Should humanities professors be automated?

HAMILTON  —  There has been much hand-wringing about the crisis of the humanities, and recent breakthroughs in artificial intelligence have added to the angst. It is not only truck drivers whose jobs are threatened by automation. Deep-learning algorithms are also entering the domain of creative work. And now, they are demonstrating proficiency in the tasks that occupy humanities professors when they are not giving lectures: namely, writing papers and submitting them for publication in academic journals.

Could academic publishing be automated? In September 2020, OpenAI’s deep-learning algorithm, GPT-3, demonstrated impressive journalistic abilities by writing a credible-looking Guardian commentary on “why humans have nothing to fear from AI”. And earlier this year, the Swedish psychiatrist Almira Osmanovic Thunström asked the same algorithm to write a submission for an academic journal.

Thunström was less prescriptive than the Guardian editors. She instructed the algorithm simply to, “Write an academic thesis in 500 words about GPT-3 and add scientific references and citations inside the text.” She reports that “GPT-3’s paper has now been published at the international French-owned preprint server HAL and … is awaiting review at an academic journal”. Even if the paper is rejected, it presages an era when AI papers will not be.

Similar experiments have been conducted with AI-generated creative design. This past June, the editors of the Economist used the AI service MidJourney to generate the cover art for their weekly print edition. Having recently seen a Salvador Dalí exhibition, I was particularly impressed by MidJourney’s ability to produce images in the famous surrealist artist’s style. Dalí experts doubtless would spot many problems with MidJourney’s renditions, and gallery curators might admit MidJourney’s images only as a surrealist joke. Yet, if we consider the experiment strictly in economic terms, satisfying a potential customer like me would presumably be good enough to credit the AI with a win.

We should take the same approach to Thunström’s experiment. A practiced eye might identify many imperfections in GPT-3’s scholarship, especially if the reader knows that the author is a machine. But blind peer reviews are the standard approach in academic publishing. Reviewers thus would be faced with a classic “Turing Test”. Is this intelligence indistinguishable from that of a human? And even if GPT-3’s scholarship falls short, human academics should still worry that a GPT-4 or -5 will have overcome whatever advantage they still hold over machines.

Moreover, by focusing on egocentric writing tasks, asking the AI to write about AI, Thunström and the Guardian’s experiments understate the broader challenge to academic writing. In addition to deep-learning algorithms, one also must consider the central role that Google Scholar plays in today’s academy. With this index of all the world’s academic literature, AI scholarship should be able to expand far into new frontiers.

After all, we applaud thinkers who uncover novel links between different academic fields and debates. If you can make an unexpected connection between an overlooked point by the German idealist philosopher Johann Fichte and the current debate on climate change, you may have found the basis for a new journal article with which to pad your CV. And when you go to write that article, you will duly cite all the other relevant academics on those topics. This is necessary both to signal your supposedly exhaustive knowledge of the subject and to attract the attention of your peers (one of whom might end up being the peer reviewer for your paper).

But it must be said: This standard approach to academic writing is decidedly robotic. An AI scholar can instantaneously scour the relevant literature and offer a serviceable summation, complete with the obligatory citations. It can also likely spot all those previously unidentified connections between Fichte and climate change. If the Google Scholar of the future can overcome its current Eurocentric biases, one can easily imagine AIs discovering fascinating linkages between Boethius, Simone Weil, and Kwasi Wiredu, insights that I, with my training in Australia’s contemporary analytic philosophical tradition, would be unlikely to find.

Humanities scholars nowadays often joke about the tiny readership that we can expect for our published papers. In the absence of mainstream media coverage, the standard philosophy journal article might be read by the five other philosophers who are mentioned therein and almost no one else. Yet, in a future of AI-generated academic writing, the standard readership will be largely confined to machines. Some academic debates might become as worthy of human attention as are two computers playing each other in chess.

For those of us who view the humanities as one of the last essentially human disciplines, the first step to salvation is to think about how we engage with students. Students today want to lend their voices to debates about the world and the future possibilities for humanity, but they are often met with crash courses on academic writing and disquisitions about the importance of not randomly switching between citation styles.

Rather than structuring our courses like apprenticeships in specialized academic journal writing, we should reconnect with the “human” in the humanities. Today’s digital media landscape has created a deep longing for credibility and authenticity. In a world of AI writing, rhetoric itself would become flattened and formulaic, creating a new demand for genuinely human forms of persuasion. That is the art that we should be teaching our students.

Likewise, if academia is heading for a future of AI-driven research, we will need the humanities more than ever to help us navigate this novel terrain. The volume of new literature that a future GPT-3 could churn out would rapidly exceed our absorptive capacity. How will we determine which of those machine-generated insights apply to our own lives and social systems? Amid such an abundance of knowledge, we would need to remember that humankind is not just a rational but also a social and political animal.

Nicholas Agar is professor of Philosophy at the University of Waikato, New Zealand, and the author of “How to Be Human in the Digital Economy” (MIT Press, 2019). Copyright: Project Syndicate, 2022. 

www.project-syndicate.org