Business Musings: AI And Mediocre Work

To Get a Free AI Audio Version of this blog, click here.

I write better in Spanish than I speak. Much better. Some of that is because I have read Spanish with varying degrees of competency since I was eight. Many of my problems speaking the language come from either a lack of vocabulary or, in the past year, an inability to access that vocabulary in the moment.

However, much of the reason that I write better in Spanish than I speak is because of AI. And because of an acceptance among the professors at the University of Nevada, Las Vegas, that the students will use the website tools available to them—in English and in Spanish.

I am a dyslexic. I have never been able to spell well. In fact, I spell better in Spanish than in English because Spanish makes sense. (The sounds of the language appear in the spelling for the most part, unlike English.) I still transpose letters in Spanish, but I will look at them later and think, Whoops. That would sound weird.

I never do that in English.

When I write a paper in Spanish, I use a Spanish spell check and grammar check. Two years ago, when my classes started requiring essays, I would write in Spanish until I couldn’t find the words or the proper structure. Then I would put an English phrase in brackets and continue.

I’d go through the essay, pull out the brackets, and put that phrase in three different translate programs. I wouldn’t go with the most common translation, because some of the programs are quite literal. I would go with the one that looked like something I would write—right or wrong.

Nowadays, I dumb down my essays a bit. If I can’t figure out how to use the English phrase, I figure out a different way to say the same thing in Spanish.  I still look up the odd word. (My favorite from last semester? Parachute, which is paracaídas, or, in English, “for falls.”)

If I’m feeling really insecure about a paragraph, I will put it into a translate program, translate it back into English to see if it says what I think it does, then translate that English into Spanish and see if I even get close.

Mostly, I’m doing okay. Sometimes I’m still way off. The correct version, however, goes into the paper.

The distance between my spoken Spanish and my written Spanish is vast. Vast enough that I, were I some professor faced with a student like me, might think that I let someone else (or AI) write my papers for me.

However, every professor I’ve had has required us to do some free writing in class and turn that writing in. Sometimes that writing is in an exam and, last semester, it was almost daily about something we read.

My vocabulary in the free writing sessions is better than my spoken vocabulary, but it is not great. And I occasionally forget an accent mark or I transpose letters, and don’t catch it. But my written Spanish remains consistent. It’s clear that the woman who writes the papers using the AI tools I mentioned above is the woman who has done the free writing in class.

When ChatGPT filled the news, we were starting a new semester. My Spanish prof had just handed out an assignment to write a major research paper as 50% of our grade. She mentioned the AI program and added, Just let me know if you use it.

She made no judgement about it at all. Just acknowledging that the tool was there, and she wasn’t going to fight it.

I spoke to my research paper partner, who is taking something like six classes this semester, and asked him what his other profs said about ChatGPT. He said that, with one exception, they all said the same thing.

Just let me know if you use it.

It sounds like the university itself has already moved to an acceptance that the tech is changing and it will have an impact on what students do.

The tech is changing in all different creative venues, from AI art, which we will discuss next time to AI audio, which I’ve been focused on in these past few weeks, to AI writing.

These tools will be available from now until the internet breaks (if it ever does). I’m sure that as spell checkers became available and as grammar checkers started being the default in most word programs, professors worried. How would they know if their students can write well?

I’m sure the profs of language programs were quite worried.

I would have been. In fact, early on in my renewed Spanish career, I tried to operate without the spell check or grammar check, and then decided that I was the only one who cared about that.

To be completely honest, I learn more by going back and forth between my words and the computer program’s offerings. I can see now which programs are literal translations and which ones actually offer nuance. I also see how easy it is for those translation programs to misunderstand something in basic Spanish and create an amazing word salad in English.

These are tools, tools that I will probably use for the rest of my academic career.

I could write an entire paper in English, hit translate on the most reliable program, and have some rather cruddy Spanish translation of what I have done. It would confuse words and gender, and sometimes just give up and use the English word when I would want a Spanish one.

Editing that is much too much work for me. Even if I took the time to correct it, I’m sure I would miss something.

