I want to dispute the entire framing of "using AI." That might be a useful framing when you are having AI act as a search engine, but if someone actually uses AI to generate art or text, they didn't "use" AI, AI did it for them. If I go to McDonald's and order a burger I didn't "use" McDonald's as a tool the same way I would use a stove and frying pan to make one at home. Pope Julius II didn't "use" Michaelangelo to paint the Sistine Chapel.
Do you take a photograph, or does the camera take it for you?
I haven't used AI image generation in a way that's significantly different than "do it for me" but for writing there are a lot of degrees of freedom between "AI does everything" and "human does everything". I used Claude for rubber ducking slash brainstorming when I wrote my Tom Lehrer parody ( https://www.lesswrong.com/posts/MiEejaWqC2dbiezTE/i-wrote-a-song-parody ) and did end up incorporating a line Claude came up with when I tried and failed to think of anything better.
I think the major difference is how many deicsions you are making. With a photo you are still making the a lot of decisions about how the image will look. With an AI generated image you are making the same amount of decisions about the image as you would if you had paid another human to do it for you.
In my experience, generating a good AI image doesn't happen on the first try - you'll have to fiddle with the prompt and wording a whole bunch and regenerate it multiple times before you get anything good. Sometimes the AI has dumb hangups that stop it from giving you what you want, like you ask for a specific pose but it gives you a different one that's more common in that type of scene, or it doesn't know what a "pauldron" is and you have to rephrase it.
Not as much work as drawing it yourself, but probably comparable to taking a bunch of photos of an event and picking the best ones to publish.
> There’s incredible momentum towards a world where AI can be used anywhere, for any purpose, with no disclosure, and that’s fine.
This is definitely not true. I write for a living, and if somebody commissioned something from me, and I used AI for it - not as brainstorming, or "hey, what happened to this character?" or whatever, but actual writing - I would probably lose much of my business. The only way I wouldn't is if I didn't/couldn't get caught. Even the stuff I read for free, and enjoyed, when I've noticed AI chunks creeping in, it really kills my desire to keep reading that work.
It's possible to be a pretty lazy writer without AI too... There's one erotica writer that I suspect writes sex scenes by starting with a copy-paste of one that's already been written because of the occasional bizarre error I've found and how repetitive they get.
> There’s incredible momentum towards a world where AI can be used anywhere, for any purpose, with no disclosure, and that’s fine.
Pressing X to doubt on this one! I am skeptical that the current level of access the layperson has to ChatGPT et al will still be a thing in 5 years, and I think it is very much NOT fine that people are passing it off as their own writing/art.
Even though (if current political trends continue) normal people may not have access to the best AI models, they will probably still have access to open-source models which will be worse than the best contemporary models but definitely better than ChatGPT & possibly better than today's best publicly-accessible models.
(I agree, though, that it is immoral for people to present even partly AI-written work as their own, & unless future AI writing gets good enough to be entirely indistinguishable from decently-written human work, even on the level of plot/ideas, people will continue being upset about it.)
If anything I think the opposite is true, that AI inspires intense witch-hunting as people try to guess whether something is human-made. Most people do not want to be reading AI writing or looking at AI art.
> I want to get several different perspectives on a nascent idea, and modern AIs can only give me one perspective.
Unless you explicitly ask it to give you a different perspective! E.g. if I've been arguing something with an AI and felt that it's giving in to me too easily, I've used a question like a "what would a smart and knowledgeable expert who defended the opposite position say in response?" to get it to flip to a different perspective.
> What’s the equilibrium? I expect that some people have already cut back on reading essays, at least from non-established authors. Over time, I expect this will lead to fewer humans writing essays, further increasing the density of AI-generated content, driving more people to cut back on reading, et cetera.
This isn't necessarily true. When I suspect something I read online (fiction or nonfiction) is AI-written, I put it into Pangram to check; this means that even if I have to be alert for AI slop, that doesn't really stop me from reading actually human-written blogs or stories, because I can easily confirm that they are, in fact, human-written. Currently the 4 uses/day of a free account are more than sufficient for me, but if I were a more adventurous online reader, I would at least consider buying a subscription for this purpose. Admittedly this solution could break down, in general if AI advances make semi-reliable AI-written-text detectors impossible or impractical, on particular sites if those are flooded with so much AI-written text that most of what you might read is AI.
AI cannot be forced to drink hemlock and therefore is not a legitimate participant in Socratic dialogue
I want to dispute the entire framing of "using AI." That might be a useful framing when you are having AI act as a search engine, but if someone actually uses AI to generate art or text, they didn't "use" AI, AI did it for them. If I go to McDonald's and order a burger I didn't "use" McDonald's as a tool the same way I would use a stove and frying pan to make one at home. Pope Julius II didn't "use" Michaelangelo to paint the Sistine Chapel.
