Thursday, January 19, 2023

AI: Art, Images, and Artificial Intelligence

A particular kind of artificial intelligence has been appearing in the news and commentary for several months. Stable Diffusion (released in 2022) allows the user to input a text prompt ("robot artist holding paintbrush") and a deep-learning artificial intelligence (AI), trained on an enormous database of images and styles, will respond with a computer-generated image. 

Artists and commentators are wringing their hands about this development, concerned that human expression, the jobs of artists, and creativity will all be terminated.

Stable Diffusion isn't even the first thing like this. It had predecessors in things like DALL-E, but Stable Diffusion marked a step up -- it works better, and almost anyone can run it on their home computer with minimal technical skill. 

Other related AI systems like ChatGPT (itself an evolution of GPT-3) have people worried and excited that soon we won't know whether we are talking to a computer or a person and that writers will be replaced by machines as well.

I think this is a bit overblown. I also think some artists' jobs will be replaced, but I also believe this is inevitable and perhaps not that bad.

John Henry and the Steam Hammer

Prompt: "robot artist holding paintbrush"
This all sounds like John Henry and the steam hammer again. The relentless, soulless corporate machine against the honest blood and sweat of a human. John Henry won (in some tellings), but he died doing it. And the steam hammer (and its descendants) are what we use for drilling tunnels, chopping down trees, and doing other difficult work. 

We don't give the steam hammer or most replacement of humans with machines a second thought. One reason is because we assume this kind of technological replacement is vastly safer and more efficient solution than having individual, unaided humans doing dangerous, tedious, back-breaking labor. It usually is, and society benefits in many ways. 

Another reason is a kind of societal and behavioral inertia -- if we can create a machine that can replace human labor (with something perceived as cheaper, faster, safer, or "better"), we not only should, but we must. In fact, we feel as though we cannot stop it. There is a kind of inevitability built in. Someone will do it eventually. So we end up with billions sunk into self-driving cars, robot tree-fellers, and robot vacuum cleaners. 

Perhaps a better question would be if these things need doing at all, or doing at faster and greater scale. 

The New York Times ran a piece where human artists more or less griped about Stable Diffusion and other AI-powered image generators. It was what you would expect for this kind of thing. Heartfelt handwriting, cutesy artwork, emotional, warm, sentimental, and conservative. If it had been on NPR, you would have heard acoustic guitar and downtempo beats playing as the narrator talked. You can imagine the images on your friends' social media feeds (without attribution of course, and swiped from the NYT site). 

The artists made a number of points which are, from my perspective, rather flimsy.

They start off by noting the illustrations "were made by hand using paint and ink. Many hours of work went into them. And in the humble opinion of the people doing that work, it shows.[emphasis added]" 

One would hope the people who spent all that time would feel it showed, but they're hardly impartial judges. And "I spent a lot of time and worked real hard on this" has little or no bearing as to how most other people perceive the quality or value of work. 

Many of these artists blur the distinction between making what they want and commercial work (which is what someone else wants, but will pay you for). They are also raising the issue that illustrators and freelance artists don't get paid enough. There are very few workers I can think of who believe they are paid enough for the work they do. 

"AI will strip artists of their livelihood". This came right after an artist talking about "this never-ending loop of getting paid $500 for a visual that you spend three days killing yourself over". Uh, why would you want that job? Is that really a livelihood? It's a smidge over $20 per hour, so better than California minimum wage, but...this is what it was like before AI. It seems to me by-hand-illustration ain't a great way to make a living as currently framed. How much more can this human labor be devalued?

We also must clarify the difference between a capital-A "Artist" -- someone who creates what they want, for themselves, as a kind of gift to the world -- and a commercial artist, who is hired to create something specific for a set fee. 

Capital-A "Artists" are in no danger here. They aren't choosing art as a "livelihood." It is a vocation (and perhaps a curse). Artists create because they must, or because they simply want to. The money, if there is any, is a secondary concern. 

Commercial artists may be in trouble, but the truth of the industrial age is that if your job can be done by a machine, sooner or later it will. None of us is guaranteed a living doing exactly what we want to do for the amount of money we want.

"This will devalue art"

Prompt: "robot performer playing electric guitar singing"
Join the club, pal. Look at the music business. On the one hand, anyone and everyone can now make music inexpensively and distribute it globally for free (or close to it). On the other hand, everyone does now, and nobody cares. The music business is a faint echo of its former self financially and otherwise. Apparently it's all about social media now.

Or photography. When the internet started to go big, print magazines began to decline in relevance and then budget. Professional commercial photographers used to have big budgets and abundant time for shooting covers, spreads, etc. Technology chipped away at that. And when smartphone cameras began to proliferate, professional commercial photography took a bigger hit. Anyone could take a pretty good photo for cheap, and digital tools made basic clean-up faster and easier. More competition resulting in price drops. I know several professional photographers who walked away from careers because they could no longer do it the way they wanted -- no more time, no more money. 

This also hit the stock photography business in a big way, and the same thing happened. The internet was flooded with cheap stock photography sites. Maybe they aren't as objectively "good" as what was there before, but there are now oceans of stock photography, all dirt cheap. People who used to make a living doing high quality stock photography are having to find other ways to make a living. 

