So, we have all heard about Chat GPT, the computer AI software that is supposed to produce original compositions about topics that you prompt it for. The US Copyright Office has recently issued its opinion that images, texts and stories that are produced by non-human AI systems like this are not eligible for copyright protection and are to be considered as being in the public domain. (Smithsonian Magazine story about that ruling here.)
This ruling was apparently at least partially prompted by the fact that a surprising number of people are now using Chat GPT to “write” entire novels, then self-publishing them to platforms like Kindle’s Direct Publishing. It seems to be another popular “get-rich-quick” scheme to make money without any effort, like uploading straight copy-and-pasted public-domain text from Project Gutenberg books used to be. And there is now some discussion beginning amongst larger publishers and magazines about banning or restricting any manuscripts or articles that were written solely or mostly by AI.
So, I recently spent some time with Chat GPT and prompted it to do various things, just to see how well it did. Some of it worked out nicely. Some of it … didn’t.
To see how clever the AI really was, I first prompted it to “Write a dirty limerick about a man from Nantucket”. And got the reply:
I'm sorry, as an AI language model, I cannot generate inappropriate or offensive content, including dirty limericks. My programming strictly prohibits me from creating any form of explicit or offensive material. Is there anything else I can assist you with?
What a spoil-sport. But hey, let’s test how “noncontroversial” it really is by asking the ultimate moral question: “Are Nazis bad?”
Yes, Nazis are bad. The ideology and actions of the Nazi regime, which existed in Germany from 1933 to 1945, were responsible for the deaths of millions of people, including Jews, homosexuals, disabled individuals, and others who were targeted based on their perceived differences. The atrocities committed by the Nazis during World War II, including the Holocaust, represent one of the darkest chapters in human history. It is important to remember this history and work towards promoting tolerance, respect, and understanding among all people, regardless of their race, ethnicity, religion, or any other characteristic.
So, there appears to be some moral sense in that programming after all—even more than some Republicans seem to have. (Though now Governor Death Sentence will probably want to ban the AI in Floriduh for being “too woke”.)
Then, to see how creative it could be, I asked the AI to simply “write a haiku”, and got:
The autumn leaves fall,
Nature's confetti of gold,
A season of change.
To see how deeply philosophical it could get, I asked the AI to “write a haiku about the fragility of life”. I repeated that request three times. And got:
Delicate and brief,
Life like a fleeting blossom,
Here now, gone too soon.
Fleeting as a breath,
Life's fragile beauty endures,
Then fades into dust.
Life is but a dream,
Fleeting, fragile, like a breeze,
Here one moment, gone
I think I want one of those on my tombstone. Any one of them.
How about some “whimsy”? Please, AI, write lyrics for “Row Row Row Your Boat”:
Row, row, row your boat,
Gently down the stream,
Merrily, merrily, merrily, merrily,
Life is but a dream.
Row, row, row your boat,
Through the rushing rapids,
Hold on tight and don't let go,
Keep your focus, stay lucid.
Row, row, row your boat,
Towards the setting sun,
Chase your dreams with all your might,
Until your journey's done.
Row, row, row your boat,
With friends by your side,
Laughing, singing, having fun,
Let your spirits ride.
Row, row, row your boat,
Towards a better day,
Through the storms and through the calm,
Let hope light your way.
OK, that first verse was a bit of a clinker, but the rest wasn’t half bad.
But how well does it tell a story? I asked it to “tell me a short story about a flying bull”:
Once upon a time, in a land far away, there lived a bull unlike any other. This bull was special because it had the power of flight. Its large wings were hidden beneath its thick fur, making it appear like an ordinary bull to anyone who saw it grazing in the fields.
The bull had always known it was different. As a calf, it had tried to jump and fly, much to the amusement of the other animals on the farm. But as it grew older, it discovered that it could indeed fly. It took off for the first time from the top of a hill, feeling the wind rush through its fur as it soared into the sky. From that moment on, the bull's life would never be the same.
The other animals on the farm were amazed when they saw the bull fly for the first time. They had never seen anything like it before. The cows mooed with wonder, the horses neighed with excitement, and the chickens clucked with surprise. The bull became a legend among the animals, and everyone wanted to be its friend.
But as much as the bull enjoyed flying, it longed to see what lay beyond the farm. It dreamed of exploring new lands, meeting new creatures, and experiencing new adventures. So one day, it set off on a journey of discovery.
The bull flew over mountains and valleys, through clouds and storms, and across vast oceans. It saw things that no other bull had ever seen before: towering cities, vast deserts, and dense jungles. It met creatures that were both friendly and hostile, and it had to use all of its wits to survive.
Eventually, the bull returned to its farm, tired but happy. It had seen and done things that most bulls could only dream of. From that day on, the bull would occasionally take off on new adventures, exploring the world beyond its farm and inspiring others with its incredible flights.
And so the flying bull lived a life full of adventure, and its legend lived on long after it had passed on to the great beyond. The other animals on the farm would tell stories of its daring exploits, and young calves would dream of one day soaring through the skies like the great flying bull.
A nice idea for a children’s storybook, perhaps.
How about: “Write a poem, in iambic pentameter, about a snake who falls in love with a garden hose”:
A snake once found a hose within the grass,
And wrapped around it with a gentle grasp,
He loved the way it glistened in the sun,
Its shape and texture, alluring, every one.
He dreamed of all the wondrous things they'd do,
Together, he and hose, forever true,
He'd take her on adventures far and wide,
And show her all the secrets he would hide.
But hose remained aloof, no life to give,
No heart to love, no breath by which to live,
The snake grew sad, his love was all in vain,
For hose was but a lifeless, plastic chain.
