Word Worries

Greetings, and welcome back to this very occasional column!  As I have mentioned in previous editions, my pre-newspaper professional life mostly involved teaching college freshmen, who are just old enough to be on their own and just young enough to have no idea what to do about it.   In some ways, they’re like kids raised by vegan yoga instructors who have to spend a weekend at their aunt’s house while their parents attend the wheatgrass extract Expo in Des Moines, and when their aunt goes to her bridge night they get left alone with all the leftover Halloween candy.  Or at least some of them are like that.   The others face the near constant deaths of relatives—some of whom die more than once—that necessitate their turning in papers weeks late, and which are still miraculously written at the last minute.

Anyway, I guess my point is that I’m glad I’m not teaching in this new AI era, because the age old problems are now new age problems that don’t yet have solutions, as far as teachers are concerned.

In the olden times, as anyone under twenty-five or so might call them, there were places we used to call “paper mills,” and not the wholesome kind like Charles Ingalls worked at on the show Little House on the Prairie.  I don’t remember if he worked there in the books, because I was a much bigger fan of the show (especially the later seasons when they ran out of source material and began putting everyone in mortal danger every episode—fires, wolf attacks, serial killers, whatever it took to chase those ratings!).  

Paper mills were shady “research partners” that could be found through the early precursors to social media, which were basically message boards for specialized topics.  Or meeting places for nerds, as we would have said then.  You would come across a thread on a listserv with a subject line like “Looking for paper advice?” or something fairly innocuous like that, and then initiate contact with the poster, who usually either worked alone or with a small group of poor students to give you “advice” in the form of a customized paper fitting the needs of your assignment.  Payment would have to be through money orders, primarily, since this was well before Venmo and PayPal.

You would get your paper, retype it to personalize it (although sometimes students forgot to do so), turn it in and hope for the best.  I’ll let you in on the fact that as teachers we would excitedly show off especially terrible examples to each other and laugh and laugh, all while dreading having to fill out the paperwork involved in academic dishonesty reports.

These papermill writers would do an awful lot of work, obviously, and would generally charge by the page and course level.  It sounds crazy, but I solemnly swear it really happened.  

In a lot of ways, the whole thing sounds romantic when positioned next to the current version, which just output data.  What used to be an artful profession based on the craft of rhetorical mimicry and delicately nuanced mediocrity is now just an answer to soulless order like “Chat GPT, please write a five page paper with six credible sources in MLA format on the topic of water rights in the west and their effect on land development for a second year college course in pre-law that is likely to earn a B- and has a standard number of grammatical errors and a believable amount of sourcing sloppiness.” 

(For any college freshman, that’s free advice.  You have to tell AI to do the equivalent of soaking it in tea, the way that document forgers do to make something look older.  There is nothing more suspicious to an instructor than a paper without mistakes.)

Paper mill writers used to have to do that on their own, and it was a delicate task.  Sometimes, they would be from, say Canada or some part of the United States with completely different idioms than our state schools.  One easy strategy for “dumbing down” writing is to lower what is called “register” in your language.  Register is basically a way of describing formality.  A good rhetorical mimic will use a combination of vocabulary shifts, syntax idiosyncrasy, and punctuation gaffes to lower the written register of a document.  In other words, they might take the previous sentence and turn it into “if you want to fool someone, screw up your writing a little too make it more believable.”

AI takes all the fun out of it.  Since Learning-AI has access to millions of C papers from millions of struggling students, it can make a meh paper in literally seconds if you ask it to.  Most students don’t ask it to, though.  They assume that if they’re going to bother to cheat, they might as well get everything they can out of, and that is usually their downfall, since instructors read a LOT of papers.  When a student whose grades are consistently near the danger zone suddenly turns in a paper that begins “Zeno’s paradox obtains to good intentions in an infinite regression, which is why modern milling practices obviate the utility of paving as a metaphor for constructing the path to the underworld” they are not doing themselves any favors.

Currently, there is something of an arms race between AI programs and AI detection programs, which are likely written by AI.  It’s a longer and more complex topic, but in my opinion it will not be possible to “outthink” AI—it’s simply too nimble, fast, and learns without pause or constraint.  What we will really need is a strategy for using the realities of AI to our advantage, rather than a threat.  

In the meantime, I promise that we will continue to put out a paper with real, human errors of the type you just can’t fake!