It’s fashionable to talk about crises in education — and that goes double for literacy. Books and articles called “Why Johnny Can’t Write” have proliferated since the 1970s, and teachers, university faculty, and the general public frequently claim that today’s students are worse at writing, or reading, or communicating in general than they used to be.
As an experienced instructor of academic writing, I don’t have a lot of patience for this rhetoric. People in my line of work have been complaining about student writing since the job “writing teacher” was first created. To take one example, Adams Hill lamented the “bad spelling, confusing punctuation, ungrammatical, obscure, ambiguous, or inelegant expressions” of his students. And this guy ran a writing program at Harvard. In the 1870s.
This advertisement has not loaded yet, but your article continues below.
Our latest crisis is ChatGPT and similar large language models. Some see these programs as an inevitable step on the path to technologizing and automating all work of the kind we used to think only humans could do. If robots can do it, maybe it’s not worth trying to get first-year university students to write 500 words about, say, the application of political theory to the Russian invasion of Ukraine, or the molecular makeup of the particulate matter generated by forest fires.
I’m sympathetic to this perspective. I assign workaday things like summaries of short articles, which GPT seems to be able to produce accurately. I can imagine a student snubbing such assignments, instead feeding scholarly sources into an AI chatbot and reading the summaries to inform their knowledge of the subject.
The real crisis, however, is that we have already admitted defeat when we say that ChatGPT can “write” at all. Professors and pundits have ceded far too much ground to computer programs that are good at predicting statistically probable sequences of letters.
This seems to be because we are losing faith in the intelligibility of human meaning itself. Instead, we’re nihilistically embracing the techno-fatalist belief in an inevitable future where computers produce all writing, art, and other products of symbolic meaning.
This advertisement has not loaded yet, but your article continues below.
We should do whatever will help our students get jobs entering prompts into GPT-75 or whatever, right?
Wrong. A recent flowchart published by UNESCO succinctly sums up the problem. The first question it asks: “Does it matter if the output is true?” If ‘No,’ then it is ‘Safe to use ChatGPGT.’” As professor Nigel Harwood of the University of Delaware wrote on Twitter, “We can stop right there.” The fact that ChatGPT and similar products regularly spit out sentences that are not in any sense accurate should give us pause.
It’s not that it’s bad to use technology to enhance writing. We almost all do it with some kind of spellcheck program. It’s that writing at its core is meaning: grappling with ideas, bringing our unique lived experiences to bear on the world. Writing is a human encounter with otherness, a meeting of minds.
AI journalism is not just prone to errors, it is missing the touch of someone who has looked into the eyes of an interview subject or choked on tear gas. AI business plans lack the experiential knowledge of what it means to shut down a failed startup but have the guts to try again. And even academic “writing” generated by AI is missing the process of struggling to stake out one’s own place in the larger academic discipline and, indeed, the world.
This advertisement has not loaded yet, but your article continues below.
Saying that AI “can write” because it can produce grammatical sentences is like saying a boat can swim, or a guitar can sing. It doesn’t make sense to claim these things happen, except metaphorically.
Joan Didion famously wrote, “I write entirely to find out what I’m thinking, what I’m looking at, what I see and what it means.” Computer programs can arrange symbols but not think; process visual data but not see; recapitulate human texts but not mean.
Removing human agency from the creation of texts makes ChatGPT’s output a poor simulacra of what writing is, or can be. The texts that AI models produce are thin imitations of the thick and deep meaning that happen when we bring ourselves into an encounter with the word and the world through writing.
If AI language models seem to be in the business of meaning, it is only because we are able to interpret and imbue the texts it produces with meaning. Only people can mean.
If we want to be able to teach writing, and to do writing, well, we need to start with this truth: Writing is not something that can be outsourced to an algorithm, but involves showing up with your whole self, mind and body, and leaping headfirst into a risky and rewarding endeavour in which you dare to mean.
0 Comments