"Soon, asking a writer if they use AI will be like asking a photographer if they use Photoshop" – e.g., it's a foregone conclusion, and the best artists will generally adapt to using the best tools available
It's similar to music. DAWs and samples didn't kill off music; instead, it made it easier than ever for a teenager with a computer and a passing interest in music to create a song and share it with a world. As a consequence, though, the standards for mixing and mastering have gone up massively; people don't really tolerate bedroom recordings with $10 mics any more. I imagine most amateur musicians in the 90s didn't know what a compressor actually did (I certainly didn't).
Seeing the results of talented artists who are experimenting with AI[0] makes you realize that there's still going to be a massive gulf between skilled artists using SD etc as a tool, versus those who think they can be artists just by putting keywords into an image generation AI and calling it a day.
[0] https://twitter.com/jamm3rd/status/1619896080619159553 https://twitter.com/jamm3rd/status/1633758455952703488/photo... (moderately nsfw I guess)
the fear from the current professionals is that the new ai raised floor is good enough to be usable commercially, and thus, cut a chunk of economic value from the professionals.
There is no fear that AI stops people with the intrinsic interest in creation to stop creating - their ceiling and productivity would sky-rocket as a result in fact. And it would be _these_ productions that wow an audience.
Most small town papers are filled with verbatim Reuters reports.
Not exactly. It's mostly AP reports, not Reuters, covering stuff from outside that small paper's area. If there's a flood of national interest in that small town, it's probably one of the reporters from that paper that will write the national wire story.
That's kind of the point of the Associated Press, lots of small local organizations banding together to offer comprehensive local/national/international coverage.
https://en.wikipedia.org/wiki/Associated_Press:
> Many newspapers and broadcasters outside the United States are AP subscribers, paying a fee to use AP material without being contributing members of the cooperative. As part of their cooperative agreement with the AP, most member news organizations grant automatic permission for the AP to distribute their local news reports. The AP traditionally employed the "inverted pyramid" formula for writing, a method that enables news outlets to edit a story to fit its available publication area without losing the story's essentials, although in 2007, then-AP President Tom Curley called the practice "dead".[6]
and there aren't very many of these kinds of journalists (let's call them "correspondents" to differentiate).
The journalists copy/pasting from Reuters, or the ones that paraphrase a press release, is going to go the way of automation, but i suspect that this amount of jobs is still quite large today, but will disappear relatively soon.
It's one thing to ask for a pong game, it's another to ask for a program that "does a kyc process" or something similarly complex
You're unlikely to be able to ask an AI tool for "flight management software" and get what you want. This stuff is extremely complex, with lots of moving parts and interactions with other components and with government regulations. You may be able to successfully ask for some specific function, but again, in order to understand that level of detail to even know what to ask for, you have to be pretty involved. It can't just be some non-engineer asking for things they don't understand.
Never mind the challenge of certifying the resulting software for use in commercial airspace...
And while I think AI tools may be a help (not unlike Stack Overflow), I think it would be a really big mistake to start replacing novice junior engineers with AI tools, even if we could do so, because we need senior engineers who really understand this stuff, and most of the training for aerospace work is on the job.
Assembly isn't that hard. You'd have still gotten a job, but your employer just wouldn't have gotten as much productivity out of you (so they'd either have had to hire more programmers or write less software).
Whenever a customer ask me to develop an app, i don't consider that he did the job. I did it.
So I ask the AI to rephrase it for me. Now I can pick and choose an option from those the AI offered, or I can just keep my version if it turns out it's good enough. I feel this is an acceptable use, and you can still maintain authorship, just like book authors maintain authorship after an editor has overhauled their writing.
Your DAW example would work if DAWs could generate samples, beats, and whole songs by the click of a button. But are you truly the author, then?
They absolutely can, at least with 15-20 clicks rather than 1 at the moment. Drag in drum loop, drag in melodic loop, use Scaler2/Cthulhu/Captain Chords to come up with a chord progression, use a synthesizer preset to play the chords, drag in vocal sample, put on Ozone AI Mastering on the master bus, done. If you sent me back in time to 1990 with that track, people would be really impressed.
As to whether you can claim you're the author, this gets debated a lot. I'll just paste the famous goat farming quote.
> I thought using loops was cheating, so I programmed my own using samples. I then thought using samples was cheating, so I recorded real drums. I then thought that programming it was cheating, so I learned to play drums for real. I then thought using bought drums was cheating, so I learned to make my own. I then thought using premade skins was cheating, so I killed a goat and skinned it. I then thought that that was cheating too, so I grew my own goat from a baby goat. I also think that is cheating, but I’m not sure where to go from here. I haven’t made any music lately, what with the goat farming and all.
