Not a Witch Hunt: Resist AI!

For months, academics and scientists across disciplines, from mathematics to literature, have been assessing the irruption of AI on their fields. As LLMs have become so ubiquitous, so fast, assessment has turned into soul searching: what is the purpose of academic work, what is the nature of education, what is intelligence, what counts as knowledge? International lawyers have not been absent from the conversation and have not shied away from these and other questions. Jean d’Aspremont’s post is a welcome addition to this conversation. The interest sparked by this post could be explained by the prominence of the author and its provocative content. However, the attention it received may also stem from the surprise of seeing an academic known for his critical and innovative scholarship wholeheartedly endorse the use of AI.

At machine speed the post goes into multiple directions. I would reconstruct the post’s main arguments, assumptions, prescriptions and hopes as follows. First, the central argument is that international law academics and journals are in a ‘witch hunt against the machine’ because they want to preserve ‘the romantic and mythical figure of the single-minded sovereign author.’ Second, the assumption is that LLMs are simply the latest ‘regime of inscription,’ nothing more than a tool to formulate and record thoughts. Third, the prescriptions are to embrace the new technology and to assess scholarly work not from the standpoint of authorship but from the standpoint of good work, namely ‘whether the text thinks … argues, narrates, provokes … makes readers think.’ Fourth, Jean’s hope for the future of scholarship in international law relies on individual author’s self-discipline. The disciplined author must somehow delegate writing to AI but not thinking.

I will briefly consider these four aspects stressing the reasons for resisting AI and defend human research in international law.

Where is the Witch Hunt? 

First. Who has been accused and sanctioned without a fair trial? Why to use this loaded term to describe the efforts by international lawyers and editorial boards to tackle some of the problems generated by the irruption of AI in our field? Take for example EJIL’s policy. Rather than protecting the sovereign author, it demands human authors to disclose Gen AI use so that the editorial team can assess if AI (like a research assistant) complements rather than replaces human intelligence in scholarly production. Of course, this policy needs greater specificity in relation to the different tasks assigned to LLMs. And of course, as argued below, individual disclosure is not enough when the profession must collectively draw the line between acceptable and non-acceptable human-machine collaboration, not just editorial boards. However, at this level of generality, EJIL’s policy is not so different from Jean’s very same distinction between AI inscription that delegates and inscription that does not delegate thinking.

The difference is that EJIL is supposed to be in the business of preserving the sovereign author and Jean is not. That is what the post claims citing poststructuralism and Barthes. While I cannot see (as argued below) how deconstruction can be enlisted to endorse AI, I see a postmodern patina deployed to evade normative questions about the nature of AI. True, Jean mentions some of them in passing. A more complete list of questions, not limited to the irruption of Gen AI in academia, but questions interrogating AI as a general-purpose technology, must be examined to consider an inversion of the conventional argument. Rather than identifying the ethical externalities imposed by a train that has left the station, not just academics, but governments and civil society leaders as well as citizens, must ask if we want to dig a train track on the gardens where our thoughts and dreams, where human cultures have grown.

What benefit can possible justify the monstrous environmental footprint of AI? Should it be as beneficial as the steam engine, electricity, or other general-purpose technologies to tolerate the environmental costs? What about the social costs, such as the socioeconomic impact of labor displacement? Given these costs and the extractivist nature of AI, do polities have a say on the distribution of benefits? Let us not forget that LLMs have been trained on data stolen from countless human authors! What about the psychological trauma suffered by workers in the Global South, who screen violent and illicit content to train and filter AI models? Have we nothing to say about the geopolitical race to AI supremacy? What about the astonishing lack of regulation, the threat to humanity posed by AI tools in wrong hands, the liability gaps regarding autonomous AI agents and the irruption of military AI? Have we forgotten about the AI tools that enabled massive death, from Gaza to Iran? Finally, what about the threat that private tech companies pose to our democracy? And, the threat to the sustainability of our planet posed by their billionaire owners’ fantasies of galactic escape?

Those who did not ask for this devastating, soul racking train, deserve more than talk about responsible use of AI. We deserve answers to these serious questions from academics embracing AI.

Inscription as Thinking

Second. Are LLMs, as the post suggests, simply tools to inscribe thought? Can we delegate writing without delegating thinking? Unfortunately, Jean does ‘not feel the need to spell out … what … qualifies as inscription and what counts as thought,’ for he believes that each scholar ‘negotiates [this binary] differently.’ Although understood as fluid, the binary itself remains therefore unsubstantiated. This is not a useful assumption, as it is precisely on this fine line between thinking and inscribing that acceptable uses of AI depend on. 