Google Translate has existed since 2006. It’s been augmented many times and now uses a different basis than the early programs, but it’s okay for basic stuff. As I searched for the date when Google Translate was launched, I discovered a few companies that offer professional translation services that are machine-based.

Those are services that let the machine translate the document, and then human beings edit the documents to, as one service said, make sure that there are no dangerous translations.

(Their example was in the medical field. A mistranslation could cost a life.)

These services claim they are cheaper and faster than a standard human translation service. I don’t know. I haven’t priced them. But they do seem like a compromise that businesses who want some document translated into a dozen languages might use.

These are AI programs which have been around for 17 years. Add smart phones into the equation, and things change even more. Las Vegas is a multilingual city. I’ve watched people speak into a translation program on their smart phone and have the phone spit out a translation into English so that the person near them can understand them.

Sometimes that translation is word salad. Sometimes it’s close to what was said. Often it’s just what the two people trying to communicate need.

In other words, tools have their uses. We just have to figure out what they are.

The new AI programs in art, writing, and audio are tools. Nothing more and nothing less. Before I go farther, let’s acknowledge that they will create all kinds of problems.

A podcast interview I read this morning (probably thanks to an AI transcription program) between Shamitu Basu, host of Apple News In Conversation, and Nilay Patel, editor in chief of the Verge discussed some of those problems in a way I hadn’t seen it before.

Patel noted that when Microsoft marketed Excel for the first time, “entire floors of accountants at major companies around the world” disappeared. Their jobs went away.

Patel predicts this for copywriters in ad agencies who are producing “midlevel copy” for things they don’t care about. He notes that people in the bullseye of the change experience something frightening. They know their jobs are going to disappear. That’s a horrifying feeling, which is happening all over right now.

But he points out something else, which truly caught my attention. He said,

The way I’ve been thinking about it is that it reduces the cost of making mediocre work to almost zero — and that’s how we need to think about it. ChatGPT does not produce the best text on the internet. It produces acceptable text at a high rate. Midjourney and DALL-E and whatever else do not produce the best art that has ever been made. It produces acceptable art at a high rate for low cost.

The interview talks a lot about “mediocre work.” He expands on it, by talking about some of the changes that will have to happen in the culture as a result of these programs.

But I want to pause at the words “mediocre work” because it really gets at what I’ve been thinking about.

As a Spanish student who has spoken English her entire life, I do not expect my Spanish writing to come anywhere close to my writing in English. I’m trying to get through classes and write good-enough essays to get a grade.

I’m not trying to be a worldclass Spanish writer. I’m doing deliberately mediocre work…partly because that’s all I can do.

However, when it comes to audio, I can do excellent work. In 2021, I took a voiceover class at a studio here in town, partly to get my chops back and partly to learn what I had missed in the past decade or two. (I had missed a lot. I still need (and want) their editing classes.)

After we did our first cold read, the instructor asked me if I had had professional experience, because it showed. Yes, I was out of practice. Yes, I hadn’t used that particular tool (my speaking voice) in a very long time. But I still had (and have) the ability to do high quality voiceover work with the right kind of practice.

Yet, I’m now using AI voices to put out an audio version of this blog. There will be AI versions of some of my books in the near future. I explained a lot of my thinking on this in the first research post on AI Audio, which went live at the beginning of this month.

I’m fully aware that these audio blogs are not up to my professional standards for voiceover. I would change a lot if I were trying to make these better than adequate.

As I said in previous pieces, I just wanted the work out there so that people who want or need a quick way to listen to the blog have something besides the AI voice on their computer or smart speaker to help them listen to this weekly blog.

The responses I got when I put out the first audio blog based on the weekly blog surprised me. I realized that my own experiences as a radio announcer, engineer, and voiceover artist got in the way of my expectations.

Not expectations of the way that audio works, but my expectations of the way that people listen.

A number of people told me they would never listen to AI voices, which I knew going in. Even more had an opinion even though they told me they don’t like podcasts or audiobooks. Sigh. Okay, then.