The Sistine Chapel analogy really hits the nail on the head.
Nobody would call that Pope an artist.
Do you take a photograph, or does the camera take it for you?
I haven't used AI image generation in a way that's significantly different than "do it for me" but for writing there are a lot of degrees of freedom between "AI does everything" and "human does everything". I used Claude for rubber ducking slash brainstorming when I wrote my Tom Lehrer parody ( https://www.lesswrong.com/posts/MiEejaWqC2dbiezTE/i-wrote-a-song-parody ) and did end up incorporating a line Claude came up with when I tried and failed to think of anything better.
I think the major difference is how many deicsions you are making. With a photo you are still making the a lot of decisions about how the image will look. With an AI generated image you are making the same amount of decisions about the image as you would if you had paid another human to do it for you.
In my experience, generating a good AI image doesn't happen on the first try - you'll have to fiddle with the prompt and wording a whole bunch and regenerate it multiple times before you get anything good. Sometimes the AI has dumb hangups that stop it from giving you what you want, like you ask for a specific pose but it gives you a different one that's more common in that type of scene, or it doesn't know what a "pauldron" is and you have to rephrase it.
Not as much work as drawing it yourself, but probably comparable to taking a bunch of photos of an event and picking the best ones to publish.
> There’s incredible momentum towards a world where AI can be used anywhere, for any purpose, with no disclosure, and that’s fine.
This is definitely not true. I write for a living, and if somebody commissioned something from me, and I used AI for it - not as brainstorming, or "hey, what happened to this character?" or whatever, but actual writing - I would probably lose much of my business. The only way I wouldn't is if I didn't/couldn't get caught. Even the stuff I read for free, and enjoyed, when I've noticed AI chunks creeping in, it really kills my desire to keep reading that work.
It's possible to be a pretty lazy writer without AI too... There's one erotica writer that I suspect writes sex scenes by starting with a copy-paste of one that's already been written because of the occasional bizarre error I've found and how repetitive they get.
Is using AI art worse than "I did a Google Image Search and used the first result I liked?"
> There’s incredible momentum towards a world where AI can be used anywhere, for any purpose, with no disclosure, and that’s fine.
Pressing X to doubt on this one! I am skeptical that the current level of access the layperson has to ChatGPT et al will still be a thing in 5 years, and I think it is very much NOT fine that people are passing it off as their own writing/art.
Even though (if current political trends continue) normal people may not have access to the best AI models, they will probably still have access to open-source models which will be worse than the best contemporary models but definitely better than ChatGPT & possibly better than today's best publicly-accessible models.
(I agree, though, that it is immoral for people to present even partly AI-written work as their own, & unless future AI writing gets good enough to be entirely indistinguishable from decently-written human work, even on the level of plot/ideas, people will continue being upset about it.)
If anything I think the opposite is true, that AI inspires intense witch-hunting as people try to guess whether something is human-made. Most people do not want to be reading AI writing or looking at AI art.
> I want to get several different perspectives on a nascent idea, and modern AIs can only give me one perspective.
Unless you explicitly ask it to give you a different perspective! E.g. if I've been arguing something with an AI and felt that it's giving in to me too easily, I've used a question like a "what would a smart and knowledgeable expert who defended the opposite position say in response?" to get it to flip to a different perspective.
I don't think I'll ever use AI for my comments. They're just not worth it.
> What’s the equilibrium? I expect that some people have already cut back on reading essays, at least from non-established authors. Over time, I expect this will lead to fewer humans writing essays, further increasing the density of AI-generated content, driving more people to cut back on reading, et cetera.
This isn't necessarily true. When I suspect something I read online (fiction or nonfiction) is AI-written, I put it into Pangram to check; this means that even if I have to be alert for AI slop, that doesn't really stop me from reading actually human-written blogs or stories, because I can easily confirm that they are, in fact, human-written. Currently the 4 uses/day of a free account are more than sufficient for me, but if I were a more adventurous online reader, I would at least consider buying a subscription for this purpose. Admittedly this solution could break down, in general if AI advances make semi-reliable AI-written-text detectors impossible or impractical, on particular sites if those are flooded with so much AI-written text that most of what you might read is AI.
[Note: I am aware of https://www.lesswrong.com/posts/hrpQxfYvF6CBGWfJX/pangram-ai-detection-software-can-be-evaded showing that Pangram can be somewhat easily fooled; however, the method given there requires that the AI write in a pastiche of some specific other writer, & so I expect that trick to be rare in practice.]
I'm pretty sure it's obvious I don't have an ai writing for me. Editing, yes, writing, unquestionably not.