But how sterile will art become? Already most people become inured to mass synthesthesia, where pureed sterilized Disney-fied content is fast becoming the unavoidable norm in so much of our media (e.g. pop and country music). Today's AI yay-sayers are receptive to a synthetically generated window onto the world are looking only at the first step in what will surely be a progression -- from a highly variable world of human-wrought artifacts to a homogeneous digitally simplified model of the world, one that will evolve into something ever more 'palatable', until... what? Is that a trend you would welcome?
Not I. Especially with Microsoft and Google and other gigacorporations calling all the shots in the absence of informed enforced guardrails to constrain how this rapidly evolves.
It’s a bit more like a hand crank that builds a lopsided toolshed at this point. Which is absolutely incredible, to be honest, and couldn’t have been anticipated by many people five years ago, but doesn’t necessarily entail that the cathedral-button will arrive any time soon.
We kinda have to be patient.
Look at the history of woodworking for inspiration.
It used to take human sweat to make flat straight surfaces. Something like a walmart flatpack bookshelf would have been VERY expensive to make 1000 years ago. Industrial tech made that cheap and ... now its seen as tacky and out of style and only for poors.
The result culturally this century is an avoidance of "plastic wood" finish, faux finishes and faux surface prep that look old and worn and hand scuffed, and a strange hipster fetish for "live edge" raw bark on finished product. Sure, in 1875 one might have predicted based on mass production trends that furniture in 2020s would be all right angles, smooth as glass mirror, photographic "ideal" woodgrain instead of real woodgrain, plastic-y lifetime thick film finish. But thats only for poor people who shop at walmart, nobody with money buys that stuff.
I suspect what AI can cheaply produce will rapidly culturally be considered lower class and trashy. Oh, sure, completely AI generated movies look very nice and sound very nice and are ever so trendy and well written, but special effects and yesterdays memes on todays big screen generated entirely by computer algorithm is only for poors. Now, us rich cognitive elites only attend live theater plays. Aspirational middle class types watch recordings of theater plays (admittedly maybe some AI generated). You wouldn't show off your 'wealth' and 'class' by taking a date to an AI generated movie, that would be insulting, you might take her to a live theater play... I'm sure there will be some AI incursion, maybe the lights will be run by AI or AI painted backdrops or some actors will lipsync their lines to AI, or roughly every 5 years we have to tediously suffer thru a new generation of rewritten modernized classics and it'll be worse in the future with AI... but in general human will be the status symbol because it'll be expensive.
I suspect being able to ask an AI to generate non-copyrighted commercial quality music will mean massive noise pollution (much like every room required 15 digital clocks in the 80s and every room required 15 super bright blue power LEDs in the 00s) followed by the only people making money in music will be live humans playing physical instruments.
Poor people try to get by and don't spend money and when they do it'll be on infinitely cheap AI commodities. Rich people spend money on unique stuff, not mass produced AI shovelware. Aspirational people spend money trying to look rich so they will spend all their money on non-AI stuff.
I don’t find this holds true for hip hop, the most streamed genre, whatsoever. There are countless examples of people making it big recording on a cheap mic in their mom’s bed room. One example is Chief Keef. A much more prominent example is Kanye who had terrible mixing on The Life of Pablo to the point where he released patches[0] for the album. He also recorded parts of Jesus is King on an iPhone[1]. To further the point, bootleg recordings and demo tapes of unreleased songs in artists’ back catalog continue to be popular with devoted fans. Average people like good music, even if the quality of the mixing and mastering is not stellar.
Even The Beatles released songs with minimal takes from a live rooftop performance on Let It Be; hardly an ideal recording set up. Although they surely had better mixing.
[0] https://archive.org/details/2016-the-life-of-pablo-updates
[1] https://genius.com/a/kanye-west-says-20-percent-of-jesus-is-...
It's called bedroom punk now.
Asking a photographer if they use Photoshop creates a framing where the artist still did went out and took a picture to create the work and then Photoshop only modifiers their original creation.
But you could just as easily say "Soon, asking a writer if they use AI will be like asking a photographer if they have a camera". That sounds ludicrous, but that is exactly what generative AI offers: the ability to create content essentially ex nihilo.
For example: sometimes I will pass a paragraph through an LLM and ask it to return it in a more "active voice". Then I _manually_ edit my original paragraph, if I wish, to make it seem more suitable for my current goals.
This is sort of like Photoshop for photographers.
