Margaret Atwood: My Voice and Mind Are Not Replicable
theatlantic.com
theatlantic.com
I think Margaret misses some of their utility. You would never just squeeze out a novel with AIs as they currently stand. They are far too primitive and the context window if far too small right now.
However, they are excellent at speeding up the process of fiction writing with humans involved.
You get a draft together, but you feel like it could be more dramatic. So, you take a few scenes that you've written, put them in the AI, and then ask it to make them more dramatic. It comes back with okay-ish ideas. You iterate again. And again. Finally, you have something that really feels like what you're looking for.
Then you send that out to your editor or friends and family.
Instead of getting very high quality feedback in a week or a day (lucky!), you can get middling quality back in minutes. Yeah, it's not the absolute best, but it gets you out of your headspace and gets you iterating faster. You then send that better draft out for much faster feedback. It's like the rubber duck every programmer uses, only it talks back.
There are other ways that I tend to use these AIs, but the idea is the same. You're not using them point-blank. You use them as a part of the process, and it speeds up that process as a result.
Now, of course, she's not wrong here about publishers. They'll squeeze these AIs out for just crap and try to sell that, and, unfortunately, be successful. When author royalties and advances drop to $0, you can tolerate a lot in the long tail.
Thankyou!
The trick to get the AI to produce more iteratable content is to use a really long prompt. The more specific you can be, the better. So, don't say 'in the style of Hemingway' to get something like Hemingway. Say something more like 'in the style of the character of Timeaus, in Plato's Timeaus, as translated by J. Harward' to get something that feels more like an ancient platonic dialog. Don't say ' I want a heroic character who fights dragons', say something like ' I want a maiden in distress over her recent divorce from her alcoholic husband of 7 years, Fred, and the incarceration of her mother for the 4th time for drug abuse. This maiden lives in upstate New York, and ....'. Give it as much as you can.
And again, you use AI as an accelerant, not as the generator of the initial ideas. You still have to do that on your own. You can have the AI tweak things, suggest things, but having it make things up from whole cloth isn't the best. Take what it gives you, throw half away, iterate on the rest, repeat. Like with a longer prompt, if you give the AI the entire text that you need a hand with, and then ask it very specific things, you will get better results.
https://www.theguardian.com/books/2016/aug/10/speculative-or...
Sure, sci-fi is often very speculative, but lots also includes (for example) mystery elements, yet it would be strange to necessarily call all mystery stories set in a slightly future or alternate-contemporary timeline "sci-fi"
I guess you could roughly taxonomize stories on "socially speculative" and "technologically speculative" axes. A high value in the latter often pushes the former up, but not necessarily the other way
Surprisingly, nobody ever read Asimov for the positronic brains, either.
"Wtf?" -Ursula
--- I will give Atwood some slack, as less than half her bibliography is it. But still, evaluating the chasm between her and chatgpt is kinda what she wants on her terms.
Her essay is attention-seeking nonsense. She can't predict the future of art-generating AIs. Nobody can. This way she can write a "mea culpa" and admit to being wrong to get some additional attention for herself and her works once AI renders human production of new works largely pointless—because nobody will be able to verify it's human-generated rather than machine-generated, and human-generated works will cost more.
FTA:
> But judging from the attempt recently made with one of these entities—“Write a Margaret Atwood science-fiction short story about a dystopian future”—anything more complex and convincing is as yet beyond it. The result, quite frankly, was pedestrian in the extreme, and if I actually wrote like that, I would defenestrate myself immediately. The program, so far, does not understand figurative language, let alone irony and allusion, and its flat-footed prose was the opposite of effective storytelling. But who knows what the machines might yet achieve? you may say. I’ll wait and see. Maybe they’ll at least turn out a mediocre murder mystery or two.
The game AI naysayers like to play is that anything not made by a human is inherently mediocre, because value is largely subjective, and they implicitly define away the possibility of a great AI creator because knowing it was made by AI ruins it for them.
The same narrative existed for go, even the fall and winter before the Alphago–Lee Sedol match. The AI beat Fan Hui, but was still mediocre and would never challenge top professionals for the foreseeable future. There would be inherent limits an AI would run up against because it lacked some quality necessary to be a top-level competitor (or artist).
They are advancing fast and furiously, my husbands D&D group uses AI for story telling and narratives, and it's been quite compelling, interesting, and dare I say even funny and insightful.
My answer is that yes, I would care. For the same reason I would prefer the original Mona Lisa compared to an atomically-perfect replica. There is something special about unique individuals and objects, and this is not something that can be replicated by a machine.
of course. but what happens when you can replicate 1:1? would you still feel the same way? will the person that was replicated feel wrong for it? i don’t think things are as clear cut in this scenario.
We can look to people who have been adopted and how they feel about the situation as examples, a lot of people will continue to yearn for knowing who their biological parents are. It creates almost two forms of love, not lesser just different.
Unless we get replica births!
Does the fact your individuality might not be 100% replicable matter and also who does it matter too ?
I'd argue it really doesn't matter for most hollywood and consumer products. Art for art's sake is niche and irrelevant to the vast majority of people trying to scrape by. For example, people think sitcoms need to be funny, but that's not quite true. Lorne Michaels from SNL, learned the hard way that Sitcoms need to provide comfort first and foremost to be popular.
Art and writing will likely move back to the old world way of requiring patronage than relying on hollywood style consumerism as people use AI to create their own "comfort" food style media and fiction.
What a sad outcome in the pursuit of making the world a better place