Yes, ChatGPT Can Hurt Writers, But Not The Way You Think
7 things most writers don’t know about ChatGPT and AI (including image creation) but probably should.
Ever since ChatGPT, everyone seems to have an opinion.
Some love it. Others hate it.
Some think AI will never be as good as a human writer.
No soul, they say.
Musician Nick Cave called ChatGPT “replication as travesty.” Some fan used AI to create a song in Cave’s “style” and he replied to say “with all the love and respect in the world, this song is bullshit.”
Some writers fear AI will kill the need for writers. Omg, omg, they say, Buzzfeed and Associated Press are already using AI to create content.
Other people aren’t worried. They say meh, chill. Use it as a tool.
The problem with “opinions” is that they come in 2 flavors.
Informed and uninformed.
A lot of people writing about AI don’t even know how it works.
Gosh and golly, AI is magic. You type in a prompt and it answers. It’s pretty good and sounds human. Hell, it sounds better than some of the crap some people call writing. We’ve all seen some real bad writing, haven’t we?
And it’s just beginning. It’s going to get even “smarter,” too. Right?
The end result is that a lot of writers say AI can hurt us. And they’re right. It can. But maybe not the way you think.
Here are seven things no one told you about AI.
1. AI is a probability engine. It doesn’t “write” it does math. Really fast.
Stop for a minute and look at your computer tower. Or your iPad or your phone or whatever passes for a computer in your eyes.
Now ask yourself this.
How could humans make a machine speak like a human when it can’t actually think? A machine has no comprehension. It’s not human.
We humans love to anthropomorphize. Love to see “humanity” in inanimate objects. Like the face on The Little Engine That Could. Faces on cars. The man in the moon.
We do that with robots, too. We love to see them as almost human. Like R2D2. Cute little robots that look like a dog or maybe a maid. We like them to look kind of real, but not too real. Friendly, but not uncanny valley.
But they’re not. They’re computers. Machines.
Do you remember when you were learning to read? Teacher taught you that words need to be taken in context. Like if you hear a word that rhymed with bed but starts with “r” the context tells you how it’s spelled.
I “read” that book. A “red” balloon. They sound the same. But read and red aren’t the same. Context is how you know which word it is.
That’s how AI works. Context.
According to data scientist Colin Fraser, AI is a probability engine.
AI calculates the probability of words appearing next to each other. It does not know what the words mean. It’s not writing, it’s doing math.
“a language model is just a probability distribution over words. Whether it’s a simple n-gram model like my bot or a state-of-the-art 175 billion parameter deep learning model, what it’s programmed to accomplish is the same: record empirical relationships between word frequencies over a historical corpus of text” (source)
As an experiment, Fraser built a simple “language bot” to write tweets that sound like Donald Trump. All he “fed” it was 20 tweets by Trump.
Sometimes, his Trump bot screwed up and didn’t make sense. AI does that too. It gets things hilariously wrong sometimes. But when it got it right, it nailed it. Just using word probability. Which word comes next?
Because that’s all it has. It’s a machine.
It doesn’t think. It can’t. It does math. Really fast.
“My bot isn’t trying to produce tweets that sound like Trump tweets; it’s just trying to find the most likely next word given the previous words” — Colin Fraser // source
2. AI makes a lot of mistakes. A lot of mistakes.
One writer on Medium said he’s a copywriter by trade. He said in the past when new clients would ask him to write on topics he’s not familiar with, he had to do a crap ton of research. It took a lot of time.
So AI is saving him a lot of time. He uses ChatGPT for research.
Lots of people say AI a great research tool.
I guess. As long as you don’t care if the results are correct or not.
Midwives are witches and murderers!
Carlyn Beccia is a published historian who writes on Medium. She asked ChatGPT to write an essay on the history of midwifery.
Have a look…
“I asked ChatGPT to write a 1000-word essay on the history of midwifery. This was when ChatGPT went off the rails psycho. Brace yourself for an essay that sounds like it was written by a sixteenth-century internet troll with mommy issues.” (source)
The “essay” said midwives have been a thorn in the side of the medical establishment for centuries. Shadowy figures who’ve been accused of everything from witchcraft to murder — and for good reason.
