated on an exciting theory of language models that draws on Narcissus. UC Press and the journal's editors have also said that they are willing to sustain the conversation on this blog—potentially enabling the discussion to include technologists as well humanities scholars. I think that would be amazing.

We invite you to read Representations's "Special Cluster on AI," for free online for a limited time. You may purchase single print copies of the issue (175.1), in which the cluster appears, and/or other issues of Representations on the journal's site. For ongoing access to Representations, please subscribe to the journal and/or ask your library to subscribe.