As an academic, I can tell you that developments in AI text generation are causing a certain amount of panic. True to form, teachers are concerned about cheating. My great hope is that the advent of such tools will finally prompt the long-overdue shift in education from product, toward process!
As a writer, I feel like the subject deserves our attention, and a place in the Craft, Talent, Technique thread of the forum.
I visited
https://app.neuro-flash.com/aiWriter
and prompted the AI with the following:
An ape with telekinetic powers pilots an atomic spaceship to Alpha Centauri. Along the way she falls in love with one of the frozen, human, passengers.
This was its first response:
My name is Koko, and I am an ape with telekinetic powers. I was born on a planet orbiting a star in the Alpha Centauri system, and my life has been one of adventure ever since. I was just a young ape when I first discovered my telekinetic abilities. I remember vividly the day that I used them to move objects for the first time. My father was so proud of me, and he immediately began teaching me how to control and focus my powers. It wasn't long before I was using my powers to do all sorts of things, like piloting atomic spaceships. And it was on one of these ships that I met the love of my life, a frozen human passenger named John.
As our ship hurtled through space towards our destination, I couldn't help but feel drawn to John. There was something about him that was so special, and eventually I realized that I had fallen in love with him. Although I knew that it would never be possible for us to be together in the traditional sense, I didn't let that stop me from loving him with all of my heart.
This is its second:
The ape with telekinetic powers had been alone on the atomic spaceship for so long that she had almost forgotten what it was like to feel love. But when she found one of the frozen, human passengers aboard the ship, she couldn't help but feel a deep connection to him. As the two of them journeyed together towards Alpha Centauri, they slowly fell in love with each other. Though they knew that their relationship could never be fully realized due to their different forms, they cherished every moment they had together. In the end, as they gazed upon the beautiful alien world that was their destination, they knew that they had found something special in each other that would last forever.
A couple of observations:
1. I feel like my prompt has more of an AI generated "feel" than the replies themselves. I literally ran a sort of program in my head, like a Mad Lib:
• A _noun_ with _super power_ etc. etc ——
If you haven't built a similar scenario output machine in your own head, I recommend getting one!
2. The texts generated miss out on (at least) one crucial element of storytelling: Delay.
Perhaps because they don't expect to go on for very long, they each get to the points of the prompt rather directly, with none of the resistance crucial to narrative.
I should mention that I don't intend this analysis as any kind of comfort for writers banging it our on their own —I'm sure these systems will evolve rapidly!
Thoughts?
@DonMarkmaker
I've played around with ChatGPT, and the fiction it creates is very generic. I doubt machine learning is going to come up with original ideas and put that into a professionally publishable story any time soon. It'll struggle with many nuances that make a good story, like the resistance you mentioned. Some of the output was good for laughs (e.g. with a few inputs, I prompted it to write a flash fiction about a group of my friends that made us laugh, but partly because the narrative jumped about).
The non-fiction output is very dry, repetitive, often obvious, and gets facts wrong. I've found it good for summarising books, both fiction and non-fiction, though even there it gets things wrong. If ChatGPT is all students use to cheat, they'll be caught out. Facts need verifying and text needs rewriting. If five out of fifty students handed an essay in written from text prompt output, I'm pretty sure it'd be obvious to the teacher. It wouldn't prepare students for exams either (though I am with you on process in education).
Machine learning is trained on "Big Data" which often leads to issues - what the masses think rather than individuals or nuanced views. Training data often creates output that comes across as racist, sexist etc. I tweeted about the system Google use for their SafeSearch earlier this year, and pictures of women are much more likely to be considered "Racy" whereas pictures of men never were in my sample, but were much more likely to be highlighted as "Violence": https://twitter.com/AlexHarfordSEO/status/1493579016711942146
It will get better, but I don't think A.I. going to replace good writers any time soon, although it's worth writers investigating as there are ways it can help.
ChatGPT is estimated to cost OpenAI up to US$8million a month to run, but think of all the data from all the people using it that are helping to improve the systems!
I used ChatGPT to write simple PHP code for my website that was error-free.
The art generation is capable of some very good results, and I think it will take artists' jobs. Sadly the law never keeps up with technology, and it's stealing artists' styles, which I think should be considered theft. Throughout history, artists have 'stolen' from other artists, but the scale and ease machine learning does it is frightening.
(edited to updated the OpenAI cost fact, as I found out the reliable source was not so reliable after all and humans make errors too)
35: - R R R | 36: R HM R R | 37: HM HM HM SHM | 38: HM HM HM HM | 39: HM HM HM SHM
On a related note, I've just seen a post here I think was written by A.I. They're fun to spot. I've seen a few over the past 2-3 years, but it's best not to jump to conclusions in case they're not A.I. written. And there could be genuine situations like a non-native English speaker wanting to contribute and using A.I. to write posts and replies.