Given the assumption that if put to good use AI does not affect thinking, Jean ignores the existing and growing literature showing the serious cognitive effects AI inflicts upon its users: cognitive offloading, cognitive debt, cognitive complacency. As one begins delegating tasks to LLMs, even if according to Jean’s criteria only as an inscription tool, where and how to draw the line to avoid cognitive surrender? Because of the post’s assumption, nothing concrete can be gathered about the uses of AI that impact academic research and thinking.

In contrast, EJIL’s policy is explicit about where there is and isn’t a problem. The issue is not formatting footnotes and correcting grammar. The problem emerges when the use of AI eclipses authorship and responsibility over the generation of a text. However, like seismic waves that once sensed, announce an earthquake that has already arrived, the problem with AI arises long before generative AI is used. And like the aftershocks of a large seismic event, conventional ways of doing research will continue to be shaken as the role of scholars and universities will be unsettled long after the irruption of Gen AI.

Long before generating texts, the problem surfaces when AI is used to generate literature reviews and reading summaries, when used for mining vast swathes of literature, when used for brainstorming and ideation, for framing research questions, structuring a paper and identifying contradictions and when prompting replaces writing. The problem begins here, because these tasks are not just too close to thinking, they are part of thinking. Reading is thinking. Writing is thinking. It is difficult to conceive delegation of inscription without surrendering thinking, because reading and writing do not simply spring from the use of preexisting cognitive resources, they create new resources for thinking. Through reading and writing humans have become aware of the components of language that are unconscious in speech, creating new resources for systematic and rational thought. 

It is puzzling to see Jean supporting the distinction between thought and inscription, which privileges speech as unmediated reflection of thought and truth, over writing as mere representation or copy of thinking. It is puzzling, not only because it reflects attachment to a metaphysics of presence, but also because it overlooks the long human history of inscribing as a tool for thinking –in cave paintings, into skins, including our own, in the notebook. The distinction discounts our experience of the unity of thought, which extends from the formulation of the whole idea before uttered, to what occurs after deliverance when we think again the whole idea, including what happens in between, as each word is pronounced during deliverance, we feel meaning as it passes.

If we no longer experience meaning as it passes through deliverance because it is the machine that is inscribing, we are losing an essential experience with repercussions that are already becoming clear and will definitely be long-lasting. The severity of the aftershocks resulting from this radical transformation will depend on how scholars, universities and governments respond to the encroachment of AI on our epistemic sovereignty, and on how the resulting damage is prevented or repaired. 

The Case for Resisting AI in International Legal Scholarship

Third. Jean invites international lawyers to embrace the machine in order not to fetish authorship but to value texts that provoke thinking. I could not agree more with the second part of the prescription. However, for the prescription to work, concerted action is needed first to recognize AI’s threat to literacy in order then to reimagine literacy in the time of AI. Only then LLMs may become useful communication partners, rather than short-sighted collaborators, like parrots that producing immediate answers, cloud our judgement and prevent us from overcoming the difficulties necessary for learning.

Still, I invite legal scholars in international law to Resist AI! We must resist to protect the diversity required for intellectual vibrancy and innovation. We must resist to prevent the methodological, topical and linguistic convergence and the citation monoculture that AI use generates. The multiplication and concentration of research resources enabled by AI may limit new discovery, as breakthroughs depend not on concentration, but on diverse attention. Today, we must protect diversity of attention to achieve the urgent breakthroughs needed in our field.

During normal times, ‘normal science’, to use Kuhn’s terminology, operates within the dominant paradigm. And working within the paradigm, academics may efficiently turn on Gen AI to churn out massive amounts of descriptive and problem-solving scholarship. This is happening in every field including international law. Why would it not be great delegating on machines the tedious work of scraping the digital barrels to find the arguments and sources needed to understand and interpret treaties, court rulings and arcane legal doctrine? Great it is for increasing impact metrics. But if you care about the state of global politics and about the health of our planet, you know that we live in a moment where the rule of law in international relations is more honored in the breach than the observance, where international law has become irresponsive to the most pressing global governance problems: war, climate change, genocide, refugees, pandemics, and the AI race between China and the USA. Then, you know that the dominant paradigm needs rethinking.

Machines are not good at rethinking. Only we human scholars have the experience, moral compass and ability to pass judgement on our international legal order. Delegating academic writing to AI is to place reflection under the limits of algorithmic combinations and predictions that reflect the existing universe of material, which we know is not universal material. It is the digitalized, mostly Western, mostly English-language material. Imagining new laws for a new planetary order demands forward-looking reflection not backward-looking algorithmic recombination. Imagination demands inspiration. There is no imagination or inspiration in mechanical recombination. So where else should we look for inspiration but beyond the Western AI monoculture and its digital archives, and search of times lost?