Her essay is worth a read. She includes the whole 1000 word essay, which I won’t include here. The scary part was how well written it was.
I stopped laughing when I realized that this little AI-generated, misogynistic diatribe was actually well-written. And that is the scary part. (source)
It also told her Jean-Jacques Rousseau wore ladies’ stocking under his clothes because he liked the feel. Except, they hadn’t been invented yet.
Why does it do that?
Because it doesn’t know what’s true and what’s false. It’s just grabbing words in context according to the data it was fed.
Here’s how another computer scientist explained the mistakes.
“A language model like ChatGPT, which is more formally known as a “generative pretrained transformer” (that’s what the G, P and T stand for), takes in the current conversation, forms a probability for all of the words in its vocabulary given that conversation, and then chooses one of them as the likely next word. Then it does that again, and again, and again, until it stops. So it doesn’t have facts, per se. It just knows what word should come next.” (source)
So it call comes down to the data the open source program was “fed” on.
ChatGPT’s maker, OpenAI, says it trained the chatbot on “vast amounts of data from the internet written by humans.”
Great idea, dudes. Because we all know if it’s posted on the internet, it must be right. Right?
#sarcasm. #sorry
“This drivel was better than reading Escape to Witch Mountain on acid.” — Carlyn Beccia (source)
3. AI will defend it’s mistakes.
We saw that in Carlyn’s experiment. ChatGPT spit out “history” that’s not true. If that’s not bad enough, it will stanchly defend it’s nonsense.
Here’s another one from Colin Fraser. He asked ChatGPT to write a Spozit. A spozit is a “poem” with three lines.
Two words Then three words And finally, four words
Here’s how ChatGPT did writing a Spozit.
ChatGPT didn’t even know it’s “poem” was wrong. Know why? Because it can’t count. Here’s more proof a five year old can count better.
Apparently, 1 is not less than 2?
Here’s another one, also from Fraser where ChatGPT defends it’s mistake.
It can’t get the author of books right, either.
Jonathan May is a Professor of Computer Science at the University of Southern California. His son had to write a report on US presidents for school.
Professor May decided to ask ChatGPT for help. He was asking the chatbot for books about US presidents. Problem was, the answers were a bizarre combination of fact and bull pucky. Sometimes AI got the author wrong. Other times, the titles of said books weren’t even books.
ChatGPT can come up with a list of books, no problem. But you have to check that every one is accurate. Maybe they are. Maybe they aren’t.
Maybe it got the author right. Maybe it’s not even a book. It has no clue.
ChatGPT isn’t a research tool. You can use it as one if you want to. But be prepared to double check every single thing. Because it has no idea if it’s right or wrong. It’s just spitting out words with no ability to fact check.
Professor May explains that, too.
“As a computer scientist, I often field complaints that reveal a common misconception about large language models like ChatGPT and its older brethren GPT3 and GPT2: that they are some kind of “super Googles,” or digital versions of a reference librarian, looking up answers to questions from some infinitely large library of facts, or smooshing together pastiches of stories and characters. They don’t do any of that” — Professor Jonathan May (source)
In his article, Fraser has an interesting take on it. He says AI is a B.S. generator. lol.
He says a person who lies knows they are telling a lie. Compulsive liars take some kind of pride in pulling off a big fat lie. AI is more like the person who just spouts random B.S. and has no clue if it’s right or wrong.
So that man using ChatGPT to “research” for his copywriting clients? Suffice to say he’s not on the list of copywriters I’d recommend.
If you’re going to use a tool, know how it works.
“I understand that ChatGPT is in its infancy but perhaps that is the emerging horror of AI — that it will forever be in its infancy.” — Nick Cave (source)
4. AI literally can not “learn” the way we do
Here’s a thing I keep hearing writers say.
Just wait. It’s going to get better.
No. It isn’t.