AI doesn’t “do everything for you” any more than word processors, music sequencers, or CAD does. AI allows people to operate at higher levels of abstraction, is all. Yes, fewer people will type repetitive copy in to word processors. But that’s fine.
You are mistaking activity for productivity. Few people go into writing because they enjoy typing. AI offers leverage, which is the most amazing thing for anyone interested in producing work.
Ask any professional writer: revisions are a chore. Authors often say a novel is 10% writing, 90% rewriting, and they don’t enjoy the rewriting part.
So, IMO, contrary to your doomer take, I think we will see more people producing better work because they can focus on the parts they love and not do the grind that was only ever a necessary evil.
(The exception being those who love the grind and not the product)
The end has come.
Netflix also infamously used AI to generate background images for an anime, which is work that previously would have put food on the table for illustrators.[0]
It's hard to look all of this and think that all the jobs are going to stick around. I know at least 5 copywriters who have made their entire living as adults writing banal junk for small companies. All 5 of those careers will vanish in the coming years.
[0] https://www.cined.com/netflix-uses-ai-to-generate-anime-shor...
Sooner or later studios will use it to leverage the work of their artist. If the same quality of animation requires half the staff to make it, some will use it as a cost reduction, others will use it to make better and more complex art. Maybe we'll finally get 60fps Anime instead of the 10 - 24 that is typical now. Your favorite series that was cancelled after a single because they didn't have budget, now they can make 5 season of it with the same budget.
The problem is that it automates the best parts and transforms the user into a curator/manager.
Some people want that, but other enjoy the creative part more.
To bring it back to photography, that's the problem with digital, you blast through hundreds of photos and spend most of your time selecting and editing at the computer.
With analog, and especially wet plate or direct positive paper, it takes a while to compose your picture and you only have one chance to take it and develop it. It's very easy to mess anything which is what makes the process more meaningful. And you're left with a unique copy, not an artificially limited single copy.
I almost never use the prose word-for-word, but goodness is it helpful.
It supports a context length of 32k tokens (approximately 25,000 words or about half a novel).
Of course, it's not my place to say who should or shouldn't call themselves a writer, but I'll simply personally respect someone more if their stories aren't ghostwritten by the AI.
I don't think "AI" is the way most people would complete that sentence. Perhaps "Word"? Or "LibreOffice" if "Photoshop" were replaced with "Gimp"? Personally I use Emacs or Vim. You could incorporate AI into any of those tools, but how exactly? And would it be a core part of the functionality? And would it be something that the best artists make much use of?
If you used AI to make the spelling/grammar checker more reliable, probably most people would use that (assuming no privacy problems), but if there's an AI-driven autocomplete for sentences probably a lot of people, including the best artists, would turn it off, just like I turn off the autocorrect on a word processor today.
If an AI can complete the sentence, then perhaps the sentence isn't worth completing, at least if we assume that the reader is as clever as the writer and has similarly sophisticated tools. Perhaps the AI-driven tool should instead put some kind of wavy line under the second half of the sentence to indicate that it's boring and obvious and doesn't need to be there.
It’s tremendously useful technology for many domains already and you can see the stepwise refinements that will permeate many parts of our lives. Big money corporations are productizing those elements already.
But to be as fair to the pessimists as the optimists, its actually still a very big leap from Modjourney and GPT-4 to something that becomes the next camera or typewriter in terms of ubiquity. Because we saw a huge leap recently, we feel close and excited — and we might prove to be in hindsight — but we also might see that there are some hard conceptual limits that we won’t see anyone break through for another fifty or hundred years.
For image-making, there is a decision in the prompt and selecting images, but that is very different to making a decision about each color and brushstroke, and working to finish a painting. It's orders of magnitude more difficult and why great masters are celebrated. Creating stuff with AI will suit certain people, I definitely don't think necessarily the 'best' will automatically use it. It does seem to take away a lot of the fun of making images and truly original work will always push quite far outside the training set.
For literature, it could be interesting, how authors use new tools, where and in what sense. Maybe have more conclusive plots? Less inconsistencies? Have AI imitate dead poets' style?
We will value the classics more I guess, since they were done the old fashioned way.
New tools open doors, not shut them.
Another argument I keep hearing (most recently from pg), is that we'll always need non-AI training data. That, too, doesn't follow. Training new models on synthetic data does not mean we get stuck in a particular mode or style. We'll continue moving, improving, and trying wildly new things. Bootstrapping with synthetic data doesn't block evolution - it enables faster evolution, even. (I'm using synthetic data to train new models to great effect.)
People are angry that we've lowered the opportunity cost barriers and so they're expressing their frustration.