The Illusion of Self-Discipline

Fourth. The self-disciplined scholar envisioned by Jean may overcome these dangers and criticisms by exercising individual good judgement in the use of a consumer product like his or her favorite LLM. Individual market choice, however, is too reductive a framework. Deciding if AI is going to be used liberally or subject to discipline; if to resist AI or to create AI free zones, these are not alternatives for the individual academic to adopt, but choices to collectively make to shape the nature of the discipline of international law we aspire. If unrestricted use prevails, aren’t more senior scholars embracing AI sucking all the airspace, while denying the young the fun they had, making mistakes, writing embarrassing stuff and clumsily mumbling their first sentences in English, the global academic tongue? If regulated use is accepted, should junior academic’s submissions and senior academics’ publications be capped? What type of analogue educational opportunities should younger scholars have? How can we restore the luxury that senior scholars had of developing their thinking skills slowly with pen and paper? These are collective choices in need of institutional solutions.

However, many institutions that could deliberate on these choices–universities, governments and specialized international bodies such as UNESCO–seem to have capitulated to a dictatorship of no alternative but the use of AI. It is the editorial boards of academic journals, faced with exponential multiplication of submissions, that have developed regulations out of necessity. Editorial boards have found themselves on the front lines of debates about AI, as they must not only respond to Gen AI submissions, but also address questions about the ethics of AI-generated reviews, deal with reviewers’ reluctance to assess AI generated drafts and resolve the problem of (inadvertent) plagiarism in AI generated research. 

Deliberation and decisions on these questions are needed beyond the realm of academic publication. Political change rarely happens without mobilization from below. It is then the students and children whose analogue futures have been foreclosed, the authors who have been outcompeted by machines, the readers bored of AI slop, scholars troubled by the loss of academic craft, workers whose jobs are displaced, who must unite to resist AI.

We know what the AI evangelists will say: the train has already left the station and there is no other option than to get on board. Yes, trains of dystopia are in full swing now, but for how long depends on how long we will accept being robbed of utopia, of our attention, barred from the joy of reading, from the fun of playing with ideas, from what has always been a central part of thinking and discovery: daydreaming, wandering, idling, where in the transitive parts of the stream of thought we become aware of relations and associations.

Maybe international law, our discipline occupying a little corner in the larger body of human knowledge, could show why and how to reject AI. There should be no room for AI when LLMs cannot outperform our human work. Imagining better legal norms and institutions that can elicit global solidarity or give form to simple forms of coexistence requires moral and political judgement. Our disciplinary identity is rooted in the wisdom that the life of the law is not logic, but experience. The machine has no experience with the struggle for human rights of the mothers of the disappeared. We must not delegate our experiences to machines that cannot feel the fear of war and the frustration that we feel when the law fails to limit suffering and restore peace. We should not allow machines to coldly slice and recombine the elements of the crime of genocide, impervious of the moral disgust that makes humans feel sick to their stomach. There is no use to deploy agentic models that have no experience with finitude, no breathing problems caused by the smoke of apocalyptic wildfires, to design the legal architecture needed to tackle climate change.

The discipline of international law may play a small part in rejecting AI, for if we surrender to AI, we lose international law and abdicate the project of human ordering for ourselves, the more-than-human and the planet. International law is as beautiful as it is flawed and messy. Just like human life. This is our tradition, the one that was forged by colonialism and self-determination; that has given voice not only to Francisco Vitoria, but also to Titu Cusi Yupanqui; Lassa Oppenheim and Hersch Lauterpacht, as well as Wellington Koo and Abd el-Krim; and more recently, to Doris Stevens, Bertha Lutz, Hilary Charlesworth and Karen Knop; and that has had plural approaches, naturalism and positivism, Soviet approaches, the New Haven School, TWAIL. We bear responsibility for the Peace Palace and the UN Headquarters as well as for Auschwitz, Hiroshima and Nagasaki and Gaza. Just as in life, we have been responsible for both war and peace. Now, however, we international lawyers, upholding our imperfect tradition, may have become in part responsible for preserving human life as it has been. Let us Resist AI! Defending AI free zones, in the classroom, in human-only-journals. And when resisting turns impractical, let us demand scientific integrity, AI literacy and the protection and cross-subsidy of humanistic academic life. 

Human scholars of the world, unite!