But some folks truly surprised me. They only listened for less than a minute. Some didn’t even get past the introductory voice and wondered why I had chosen a male voice as my voice. My radio training came in there. I love introductory voices and added one for clarity.

I’m going to be doing a number of audio things in this little blog for clarity’s sake. That’s part of my training. A straight audio version is often confusing. In fact, if you compare the text across, you’ll find some minor tweaks that I made to the audio version because the words as I initially had them worked better on the page.

Then there were the comments that really surprised me. The ones that objected to the pronunciations, and blamed it on the AI. Some of that is the AI voice and its limitations, but some of it is just what you get with audio.

For example, I constantly correct voiceover artists, much to Dean’s annoyance. Someone either mispronounces a word on a commercial or the writing itself is terrible, and there I am making some slight adjustment or commenting on the mispronunciation. It annoys me, particularly when it can be so easily avoided now. (Back in the day, the pronunciation of some words were mysteries; you didn’t have the internet to give you the various pronunciations.)

Then there is the tone of the voice. Or the rhythm of the performance. Those all existed as benefits and problems before AI audio. There’s a voiceover artist on one of my audiobooks whose work I actually loathe, but they have a massive following on Audible, so I’ve sucked it up, even though they mispronounce common names from other cultures and their tone is nasal and grating (to me).

A lot of the problems that people complain about with ChatGPT, for example, are problems that existed before ChatGPT. For a few weeks, every writer I know had ChatGPT write a bio of them, and each bio had things wrong. Some of the things were spectacularly wrong, such as telling one writer that she had died years ago. That’s very science fictional…

However, most of what ChatGPT did were tiny annoying errors. Giving a writer an award they hadn’t been nominated for. Assigning them a book title that had nothing to do with them. Getting their age wrong or the number of books they had published or…or…or…

Everyone complained and pointed to how bad ChatGPT was. But that very week, the local paper had run an article on my neighbor who runs a variety of businesses here in Las Vegas. There was a paragraph in the middle that claimed he ran several businesses that, in reality he was an investor in, not the person who came up with the vision or who manned the desk. That paragraph also claimed a business he’s currently seeking funding for actually existed.

These problems creep up in journalism all the time and always have. Especially if the piece is an interview, which isn’t based on research. The reporter has to have heard the interviewee correctly—and then the reporter needed enough background to understand what the interviewee was telling them.

Reporters do their best to avoid these errors, but everyone is on a deadline, and budgets have been slashed. Most papers no longer have factcheckers, so these problems have multiplied over the years.

Those multiplying errors become truths. I’ve seen articles about me, written before ChatGPT, that have given me awards I’ve never earned and claimed I had written books I had never heard of. A lot of writers, especially in the early post-F&SF years, accepted my successor’s attempt to wipe out my history at the magazine. My Hugo-award winning work there is often left out of any biography about me, partly because of that partially successful attempt to pretend I had not existed.

We’ve seen yet another example of what I’m talking about with the Wired hit piece on Brandon Sanderson, which I will not link to. That piece is filled with misinformation, including a hint that Brandon might suffer from a psychological disease because he writes so much. If the idiot who wrote the hit piece had even bothered to look up said disorder, he would have realized that there was no way that Brandon, a successful New York Times bestselling author, had it. People who have it aren’t writing things that make sense.

Sigh. Grrr.

That hitman writer didn’t do mediocre work. He did terrible work. A chatbot would have done a better job.

But let’s get back to mediocre work, shall we? AI audio, AI writing, AI art are all delivering mediocre work, work that still needs a bit of human intervention to even bring it up to the standards of adequate. I spend 20 minutes or so on the audio getting it to a level that satisfies me.

There are places, though, where mediocre work doesn’t belong. The best way I can illustrate this for you is using an article from The Hollywood Reporter.

At the end of an article published in January, at the height of the ChatGPT terror, The Hollywood Reporter assigned ChatGPT to write some pitches, using a standard trope. X movie meets Y movie. (Die Hard meets 80 For Brady, for example.)

The magazine then printed the pitches and had Big Fish writer, John August, analyze them. It’s worth reading all of the pitches and his comments.

