Look, I don’t feel good including that self-checkout meme here, so let’s get it out of the way.
I’m still seeing lots of ads and think pieces and social media posts and podcasts and all kinds of content about AI and jobs. Specifically, AI and my job.
AI is coming for me! It’s going to put me out of work! I’d better just pick up a Machine Learning certification and become a technician!
AI is terrible! It can’t get anything right! It’s making stuff up instead of providing actual facts! Conversation partners are not search engines, have you even seen Cheers?
I wrote some blog posts about ChatGPT a while back on the subject. The gist is that I really couldn’t get it to produce content that was up to my own standard, nor could I get it to produce material coming close to something I’d published about a year before, even with extensive prompting.
I tried working with it more to see if there was some other utility to it, but it was really just a sounding board. It came up with the same kinds of prewriting ideas I would come up with, but it did it quickly, so maybe there was some benefit.
In the end, I was looking at it entirely wrong. Should businesses adopt AI writing tools instead of hiring writers? If you’re reading this looking for the answer, then I hate to say it, but you already know. If you’re here looking for a writer to tell you that you can’t trust AI to write for you, then that’s the case. On the other hand, if you want a writer to say that AI can be trained or used effectively to create good writing, that’s also true.
I’m not being wishy washy. I’m just remembering one of the core rhetorical principles Aristotle taught folks 2500 years ago.
Finding Available Means of Persuasion
Aristotle defined rhetoric as a thing with three components.
Its purpose was persuasion, getting an audience to think or believe a thing. It wasn’t malicious persuasion. A rhetorician wasn’t there to convince you the Earth was flat. A rhetorician’s goal was to get a concept from their mind into yours, fully intact, without you rejecting it. When I convince my daughter to put on a coat on a snowy morning, I’m using rhetorical principles to persuade her to believe that it is, in fact, cold enough to warrant another layer of clothing.
It used the available evidence to support claims and arguments. You can’t fabricate information to convince an audience. That’s lying. When your audience realizes you’re lying, they’ll stop believing you and your persuasion fails.
And, the method of gathering that evidence is finding. A rhetorician doesn’t have all of the information ready to go when it’s needed. We need to do some thought to figure out which lines of argument and which kinds of data will resonate with our audience.
Back to the coat thing: if I tell my daughter there is snow on the ground, she will grab a parka. If I tell her it’s 15 degrees outside, she’ll say “that doesn’t sound cold” and head outside (before rushing back in to grab a coat). One time I told her it was negative ten degrees. She told me that’s a made up number and there’s no such thing as negatives. (She’s in elementary school. They haven’t gotten that far in math.)
Why am I telling you this on a post about AI?
Generative AI and predictive AI each lack at least one of these features.
Generative AI cannot find evidence. That’s just not in the program. It can create text in response to a prompt, and that text can sound pretty convincing if you craft your prompt really well. But, that’s the end of it. If I told ChatGPT “convince my daughter to put on a coat before school,” this is what it would say.
Seven barely related lines of argument. One or two might work, but ChatGPT is throwing spaghetti at the wall.
Hey, just for fun…
Ten arguments instead of seven. Some are the same argument with a different lead-in phrase. Again, some of those are going to convince me, but I really don’t care about the versatility or professional image aspect. ChatGPT threw spaghetti at the wall and lucked out with a few strands, but it missed a very important part of persuasion.
It didn’t care about my or my daughter’s attention span.
Directing Attention
Hands down, one of my favorite rhetorical tools to teach was directing attention. The basic idea is that a writer/speaker gets to set the reader’s/audience’s attention on a thing by identifying and describing it first, which ends up framing the rest of the conversation. You can see it at the beginning of each list item, where the same idea gets a different frame. Item #1 sells the same basic idea, but it puts the focus on either coziness (for a 10-year-old girl) or protection (for a mid-30s man).
In this case, the AI does a poor job directing attention both times. The initial response is to let me tell you why a coat is a good idea. The speaker here claims all of the knowledge, and I’ve just got to be ready to receive it.
The frame works, sometimes. Let me tell you why dual-energy CT scans are lower risk than single-energy CT. Let me tell you why worm casings are more effective fertilizer for this garden crop than bone meal. Let me tell you why some technical solution requiring advanced knowledge of a field is what you need.
But, we’re talking about a coat here. Infants are aware of the difference between cold and warm as well as the things that provide both cold and warmth. The AI isn’t offering anything special. If anything, it’s a little patronizing to set my attention on its expertise.
But that’s the catch! The individual lines of reasoning are pretty good. A coat is comfort (I like that) and protection (sounds manly!). Putting on a coat is quick and easy, something I really need on the way out the door in the morning. ChatGPT has given me some good copy here, but I need to sift out the chaff to find it.
Do I Use It?
ChatGPT and other AI writing bots are a lot like any other specialized tool. They can produce results, but using them uncritically probably won’t create great results. That’s not really a new observation. Plenty of other articles already claim ChatGPT is a specialized skill that job-seekers can claim.
But that’s not why you’ve read this far. The question is whether you can use it.
Use-case: Common Products, Common Knowledge
Are you selling coats to elementary school kids (or their parents)? AI is going to create generic language that looks a lot like other copy your audience will find online.
That’s not really bad. You’re probably trying to get folks’ attention first because the field is flooded with everyone selling very similar products at very similar costs. In this case, truly persuasive copy is less important than getting some content out there.
Use AI!
Use-case: Uncommon Products, Specialty Knowledge
Are you selling kids’ coats that are rated down to -20F? AI is still going to create generic language, but you might be able to force it to include some specific benefits you want (or maybe just revise its output yourself).
Generative AI creates text based on things that already exist. The more specialized your product or information is, the more you are going to have to write material yourself. But, for the really common parts (in the case of coats: weatherproof materials, dual-zip, hoods, etc.), it’ll stop you from reinventing the wheel.
Use AI, but prepare to revise!
Use-case: Boutique Products, Niche Knowledge
Are you selling ruggedized, battery-power thermal wraps for outdoor workers in arctic environments? Or, are you otherwise selling or offering something with a very narrow field of competitors?
Don’t use AI! Sure, it can make text quickly on just about any subject. With the right prompting, you can get it to explain how or when AI will take over the world. You can also get it to say it will never surpass humanity. It doesn’t know anything other than the most common combination of words in connection with different subjects.
If your subject is totally new or not well represented across the web, then there isn’t a common combination of words it can spit back at you.
Conclusion?
The more I use, read about, and talk about AI writing tools, the more I think about spell check. For more than thirty years, people have worried that spelling checkers (and their modern variant, auto-correct) are ruining kids’ spelling abilities. That spell-check and auto-correct are writing for us. But, we all know that’s not the case.
More importantly, over time, we’ve all learned when to trust these tools and when not to. We’ve learned when the tool can be really helpful, and when we shouldn’t give a flying duck what it tells us.
The same lessons need to be learned for AI over time. The tool is a great time-saver for some cases, and a terrible fit for others.