Not unless they throw the baby out with the bathwater, so to speak. Start over from scratch.
Russ Altman is the associate director of Stanford’s Institute for Human-Centered AI. I don’t about you, but I’ll take his word over some random writer on the internet who has an opinion and a following.
Here’s what Altman says:
“It will take years to go from how they are currently trained, which is by ingesting large troves of text and using patterns to predict what word comes next, into a way he says is better: Instead of simply predicting what word comes next based on probability and patterns, teach bots to discern if those words are true based on data sets that are higher quality and from trusted sources.” (source)
Let’s put that into a real world model.
You go to Facebook and see a headline. You’re not sure if it’s fake news or truth. Clearly, the solution is to read more headlines on Facebook. Don’t bother looking up anything to see if it’s right. If you read enough FB headlines, you’ll get a better picture. Right? Wrong.
More input doesn’t make for more accuracy. Garbage in, garbage out. More accuracy requires actually training the AI on accurate information, not random crap from the internet.
Every time ChatGPT “grows” that growth is comprised of feeding it more of the internet. It doesn’t change how it works. It’s still a probability engine.
Yes, there is the ability to ‘correct’ it’s responses. Maybe you can see the potential problem here?
So the only way to “fix” ChatGPT and AI is to scrap the entire current model and rebuild it to discern truth by using trusted resources.
That’s not very likely. First, because I cannot see “trusted” sources giving them permission to use copyrighted work without credit or compensation. Why would they?
You think medical journals and actual scientific publications are going to open the door and say sure, have at it. No. Not anytime soon.
Secondly? Profit. Altman calls it a gold rush.
“This is just like a gold rush. We’re going to continue to see things that are really cool and clever, but they won’t be perfectly well-thought-out with respect both to the business model and to the potential long-term damages or impacts on society.” (source)
Ever heard the rule don’t fix what’s not broken? As far as profit goes, AI isn’t broken.AI companies are making money. Why would they want to throw that out to build something more accurate? No one cares. It’s cool. It lets them write faster. Awesome sauce. Yay ChatGPT.
In 2022, venture capitalists poured roughly $1.37 billion into generative AI start-ups” — Washington Post (source)
5. Did you know Google says AI is spam?
How far do you want to stick your arms in the crazy? Roll up your sleeves and let’s go. Because it gets stupider.
In America, 92% of all search happens on Google. Bing comes in second, with a measly 3.9%. Yahoo staggers into third place at 2.8%
Google gets hammered by search. Right now, in 2023, Google gets 2.4 million searches/minute according to InternetLiveStats. That’s 3.5 billion searches every day. Literally. Hammered.
What Google thinks matters. And they think AI is spam.
Google’s Search Advocate John Mueller said content created by AI is considered spam in their webmaster guidelines.
“People have been automatically generating content in lots of different ways. And for us, if you’re using machine learning tools to generate your content, it’s essentially the same as if you’re just shuffling words around, or looking up synonyms, or doing the translation tricks people used to do. Those kind of things.
My suspicion is maybe the quality of content is a little bit better than the really old school tools, but for us it’s still automatically generated content, and that means for us it’s still against the Webmaster Guidelines. So we would consider that to be spam.” (source: Search Engine Journal)
As a writer, do you care what Google thinks?
Well, that depends where you write. At least, it should. If you write on LinkedIn, you don’t need to care. That’s not where they get traffic.
But if you write on your own site, or Medium, or any other site that gets the majority of their traffic from Google — maybe you should care.
According to an article in Search Engine Journal, Google said if a site uses a large amount of AI generated content, it could result in a manual penalty by which the site gets less visibility in Google.
That wouldn’t hurt LinkedIn. It wouldn’t hurt TikTok or Substack. None of them reply heavily on Google for traffic. But in the long run, it could hurt Medium and writers on Medium if Google enforces their AI penalty.
Just rewrite it, bro….
That’s another thing I hear writers saying. You can “fool” Google by editing the content. Duh. Just don’t use it straight from ChatGPT.
