Refusing Suspicion

Why I won't be making use of Pangram.

I quite viscerally experienced one of my mantras, “writing is thinking,” last Saturday when I sat down to try to draft this newsletter and realized that I literally could not think.

The culprit? Flu, and its associated fever. I’d been going downhill over a couple of days, but it has been almost a decade since I’d experienced an illness that proved truly debilitating - save one holiday 48 hour bout of norovirus which I prefer to forget - so I figured that I could solider through as I had through previous bouts of the grippe, but no dice.

Soon enough it became apparent that I also could not even sufficiently concentrate to read, depriving me of one of the great pleasures of a sick day. Instead I lolled in my bed, nodding in an out of sleep as a series of second rate movies I’d already seen streamed on the television.

Those moments are not without their own pleasures, provided one is confident your body is healing, which it was, though more slowly than I would have wished. Even a week later I could not say that all systems are go.

But it’s not altogether bad to have a reminder of that which you value, in my case, the ability to read, to think, and make sense of the world to myself (and hopefully others) through writing. One of the reasons I do not understand the choice to outsource any part of the writing experience to AI automation is because these experiences are integral to well…my life.

Having more time to marinate my brain in replacement-level movies because my AI agent is working on this week’s newsletter does not track as an enhanced existence and yet this is the kind of positive promise the AI developers are making. Perhaps if I had a personality more attuned toward productivity or achievement I would make “good” use of this time they promise me, but this, exactly this, noodling and stringing one word after the other is what I want to spend my time doing. I have no desire for automation to liberate me from my own pleasures.

Unlike generative AI, I have a mind, and as long as it is not addled, I intend to put it to use.

The topic I intended to put it to us on is Substack’s partnership with the Pangram AI detection program, an initiative apparently meant to help readers identify AI-generated text in order to…I’m not sure, actually…avoid it?

No doubt, there’s a lot of AI-generated slop on this platform as there is a lot of AI-generated slop in the wider digital world. To the extent this kind of initiative works at a scale which slows some of the flood of crap, I suppose I’m for it, but this does not mean you will find me making any use of it, personally.

Now, out a sense of due diligence as someone who opines on issues of writing and pedagogy, I have investigated various AI detectors, including Pangram, which appears to be the best by far, best meaning most likely to render a judgment consistent with reality.

But “best” does not mean infallible. These detectors work through pattern recognition and given that LLM text is extruded as the byproduct of ingesting a massive corpus of human text, it’s not a stretch to believe that some of this text under certain prompting conditions may defy the patterns.

At the same time, human-created text may mirror that which the detector decides is AI.

Freddie deBoer was recently accused of AI use in a post from more than a year ago, a ridiculous charge if you have any familiarity with deBoer and his work. He is far too arrogant about his own writing (compliment, because me too) to outsource it to AI automation. In subsequent experiments, he was able to show how to fairly reliably generate false positives on the Pangram detector.

Charlie Finch engaged in a similar experiment, demonstrating how “fragile” the detection is, showing that by changing a single word he was able to trigger a shift in the detector from declaring something human-generated to AI-authored.

I am grateful others have done these experiments because lord…I could not imagine being less interested in doing them myself. I do not care to map the tolerances of these non-reading machines because their capacities mean nothing to me.

There are a lot of things I choose not to care about, and one of those things is whether or not a particular piece of text is AI-generated. I’m not saying that I’m indifferent to the facts of the matter as to whether or not a text is AI or human-generated. I am team human all the way, but as a reader I cannot think of a less interesting question to bring to a text than “I wonder if AI produced this?”

If I was a publisher or editor tasked with ensuring that humans had produced the texts shared under my banner I may feel differently, but I am a reader, and when I read the question I want to approach a text with is, “What’s going on here?”

Now, despite carrying this attitude into my reading experiences, while reading I have had the question - I wonder if this is AI? - creep into my consciousness. When this happens I keep reading, if it persists and is interfering with the more meaningful pleasures, I stop reading and move on.

I don’t take the text to Pangram to check if my suspicions are correct. I stop. Whatever that text is giving me is not what I’m looking for. There’s more than enough stuff to read in the world, so why spend my time on something that does not feel sufficiently human no matter its provenance?

For sure, I am likely triggering some internal false positives and what I’m reading is not AI prose, but lackluster or unengaging human writing, but so be it. Time to move on either way. It’s not that I’m eager to trust others, so much as I trust myself to know what is and isn’t worth my time.

I’m simply not going to participate in a culture of suspicion that centers AI text in the discussion. Reading from that suspicious place is not interesting or generative to my own work. The reason I read in order to write, rather than yammering endlessly with Claude or ChatGPT as my so-called “thought partner,” is because I need that spark of the other unique human intelligence to fire my thoughts in turn.

I know what it’s like to read from a fundamental place of suspicion, and it is not fun. This is where I found myself when I was teaching and confronted with a pile of student writing I would dig in and start looking for the things they did wrong so I could correct them. If you teach writing long enough you know exactly what mistakes are going to be made, and if you have more students that you should, you may develop some shortcuts for adding marginal comments - keystroke macros - and soon enough you don’t really know (or care) what a student is trying to say, you’re just trying to find the next thing that you can flag.

Thankfully, my dislike of this manner of reading set me on the path of experimentation to find a different relationship with the teaching of writing where I would do my best to privilege the ideas and expression students were offering, as opposed to the mistakes I thought they were making, but my disgust at myself in those moments has proved an indelible lesson.

If I can’t maintain my human agency over my own attention, my own taste, what’s the point of continuing to do this work?

I don’t begrudge others who feel some kind of need to verify the origin of texts, but it’s not something I have to do, so I’m not going to.

Links

Two weeks ago at the Chicago Tribune I wrote a rather painful review, admitting that for the first time, Richard Russo had disappointed me with his new novel, Under the Falls. This week I checked a writer off the list of people I’ve long wanted to read, Domenico Starnone and his novel, Trust.

At Inside Higher Ed I explored my disgust at Harvard’s choice to disband its writing center and lay off its brilliant director, Jane Rosenzweig. These places truly do not care about students, and whatever lip service they pay about the importance of learning to write is B.S. I cannot emphasize this enough.

If you’re interested in how publishing is (and isn’t) dealing with AI-generated texts, I recommend this roundtable on several recent controversies featuring Lincoln Michel, Kathleen Schmidt and others. Similar to what I say above, this is one of those areas where I’m glad other people are thinking about this, but as for me, I’m just not going to spend any bandwidth on it. I know what my work is.

Mark Twain lost his mind and his fortune with an obsession over a new kind of printing press. Great bit of history here.

Trevor Aleo has some very worthwhile thoughts on that so-called “literacy crisis.”

Via my friends McSweeney's, “I’m Timothy, the Short-Spined Raccoon Who Has Taken the Internet by Storm, and the Collective Weight of Your Adoration Is Breaking My Already-Compromised Back” by Emily Flake

Recommendations

1. The Wedding People by Alison Espach
2. Running With Scissors by Augusten Burroughs
3. Say Nothing by Patrick Radden Keefe
4. In the Heart of the Sea by Nathaniel Philbrick
5. The Heart’s Invisible Furies by John Boyle

Sallie G. - Palisade, CO

I went back and re-read some of this recently, looking for a particular mood, and I feel like that mood might resonate with Sallie: This History of Sound by Ben Shattuck.

Well. Even this relatively modest effort has proved exhausting in my diminished state, but it is a pleasure to be able to string at least some thoughts together without the text in front of my eyes swimming into incoherence.

See you next week when I shall be back to full octane!

JW
The Biblioracle