Valuing written material
Reading this Grub Street article recently, about all of the weird ways that people are acting around restaurants these days, reminded me of something I’ve been thinking about a lot: how AI impacts the perceived value of written material.
Once upon a time, a long written document—a report, an RFC, a letter, an email, whatever—carried with it some level of implied effort on the part of the person sending it. The ideas or sentiment or material conveyed within the written material may have been bullshit, but the person asking you to read it did genuinely have to invest some of their own time to actually produce the thing, and therefore communicated something alongside the writing: I cared enough about this to invest my time writing this, and I hope you will in reading it. This obviously wasn’t the only signal for a reader to pick up on, as anyone who read the first few paragraphs of an unhinged screed would be familiar with, but the heuristics for “should I keep reading” in human-written writing are something people have developed collectively over centuries.
AI has now inverted that: a one sentence prompt can be expanded into a hundred page document with zero human intervention, at effectively zero cost. Five years ago, someone dropping an incredibly detailed project plan on my desk spanning dozens of pages, with flowcharts and graphics and such, would have not only been communicating the plan itself, but relaying that they’d invested significant time to deeply understand the project and organize their thoughts in this way. Today, a plan like this implies nothing, I cannot know whether the person even has any concept of what’s written in it, or if they simply had their AI produce it with no review. This ultimately has the effect of devaluing all written material, as the reader has to develop new heuristics to identify the level of human effort put into the document.
I work with AI a lot, and I have a pretty good eye for when one sentence of real thought has been expanded into thirty-seven paragraphs by AI, but I’m increasingly noticing that most people do not have this at all.
I’ve also noticed maybe-a-trend of people complaining about how everyone wants to schedule meetings to discuss things these days. I’d suggest there’s a correlation here: a meeting still carries roughly the same human time investment it always did. A request for a meeting to discuss a lengthy document is itself a request that the person who produced it invest the same amount of time in explaining it as the recipient should in actually understanding it. An instant ask to schedule a meeting to talk through a big doc might be rude if the person receiving it has not read the doc, but is only logical if the people producing the docs have put in less effort to create them than it would actually take to give them a careful reading.
The other side effect here of this phenomenon is that recipients will simply delegate the reading of dense AI-generated stuff to their own AI, a lossy process that increasingly feels like the default: some people have a meeting (recorded by AI notetakers), person A has their AI produce some kind of document based on the meeting, forwards it to the other participants who then have their own AI digest that document. Person B has their AI write a response, and the back and forth continues until things are comfortably averaged into a bland nothingness.
I’m not sure where this winds up, but I’m making an effort to write every word I send to another person myself, even if I use AI to pull information together along the way.
There's no articles to list here yet.