It's a good thing that life's choices no longer fence us in as much. Everyone should get a chance to learn how to express themselves through art with the new regime of tools. Despite changing economics, there will still be a top 1% that do better than the rest of us.
I think it's a great thing.
This is all explained well by "Drawing on the Right Side of the Brain". People apply their childhood drawing symbols to realistic art in their teenage years and get frustrated. They never learn to see with an artistic eye and quit.
I was skeptical at first but after going through a couple of the drawing exercises I was convinced. They are so cleverly constructed to make you step outside of your normal brain and just draw, it's kind of amazing the results.
I'm never going to be an artist professionally but I do actually draw as a pastime now and I have improved substantially.
I think if I had practiced more as a kid and teen I may have pursued art professionally.
Part of the problem is that adults see a kid upset that their drawing isn't very good and say things like "That's ok, you're just not an artist. Not everyone can draw" and people internalize that. It's supposed to make the kid feel better and it might cheer them up in the moment, but it also convinces them that they cannot learn to be better at things they are not immediately good at.
Toxic Feel-Good shit.
It’s about expressing and finding yourself and developing a soul through creative action, not just a squirting out a picture that looks “correct”.
You only shortchange your own soul’s development by trying to skip the hard steps that it takes to develop a skill like that.
Because you can make an image on a computer that looks like it was drawn does not mean you can draw. Hell, I’m not even sure it means you can express yourself. The inputs you contribute when using an AI prompt are so minimal compared to the informational input required to create something in the physical space.
Don’t let anyone convince you that it’s an exclusive act. Find a pencil and get at it.
Ultimately though, I realize that trying to get everything what you want from a single prompt won't work. Tokens have implications, and they influence each other too much. Long term, one needs more than just single prompts, and the tools are still evolving to support that.
Still, you'd be surprised at how much you can express even now.
You're already the most complex model and model-maker—the procedure of developing a skill through a creative feedback loop and its output can be vastly more edifying for yourself and others. Don't take my word for it. You can only know by doing it.
These models have their place as useful tools, but are in no way a replacement for the experience of creating and all its challenges. You can't climb a mountain by paying someone else to do it for you. That analogy begs the question—what is the true value in climbing a mountain? What is the value in creating?
And saying that, I don't disagree with the idea that learning how to draw causes a larger mental development than writing prompts, and that prompters will forever miss out on that human development. It's just that not all people are motivated enough to "climb the mountain" of being a decent enough illustrator, whether hobbyist or professional. Some of them are happier with the process of getting decent results quickly than stumbling over themselves for years just to reach the point of what AI can accomplish now. Maybe the process of drawing is painful because of the way they treat the process, discouraging them from persisting, but it's hard to change people, and they're capable of defining failure on their own terms. And those people will always exist. Not everyone who wants to draw will end up enjoying it or succeeding after their preconceived notions are shattered. That in my mind means there will always be a target audience for generative art tools.
I think AI art could reveal a lot of diminished self-confidence towards life in general, and I could imagine that some portion of the people flocking to the generative programs would have become manual artists if they believed in themselves more, but they don't, and AI has won them over because they get to skip steps to have the result. Maybe they didn't want to do though the challenges of self-development to gain a new skill, but imagined artistry as nothing more than a fun hobby, and AI art aligns with that notion (mistaken or not) than doing months and months of practice, in a way that no technology has been capable of up until this point.
But does that mean those people have failed as artists and soul-searchers? I think that's still up for debate. AI could quickly become the new kitch (if it isn't already) but they're happy creating it, and many are happy consuming it. I don't define myself as an artist and have no insights as to the "value" of AI art as opposed to human art, since I'm a programmer who's stumbled into this other world and I feel that's a domain I shouldn't be making solid declarations about. But I believe both types of art have their place. Maybe a collaborative project between artists of both denominations would be useful in showing the unique qualities of both styles of creation.
A 'correct' image on a computer? Then you'll get just that and little more.
But I'm suggesting there is much more to be gained by taking part in a creative act as an experience and its myriad of facets.
Taking the stance that a language or 2-dimensional geometric model is a replacement for the latter is gross hubris. At best it can describe a potential outcome, but it does't provide any of the rest.
To the point about "accomplishing art", there is an old saying often repeated: "art is never finished, it is only abandoned". So if you think art is just something to be accomplished, a box ticked and then profited from then I think it is worth studying the subject further.
* If you grab some pencils and scribble all over a page, you'll get something that looks interesting for a minute, but it's not the picture in your head, and after a while all scribbles start to look the same.