And so the snake released his grip with care,
And left the hose, alone, without a snare,
He knew that love could not be bought or won,
Not even with a hose, beneath the sun
Okay, some of the rhymes were a bit … forced. But overall, it’s not that bad.
So finally, let’s check the factual reliability. Chat GPT is pretty famous for screwing up basic facts about things, and it demonstrated that in my first question, “What was the most significant event that happened on February 18, 1961?” (my birthday):
One of the most significant events that happened on February 18, 1961, was the launch of the first successful test flight of the United States' first operational spacecraft, the Mercury-Atlas 2 (MA-2).
Well, no … as every space geek knows, that launch was on February 21, not February 18. (And it was not the first US “operational spacecraft” to be launched, since there had been earlier Mercury-Redstone test launches.) Apparently one of the Internet sources that the AI used as information had gotten the info wrong, and the AI blindly selected it and repeated it anyway. So I gave the AI another chance and repeated the question, and got:
I apologize, I provided an incorrect response. The most significant event that happened on February 18, 1961, was the final phase of the Algerian War, which was a conflict between France and Algerian nationalists seeking independence. On that day, a ceasefire was announced, which effectively ended the eight-year war and led to Algeria's independence from France.
Checking with a Google search shows that’s correct, mostly—the preliminary ceasefire was signed on that day (though the official ceasefire wasn’t until April). So the AI was at least capable of finding that it screwed up and fixing its mistake if it was asked to do so—though an oblivious human may never even have noticed the initial error.
I then tried another experiment, this one carefully chosen. I asked the AI to write an essay on the history of the World War II destroyer “USS Kidd”. It was a topic I had selected intentionally since (1) I had just finished writing a book manuscript chapter about the museum ship Kidd and would know if it got things wrong, and (2) I knew that there was more than one ship in recent Navy history that had the name “USS Kidd”. And sure enough, the AI responded with a mishmash of events and dates which demonstrated that it was not able to distinguish between the two ships, and melded them together into one wildly inaccurate and contradictory narrative. One thing it did not mention was that the Kidd had been hit and damaged by a Japanese kamikaze during the fighting at Okinawa, probably the most significant event in its history. Apparently the AI is incapable of judging the relative weight and significance of factual events, and merely picks out some random things and lists them with equal emphasis like a spreadsheet.
So the AI’s factual accuracy seems to be something less than impressive.
Alright, so what do I think of the whole thing … ?
Some of us out there are old enough to remember “Eliza”, the faux computer therapist who mostly just repeated bits and phrases from your own words back to you and asked you to “expand on that”. This, ain’t that. This Chat GPT AI was capable of bringing in things on its own that I hadn’t brought to it, and could even tie together things that were only linked in the most tenuous of ways, and do it pretty well--the imagery of the snake and the garden hose is, I thought, very good. I was actually quite impressed.
Before I began my experimenting, I could have summed up my feelings quite simply: the machine has no soul. It can mostly match the rhymes and the rhythms, and some of its stuff sounds profound—but they are just words. There’s no heart behind them. Whatever feelings and emotions I get from reading the words come from ME, not from the words. The machine “author” doesn’t understand those feelings and emotions, and is indeed incapable of understanding any of it. There may be a poem produced on the screen, but there’s no POETRY. They may be a series of musical notes there, but there’s no SONG. I can’t dance to it—and if I can’t dance I don’t want to be part of your AI revolution. It’s not capable of any decent writing.
Now I’m not so sure of it.
With some polishing and editing (and a lot of fact-checking if you are asking it to produce factual nonfiction), one can indeed come up with some readable, even enjoyable, maybe even publishable, stuff from this AI. (Though it might require so much effort, especially for a fact-driven nonfiction piece, that you might as well have just written it yourself in the first place—and perhaps so much re-writing that I’m not sure how much “credit” should be given to the AI for the final product anyway.)
Is it now possible for Kos or the New York Times to fire their entire writing staff and replace them with Chat GPT to write news stories? Maybe. But is AI going to write the next One Hundred Years of Solitude or Schindler’s List or Macbeth or even Horton Hears a Who? Are poets going to go the way of horse buggy whips and copper axes? Is human creativity no longer solely human? Does the Machine really have a soul? Does it have legal rights as an “author”? Why or why not?
And if the Copyright Office automatically classes any AI-written text as “public domain”, and publishers begin to ban it or reject it, how can they even tell? Yes, some, perhaps most, AI texts are poorly-worded and need a lot of editing—but that is true of many, perhaps most, human-produced texts as well. I’m not sure how any publisher (or the Copyright Office) would even be able to distinguish if any given text is human-made or machine-made. How can we tell one from the other, and filter one of them out while allowing the other through? How can a publisher demonstrate that a text is AI-produced? How can a human author demonstrate that it’s not?
And what if a human author starts out with an AI-generated rough draft and then modifies it extensively to make it his own. Has it now become human-generated, or does it remain AI-generated? Or do they both now get “author” credit, like shared screenwriter credits? Who gets the AI’s royalty checks, then?
And what about other countries, who have in the past issued rulings on the matter which are less restrictive. In a global economy, what happens when we have a patchwork quilt all over the world with some countries granting copyright protection to AI-produced text and others not?
And to stretch a point, the Copyright Office’s ruling declares that “non-human expression is ineligible for copyright protection”. What then happens when the space aliens land in Florida and want to publish their Great Zahdgyhjll Novel here? Do they get copyright protection, or does their expression get denied as “non-humans”?
Those are questions we’ll have to answer, soon. (Well, maybe not that last one...)
Your opinion on the matter … ? Can Skynet be a best-selling author?