35: - R R R | 36: R HM R R | 37: HM HM HM SHM | 38: HM HM HM HM | 39: HM HM HM SHM
fiction it creates is very generic
I think i's being used as a plotting tool. Interesting article:
https://www.theverge.com/23520625/chatgpt-openai-amazon-kindle-novel
@DonMarkmaker
fiction it creates is very generic
I think i's being used as a plotting tool. Interesting article:
https://www.theverge.com/23520625/chatgpt-openai-amazon-kindle-novel
Interesting indeed. When I tried to get it to write in the style of a particular author (like Stephen King or Ursula K. le Guin) it spewed out similar rubbish each time.
Using it for blurbs is the kind of thing I found it occasionally useful for.
I think there's a danger that if every author overused it for anything (even plotting), it'd dampen their creativity, and we'll end up with nothing original ever again. The big companies like Google, Microsoft, and Apple want to control our lives with A.I; telling us what to wear, eat, listen to etc. That's the eventual aim. Not a spoiler because it's the start, but the kind of thing that happened in Ken Liu's story that was published 13 years ago: https://www.lightspeedmagazine.com/fiction/the-perfect-match/
I like this quote from Jess Peck: "The current level of response from GPT-Chat is amazing. I’ve argued that it is probably about the same level as a low-effort human being. But that’s just it-- we already have low effort content out there. Don’t you want to make something good?"
https://jessbpeck.com/posts/artificialintelligence/
35: - R R R | 36: R HM R R | 37: HM HM HM SHM | 38: HM HM HM HM | 39: HM HM HM SHM
If we look at the history of tech, tho, this stuff is bound to get better. I'll wager, at some point a screenplay or a piece of music or a short story generated by an AI will move you to tears...
No?
@DonMarkmaker
Just keep checking the 'I am not a robot' box. After all, we are only human.
Today's science fiction is tomorrow's reality-D.R.Sweeney
HM x5
Published Poetry
2012 Stars in Our Hearts
Silver Ships
When I used it, all it did was seriously bloat the things I already said. So, if my input was "man dancing on moon saves family of moon bears", it would continuously talk about how the man on the moon saved moon bears and everyone was happy and great. There's no dialogue, no actual thought or emotion into it at all.
Until AI is sentient AI (which defeats it being AI because by definition, AI is supposed to be sentient), then I doubt we have to worry at all!
~ I honestly believe that good stories write themselves. You can always start writing a story, but the characters tell their own tales and if you're lucky enough, you get to merely be in the audience watching everything unfold. ~ 8/1/2022
***
WotF - 2022, V39 Q4: R
WotF - 2022, V40 Q1: Submitted; Q2: Work in Progress;
@AnnaX You can request dialogue. Generally results are better with more instructions, which makes it arguable who's writing the story after all. 😀
If we look at the history of tech, tho, this stuff is bound to get better. I'll wager, at some point a screenplay or a piece of music or a short story generated by an AI will move you to tears...
No?
If so I think it'll have had direct human assistance, and A.I. won't do something on it's own without massive plagiarism.
35: - R R R | 36: R HM R R | 37: HM HM HM SHM | 38: HM HM HM HM | 39: HM HM HM SHM
@alexh I am currently stealing from two fabulous sources, so I'm not sure what to make of your "plagiarism"
@annax visual artist friends of mine are getting into writing python scripts for AI systems. I think what these things are missing is a complete UI. Then you'll get results!
Keep in mind, folks, this stuff is brand new. Give it 19 years... Hell, give it SIX!
@DonMarkmaker
@alexh I am currently stealing from two fabulous sources, so I'm not sure what to make of your "plagiarism"
From a human or machine perspective? From a human perspective, we all put our own spin on writing, so even if we take ideas from elsewhere, no one else would have created whatever we create. It's unique to us. And the stories (or any creative works) that are the best often have personal meaning to the creator, while machines aren't sentient.
It's the ease and scale that machines can duplicate which is where my biggest concerns about plagiarism come from. Not from us humans.
The KYD stories in Wulf's Super Secrets thread are a great example of how everyone comes up with something different from the same prompt.
I don't see that being the same with machines, although maybe there will be ways around it such as training on specific datasets. But then it'll have been a human that picked the dataset, which goes back to my point about how much of a story machine learning would actually be writing.
Either way, it'll be interesting to see how it all turns out.
35: - R R R | 36: R HM R R | 37: HM HM HM SHM | 38: HM HM HM HM | 39: HM HM HM SHM
I haven't tried any AI writing programs yet, but I've been listening to my stories being read to me on the Autocrit site.
I've begun using that same function in Word, and both of the AI's misuse and mispronounce common English quite frequently.
Today's science fiction is tomorrow's reality-D.R.Sweeney
HM x5
Published Poetry
2012 Stars in Our Hearts
Silver Ships
All I can say is the genie is notoriously hard to get back in the bottle!
@DonMarkmaker