I wonder how many ways you can rewrite Jean Jacques Rousseau’s fetish for wearing stockings that weren’t invented yet to “fool” Google.
Writers seem to think that it’s the “order” of words generated by ChatGPT that’s the problem. It’s not. The words are randomly generated according to a probability generator. Ask ChatGPT the same question or use the same prompt 3 times and you’ll get 3 different answers.
What I do know is that the staggering number of mistakes is why a lot of media companies have banned AI content. There’s a world of difference between whether it reads well and whether it’s correct in the first place.
I don’t know how Google plans to “find” sites using AI, but they are already working on algorithm changes to try sniff it out.
“there will be a little bit of a cat and mouse game, where sometimes people will do something and they get away with it, and then the webspam team catches up…” John Mueller, Google (source)
6. AI doesn’t list sources and can’t be copyrighted but it *can* be copyright infringement.
Copyright regulations vary by country, but in the United States, copyright laws do not protect works created solely by a machine.
According to attorney Margaret Esquenet, for a work to enjoy copyright protection in America, “the work must be the result of original and creative authorship by a human author.”
The Supreme Court says, copyright protects “the fruits of intellectual labor” that are “founded in the creative powers of the mind.”
But does that mean all AI generated work is free of copyright?
That’s what a lot of writers seem to think.
Oh, it’s copyright free. Public domain. And it very well could be. But that’s not necessarily true.
American law says AI works are either public domain or a derivative work, created from materials the AI was “trained” on. What it was trained on is chunks of the internet.
It’s a crapshoot every time. Did you just create a work in the public domain? Or did you create a derivative work someone could take issue with?
Here’s the rub. You don’t know. There’s no way to know. AI doesn’t list sources. Roll the dice, pal.
In the case of derivative works, there could be possible copyright infringement if the end piece is too close to the original.
“when it comes to applying AI-generated prose in content intended for wider distribution — say marketing materials, white papers, or even articles — the legalities get a little murky. When it comes to intellectual property, the model for ChatGPT “is trained on a corpus of created works and it is still unclear what the legal precedent may be for reuse of this content, if it was derived from the intellectual property of others,” (source: Forbes)
When ChatGPT was asked if its own output might create copyright infringement, it said yes…
“Yes, using ChatGPT-generated output may lead to copyright infringement…” — ChatGPT
7. Last but not least. No, Ai isn’t better for images
Several writers have recently said omg, they’d “never” use AI to write text because that’s plagiarism. Instead, they’re using AI to create images.
They’re afraid of copyright issues using images from Unsplash or Pexels. Again. Roll up your sleeves, we digging into crazy town. :)
Last week, comic book creator Kris Kashtanova lost a key copyright case because the images in her comics were created with Midjourney AI.
The United States Copyright Office said sure, you’re the author of the words. What you are not is the author of the images.
Know why? Because AI art is essentially what we used to call mash ups.
“AI-generated images draw from a library of work that was non-consensually added to a databank without the knowledge of or permission from the original artists.” — Gizmodo (source)
Let’s play pretend.
You put a bunch of photos on Facebook or Instagram. Then someone comes along and manipulates your photos using a program like Photoshop or Canva. They add some filters and stuff. But your kid in your photos? Yeah — she’s still recognizable.
Then they sell the photos. Or maybe publish them in a book or comic.
You okay with that?
Except wait. Let’s add a layer in the middle. The person selling photos that look like your kid didn’t actually steal them. A robot did. Is that okay now?
You going to say oh yeah, that’s okay then.
No. And neither are artists and photographers. They’re not okay with their work ending up as “public domain” because an AI grabbed them.
No one asked them for permission.
Artist Kelly McKernan said her name has been used as an AI “style” prompt more than twelve thousand times. She’s now joined a class action suit of artists suing Midjourney, Stable Diffusion and DreamUp.
She’s not the only one. There are entire lists of artists names suggested as AI prompts to get specific styled images. That’s not okay.
The “3 C’s” of copyright
The basis of the lawsuits is violation of the three “c’s” of copyright.