His final analysis is truly telling. He wrote,

These read like the summaries on the top sheet of script coverage. That’s not a knock; I used to be a reader at TriStar, and they’re not always easy to write. As far as the ideas themselves, they’re all taking ‘X meets Y’ far too literally. Screenwriters don’t start with this shorthand; they find their way to it to explain their take.

In other words, these little paragraphs are mediocre. Script coverage might be hard, but it’s not something that anyone tries to do in an artistic way—especially if they want to keep their jobs.

The key sentence is the last one. Screenwriters don’t start with this shorthand. They use the shorthand to explain what they have already done.

What I find amusing about all of this AI terror, particularly in the writing community, are the writers who use ChatGPT to give them ideas. Or to use as a basis for their next project. Or to help them plot.

One writer got very angry at some newbies who decided that ChatGPT would save them time. She wrote that she doesn’t want to save time when she writes. She wants to enjoy what she’s doing.

Of course she does. She’s an award-winning writer with a hell of a following. She entertains herself and thereby entertains others. That’s how it’s done.

This other crap? Those poor writers are using a tool that can barely achieve mediocre to help them be creative.

What she missed—what a lot of the writers who are worrying about their fellow writers are missing—is that these ChatGPT-shorthand writers have no hope for long careers in the first place. These folks are focused externals, like getting help plotting, rather than learning how to do it organically. They want some computer program to help them write, which is like me expecting to sell my Spanish essays to some major literary outlet in Latin America.

It’s that ridiculous.

Why start with something guaranteed to be mediocre? Are you saying your skills aren’t even good enough to be considered adequate? And you’d rather use some kind of shorthand than learn?

Yep, that’s exactly what these folks are saying.

What I’m finding fascinating about the AI audio as I delve deeper into it is that there are some built-in tools that I want to play with, as a creative and an audio engineer.

They’re still not going to make my audio blog rise to the level of art, but they do provide me with opportunities that would either cost me a small fortune to do properly or opportunities that I can’t access without a lot of time and effort.

Multiple voices, for example, are expensive when they’re real people. So the fact that I can use them here without much effort makes me happy.

But doing other things? Like an actual audio play of any quality? That’s not really possible.

And there are many things in the audio world that I don’t want to do with AI. This tool does exactly what I need, what Patel called reducing “the cost of making mediocre work to almost zero.”

My problem with audio is that I would have done work too high quality for a quick and dirty blog post, and wasted hours, if not days, on a fifteen-minute segment. That’s not worth my time.

But what is worth my time is making sure that the audio fans out there get a good-enough product to make it to next week’s blog.

That’s the proper use of a new tool.

Yes, there are still a lot of problems. And I will deal with a few in next week’s blog that creatives need to know about and think about.

But if this post does nothing else, I hope it convinces you to think of these advances in technology as tools to bring work to the level of mediocre. If you want to be better than that, don’t use the tool.

Either do the work yourself or hire it out to someone great.

We will get used to these new tools, just like we got used to Excel spreadsheets and language translation programs. They will become part of our lives.

Right now, we’re in the messy beginning, sorting out all the details. And that might take quite a bit of time.

****

Speaking of learning how to improve your writing, WMG Publishing is running a Kickstarter which has a lot of great online writing workshops. The Kickstarter is for a starter kit that will introduce you to the first books in ten of Dean Wesley Smith’s series. You’ll get the starter kit with each reward. The kit is also a writing revelation. If you want to learn how to hook people into a great book series, take a look at what Dean is doing.

The Kickstarter ends about 24 hours after this blog goes live on the website, so hurry.

This weekly blog is reader supported.

If you feel like supporting the blog on an on-going basis, then please head to my Patreon page.

If you liked this post, and want to show your one-time appreciation, the place to do that is PayPal. If you go that route, please include your email address in the notes section, so I can say thank you.

Which I am going to say right now. Thank you!

Click paypal.me/kristinekathrynruschr4e to go to PayPal.

“Business Musings: AI And Mediocre Work,” copyright © 2023 by Kristine Kathryn Rusch. Image at the top of the blog copyright © Can Stock Photo / Amaviael.