* If you throw some geometric shapes into blender and run a render, you'll get something that looks interesting for -ok maybe a couple of minutes-, but it's not the picture in your head, and after a while all geometric shapes start to look the same.
* If you throw a few tokens into a prompt and hit generate, you'll get something that looks interesting for -now maybe a couple of hours?- but it's still not the picture in your head, and after a while all generated images start to look the same.
Each time, if you want to actually get anywhere close to the picture in your head, you're going to need to practice, and you're going to have to learn the medium, and you'll have a lot of experiences along the way. "There's no mistakes, just happy little trees", right? But the actual path -and the happy little trees you experience along the way- will be rather different for canvas, pencil, blender, or stable diffusion.
Have you ever tried to hold a 7-dimensional, non-linear, chaotic landscape in your head whilst trying to resolve nightmare images of deformed limbs and eldritch landscapes?
Note: I don't really care if people use AI to generate anything but they should be forthright about doing so.
IMHO this is absolutely the case, and has always been the case. Being paid as an artist, on the other hand, is something entirely different, or having one's art recognized as good by other people. But I am sure everyone remembers other more popular artists telling them they're "doing it wrong" in art class, on forums, or even all the way back to preschool coloring books if someone would deign to color outside the lines.
There have always been those who seek to gatekeep what art is and who they view as legitimate participants in it. I suppose everyone does in some sense. But I am very excited by the prospect of art as a personal developmental journey becoming more open to everyone.
Beyond that, if you think AI algorithms will replace art, you don't understand art. It replaces much of the commercial utility of many artists and creating many types of images, but the idea that it threatens art, generally, is preposterous. People who think art school is only about learning to physically make art are like people who think computer science degrees are only about coding. Most of what you learn in art school is conceptual thinking, communication, really deeply seeing things in a way that most people don't, and stuff like that.
Factories didn't replace potters when they started making pots and dishes or sculptors when they started making cheap home decorations-- it just meant the artisans no longer made money creating a commodity because most people were satisfied buying chintzy shit cheap. Many potters who lived through that change probably died in poverty because of it. Talk big numbers about the way industry shifts all you want, but not everybody can switch to an entirely different category of employment just because someone taught a computer to do it.
There was never anything stopping anyone from picking up a pencil and learning to draw. The idea there's some inborn "talent" for art is absurd. There is only people who practice and people who don't.
What you are actually "democratizing" is the ability to produce finished products without ever putting in any kind of work.
Which means you are absolutely devaluing the people who actually did put in the work.
It is a terrible thing. A miserable race to the bottom of human expression.
Jokes on you, I agree with everything you just said, fully unironically
The web is way worse where people have to put zero effort into uploading their content
The best places on the web are the places where people have crafted the page they want.
Drawing was for anybody who put in the work.
You may as well say football wasn't democratized until the release of FIFA International Soccer in '94.
Artists, by and large, don't do art because they enjoy having art--they do art because they like /the process/ of producing art. If that process can be done faster and better by AI, then yeah, sure, they /might/ be able to still do art for a living (some artists will be able to leverage their experience to maintain an advantage; other, less flexible ones will lose work)--but the work they do will likely not be commensurate to the work they were doing before, and will likely be less enjoyable to them.
The thing that worries people about AI is that it'll make all creatives into middle-managers.
Then I spend 95% of my hobby time tediously sanding boards and applying finish and this is NOT the fun part of my hobby.
I will say the only time I've gotten compensation directly as been for the latter because nobody likes doing that stuff.
I'd say AI will make creatives into micromanaging middle managers, the type that gets way too involved in a critique of your every decision is still going to micromanage, they'll just be harassing an AI instead of humans.
Fundamentally the creative act of making book cover art has always been specification. You can say to an artist "make me a book cover looking like ..." or you can hire a guy and give him no direction and trust him to decide what to do. But the creative act was the specifying, not the mechanics of turning that idea into printing ready art.
Another problem you run into is AI is as currently produced, a groupthink enforcement technology, and most people will be turned off by that. The blandest least offensive corporate dining experience is likely McDonalds. How much do hipster foodies prefer going to McDonalds over ... anywhere else on the planet? AI can only create the ideal big mac, and you can make money shipping big macs, but only selling to poor people, everyone else wants something else, even if in a strict corporate inhuman sense its inferior in every way to a big mac.
Hence I disagree. Even of the training data curators are perfect, there is a risk of "model inbreeding."
And the curators are not perfect. In fact, they seem to prioritize (by necessity) volume and speed over quality and curation when training these LLMs and diffusion models.