What I have to show for a year of living (dangerously?) with AI as a self-employed teacher in Japan
The Bulletin (Issue 49)
It’s been close to a year since I wrote this about Artificial Intelligence and teaching English. The sky hasn’t fallen yet, and AI hasn’t had any major effect on my teaching business either. Yet.
It’s, I think now, way too soon to make any definitive pronouncements on AI and the future, and I would not be the one qualified to make such pronouncements.
But…
I have been listening to what other people have to say about it all. Consider marketer Daniel Priestley (my half-remembered version of what he said here):
In the AI economy, you will either be a consumer, sliced and diced by AI algorithms to accept more and more products and services, or you will be a creator/producer harnessing AI to do so much more than was ever imaginable before. Your choice.
Well…
I’m not keen on binary choices, the truth could well be somewhere in the middle, or neither of those options. But I have been using AI in a small way as a creator/producer. I used an AI picture generation tool to help me make over 50% of the artwork in my latest conversational English textbook for adult learners, and I used ChatGPT to generate the text of five of the 42 lessons. Here’s what I found:
The pictures ranged from awful to amusing, with a few good ones. They work in the textbook as illustrations that don’t have to be realistic. They are like the illustration I used at the top of this post (I gave the prompt “A robot looking into Pandora’s Box,” and this was the only one I considered usable after 12 iterations). I think of them as a poor man’s New Yorker cartoons. Since my textbook was designed to stimulate conversation, even bad pictures that come off as slightly surreal (like the one above) work fine.
If a picture really was too awful, a quick click of the “regenerate” button or rethinking the prompt to something else would work a treat. For some reason it just couldn’t generate “a piece of cake” that looked anything like what I think of as a piece of cake without 40 re-generations.
I only resorted to AI text for topics that I personally knew next-to-nothing about and didn’t have the time to learn about. But I verified some of the factual assertions with Google searches and I think I got away with it.
But maaan, the AI-generated text (at least by the early iteration of ChatGPT I used) was grammatically OK but often repetitive, dull and needed re-editing and sprucing up by hand.
Here were the benefits though:
The similar style of the illustrations meant the book had a more uniform appearance than if I had sourced all the illustrations from a stock pic website, which was my pre-AI first port of call.
Also, for the limited purpose of finding simple illustrations for a textbook, it saved me maybe a couple of hours of searching through websites for usable artwork.
The artwork was all copyright free, and therefore usable to me for free, saving me maybe ¥10,000 to buy stock art.
The text-generation saved me research time, say four or five hours, but necessitated a little more editing and checking than I would have normally done, adding maybe an hour to the usual process, so in total AI text saved me say three hours of my life.
All in all, AI saved me maybe 10 hours and ¥10,000 on a project that would otherwise have taken 50 hours to complete. A 20% saving was good, hardly a 10x improvement to be sure, but the end result was probably of a higher quality than I could have done without AI. And since I wasn’t going to hire an artist or editor for my textbook anyway, I don’t feel that I cost anyone their job, apart from the stock art website owner, eventually.
But this is just one small step.
What do you think? Did I do the right thing by using AI or have I opened Pandora’s Box? How do you see AI affecting teaching? I and your fellow Freetalkers would love to know.
See you on Friday for the Staffroom video podcast where I discuss what to do if you find yourself feeling trapped as an English teacher in Japan.