Of all the questions I receive from readers at Too Much TV world headquarters, the most common one is some variation of “Do you use A.I. on your web site and in your newsletter?”
That is not a crazy question, because the use of artificial intelligence in journalism is growing. The Wall Street Journal just announced it was fine to use A.I. to help write op-ed pieces. Plenty of sites are now using A.I. to crank out local news, financial stories and other pieces that can easily be cobbled together by a computer that scours the Internet looking for relevant data.
There are newsletters devoted to helping journalists find the best A.I. programs and I know quite a few writers who use A.I. to search the web for relevant stories they might have missed.
I tend to be in the “less is more” camp when it comes to using A.I. driven software. Ed Zitron recently posted a lengthy “I Hate A.I.” screed in his indispensable newsletter Where’s Your Ed At?, and honestly, I don’t think I could better describe my feelings towards the technology:
You know what I’d love? Spell-check to work in Google Docs rather than putting a red squiggly line underneath and saying “yeah there’s probably something wrong with this, I dunno what though.” I’d like Microsoft Word to stop crashing because I have too many end-notes. I’d like Riverside to not have 10 different menus to click through to get to a link to send a person to join my podcast. I’d like my email to not be full of spam. I’d like things to “just work” rather than constantly fighting some sort of broken app or broken UX element or weird bug or intrusive pop-up about a feature that I don’t want. I’d like Slack or Discord to not feel like digital escher paintings of different notifications.
LLMs are sold as some sort of magic tool that can fix “anything” without ever specifying what that thing might be, mostly because they cannot be trusted, even in things that they mostly get right, to do things right every time. While they can do “more” than they used to, the extent of that “more” comes with it the danger of giving a mindless software tool access to your computer’s files, which it may choose to delete in pursuit of “efficiency,” which makes investigating what they might be able to do equal parts convoluted and dangerous.
One critique of my work is that I’ve never used LLMs. I have! I experiment with them from time to time to make sure I haven’t missed something. I used one to debug a problem with my son’s Minecraft add-on the other day, and it took 30 minutes of fucking around trying things to eventually sort of work it out. The other day I used one to install a Pokemon Minecraft mod, then when I asked it to make sure the PS5 controller worked with the menus it broke a bunch of stuff, though I’ll concede it was useful that it installed something and it sort of worked.
For me, I have found that I can’t teach an A.I. program (I refuse to call them “agents,” because that sounds near sentient) to find stories for my newsletter, because half the time I don’t know what I am looking for until I stumble across it. I subscribe to dozens of newsletters and a fair number of the topics I end up writing about were sparked by some piece I read in a newsletter focused on South Asian spices or the thriving Albanian street dancing scene.
I certainly wouldn’t outsource my writing on any level to artificial intelligence. Yes, maybe it could help me clarify my thoughts. But that process of clarification and editing is how I learn and the way I create something that is me. For better or worse, when you read my writing, it’s clear who wrote it and that the person was human. Flaws and all.
There are two very specific circumstances where I use computer-aided technology in my daily editorial workflow.
When I do an interview, I run the audio file through Turboscribe, which creates a rough transcript I use as a starting point. Then I listen back to the entire conversation, cleaning up and correcting the transcript as I go. That cuts an immense amount of typing time out of the process, particularly since I am still a hunt-and-peck kind of typist. But I end up with an interview that is accurate, which wouldn’t always be the case if I relied on technology.
The second circumstance where I use technology is once I have finished a piece. Longtime readers might notice I have a bad habit of forgetting a word here and there. I don’t even notice it many times when I am re-reading it before publication.
So I run the piece through Grammarly, which highlights misspellings and obvious missing words and grammar problems. But that is the extent of my use of the technology. I don’t accept Grammary’s suggestions about rewriting passages and I certainly am not interested in allowing it to alter my work to make it more “readable.”
And that’s it. Everything else is me and I think my use of technology hits that sweet spot of making my job easier without substantively changing anything about my work.
I think every journalist should be transparent about whether they use A.I. and I hope this explanation answers all of your questions.
Rick