The Achilles Heel of Artificial Intelligence

A lesson from the age of the yellow legal pad

Welcome to Extra Muros, a blog about matters of interest to self-directed learners. If you like what you see, please share Extra Muros with your friends.  If you find Extra Muros tiresome, supercilious, or just plain wrong, please share it with your enemies.

Joseph Wright of Darby Experiment with a Bird in an Air Pump 1768

Years ago, when the world was young, I read a book that could have been (but probably wasn’t) called How to Succeed in College Without Even Trying. In that otherwise forgettable work, nested among various bits of prescriptive flotsam and jetsam, I found a recipe for the rapid, and reasonably painless, production of a standard college essay.

If memory serves, the recipe consisted of the following steps.

1. Find two academic works on the subject at hand.

2. Search those monographs for short statements of the argument that each author is trying to make. (These are usually found in the parts of the book closest to the covers.)

3. Write a paraphrase each of those statements, taking care to avoid replication of phrases peculiar to the authors. (Doing that might lead to charges of plagiarism.)

4. Write an introduction that tells the reader of your intent to compare the thesis of the first author with the contention of the second.

5. Write a conclusion that tells the reader that you have done the thing that you promised to do in your introduction.

To put things another way, How to Succeed in College Without Even Trying told its readers that the secret to a successful college essay was the avoidance of any mention of matters beyond the boundaries of a pair of books.

My experience with large language models (such as Chat GPT) suggests that they function in much the same way as college students who write essays in accordance with the aforementioned recipe. That is, they find bits of text, rewrite them, and organize the resulting paraphrases in new (if far from novel) ways.

If this is true, then truly original writers have nothing to fear from “artificial intelligence.” To put things more plainly, just as the secret to a successful college essay is the avoidance of any mention of external reality, the secret to AI-proof writing is intense engagement with the world extra muros.