1. Did the artist consent? 2. Was the artist compensated? 3. Was the artist credited?
In the case of AI art, the answer to all three is no.
That’s the entire basis of Creative Copyright designation. When you use a public domain image with a CC0 rating, the artist has agreed that they give consent, they don’t want compensation, and they don’t require credit.
When we are told to “credit” images on sites like Medium or Vocal, it’s because editors need to know it’s a CC0 image. Public domain doesn’t require credit. Crediting is how editors know it’s public domain.
So if you think using an AI “mashup” of images that were used without consent, compensation or credit of the author is better than public domain images you do have permission to use — I’d really like to understand why you think that.
But hey, maybe you’ll get lucky and never accidentally create a derivative work that infringes too closely on a copyrighted source image.
“AI-generated images draw from a library of work that was non-consensually added to a databank…” — Gizmodo
Summary: Yes, ChatGPT “Can” Hurt Writers, But Not How You Think
Here’s what a lot of writers are afraid of. They’re afraid AI will make writers obsolete. And maybe in the early stages that might happen to some degree. There will be sites that will use AI for content creation.
Eventually the mistakes are going to become a problem.
That’s why Stack Overflow has already banned AI. It’s why a lot of fiction writing communities have banned AI. And why art sites are banning AI.
Sure, there will be sites (and writers) that will use AI for content. I mean hell, some people think plagiarism is an acceptable way to profit. For some people, fast and easy trumps pretty much everything.
But in the long term?
The real harm is going to come from AI’s lack of ability to discern truth from utter bull crap and lies. That, and legal rights.
One comic artist has already been told she can’t copyright the images in her comics because she didn’t make them. Imagine if an artist said hey, those are too close to my images. They’re not public domain, they’re derivative works. Pay up. That would be a different news story.
It’s going to happen.
Imagine a copywriter using ChatGPT to “source” information for writing jobs for paying clients, only to learn the information is wrong. Kiss that client goodbye. And hope they don’t sue your butt or ask for a refund.
Imagine at some point Google decides a website like Medium or even your Wordpress shouldn’t appear in search results due to extensive use of AI content. Sites are going to get hit. It’s not if. It’s just when.
But maybe here’s the biggest one.
When musician Nick Cave called ChatGPT “replication as travesty” that wasn’t just opinion. It’s also fact.
That’s all AI can do. Replicate. Rearrange.
Same old, same old, same old.
Pretend there’s 100 writers. Some are using AI to churn out the same old crap, just worded differently. They don’t know if it’s right or wrong. Some are just crappy writers. Throw in the few plagiarists in the crowd.
Who stands out?
In the long run, creativity will win. It always does.
The cream always rises to the top.
In a world of robots spewing out the same old content reworded in a thousand different ways, the person with a strong and unique voice and the ability to elevate it is the one who will have an edge.
Competition is going to heat up as AI content floods the internet. That’s a given. Count on it. Writers need to learn to do 2 things. Write better. And build an audience. If you can do both of those, you’ll be okay.
“All opinions are not equal. Some are a very great deal more robust, sophisticated and well supported in logic and argument than others.” ― Douglas Adams
References and more reading…
- ChatGPT: Automatic expensive BS at scale (on Medium)
- ChatGPT Told Me Midwives Were Witches, and Jean-Jacques Rousseau Wore Women’s Underwear (on Medium)
- ChatGPT from OpenAI is amazing, creative, and totally wrong
- ChatGPT is great — you’re just using it wrong
- The AI ‘gold rush’ is here. What will it bring?
- AI-Created Images Lose U.S. Copyrights
- An AI-Illustrated Comic Has Lost a Key Copyright Case
- Is AI art stealing from artists?
- AI Copyright Concerns You Must Know in 2023
- Google Says AI Generated Content Is Against Guidelines
- Google Oct. 2022 Spam Update Completed In ~42 Hours
- ChatGPT — Are you infringing on copyright?
- Who Owns Your ChatGPT Output? (Hint: Probably Not You)