Before AI, people were just grabbing stuff off of Google Image Search, or Getty Images and not bothering to provide any attribution, much less pay for a license. These aren't just rinky-dink mom-and-pop shops. They're major video game companies. The Republican Party

This sort of cavalier and thoughtless use of artwork undermines the livelihoods of artists, of course, but more profoundly, the action belies an underlying belief: These people clearly don't value art in the first place. Digital images, photography as art, visual art itself have all already been devalued.

If people really cared about the "art" they're generating through these AI tools, they wouldn't be using the AI tools. They'd be paying an artist to do it and be heavily involved in the process. And they people "hiring" the AI would choose a human option if they thought it mattered to the audience purchasing and viewing the work. 

This cheapening has been going on for a long time. Commercial artists or professional graphic artists will tell you budgets have been falling for years, and that nobody cares about quality illustrations anymore. 

Most of the high-profile cases of paid work going to AI is because AI-generated art is a gimmick -- it is "new!". But most people don't want to play creative AI image roulette unless they had no intention of trying to find an artist, negotiate a rate, set a creative brief, and go through the process of creation

"Intellectual property issue because A.I. programs scrape human artists' work to Frankenstein them into a new creation." 

"Scraping other artist's work" and "Frankensteining" it into a new creation is also known as "having influences" and "learning from the masters." It is not "copying", and it is precisely the kind of fair use all artists benefit from and should continue to defend. I have seen plenty of work -- both fine art and commercial art -- that was clearly derived from, inspired by, and/or referencing the work of other people. Sketch artists do this sort of thing all the time. 

Any first-year art student can expound at length about how there is no such thing as "originality." If AI has looked at every image it possibly can and is able to draw on what it has seen to make something that is "like" something else, how is it any different than a commercial artist being asked to draw something similar to X or in the style of Y?

What AI is doing by analyzing big databases of images is no different from what artists do when painting "in the style of..." someone else. We could, should, and do allow that kind of derivative work.

"The job of the artist will change"

Prompt: "robot using typewriter by Gustave Klimt"
The implication here is the job will be different, and perhaps degraded. This is the nature of technology, whether applied to work, craft, or art. When the first crude drum machines came out, there was a lot of hand-wringing about how drummers would be out of work. Here's what happened: 

  • People realized drum machines were poor replacements for human drummers, and that if you wanted a human drummer, that's what you should use
  • There was a period where some people found jobs as "drum machine programmers", both because they knew how to work the machines and they knew how to make "good" drum beats
  • People realized drum machines could do things humans could not, and if you wanted a machine, that's what you should use -- machines felt different and were for different creative purposes
  • There are artists who manage to make machines sound remarkably human, but it is a lot of work using a skill set not everyone has
  • There are artists who manage to play live drums and sound remarkably like a machine, but it is a lot of work using a skill set not everyone has
  • The palette of sounds available to musicians radically expanded
  • More musicians had and have more options for drums and drum sounds
  • The cost of getting good or interesting drum sounds dropped to zero
  • New music and new music genres blossomed and flourished
  • The technology continues to develop and most records involve a hybrid of human work and machine work
  • Most people listening to music neither know nor care whether the beats were created by a human, a machine, or interaction between the two

People responded to the technology with fear of loss. Loss of human drumming as a job, of human "feel", of the value of humans playing drums, of creativity. But what actually happened is nearly the opposite of all of those things. Drum machines created new and different kinds of music, new and different jobs, helped people who otherwise would not have made music at all create something, and helped everyone -- musicians and listeners -- better understand the value of human drumming and "real" drums.

The job of the artist is always changing. Maybe part of your job is designing the seed image for the AI, part of your job is knowing how to work with the AI, and part of your job is sifting through a bunch of terrible images looking for the good ones. 

If nothing else, these new AI tools will inevitably produce artists who focus on AI interaction and output as their primary medium and make it the foundation of their career. Someone will eventually do some striking and creative things. Today, we are still in the early stages of “I asked this AI to generate something, look at how weird/bad/funny/adequate it is.” Tomorrow, “collaborations” between AI and human artists will be in modern art museums. 

Good Enough

Current AI art is like a well-read but uninspired artist. AI is good at cranking out consistent, mediocre, derivative work that occasionally surprises or delights when it brings you what it thought you asked for. It will never be brilliant. It will never "change your life." It will never make you think or cause a strong, polarizing reaction. 

Perhaps I am wrong. Art’s perceived value comes from the audience. There is plenty of mediocre work by human artists — movies, music, books, paintings — that matters to people. I imagine there will be people who find some AI works to be great, life-changing, powerful things, even if critics or the larger public consensus do not.

But AI art will absolutely be "good enough." Particularly for cheap commercial endeavors. If you are a commercial artist, you are best off leaning into this technology in a big way, and thinking about how you can leverage it to make your work better, faster, and easier.

Many of these points are similar or identical to things I have said about being a professional musician for many years. You can do what you want, but you can't decide how people will value that work. From an economics perspective, you're willing to get paid less (probably) because you are doing what you love (in theory). That's the choice you made. If you really cared about making money, you wouldn't be an artist. You'd be a banker or a lawyer or whatever. We can pick our jobs, but we usually don't get to also decide how much we get paid to do it, or how much society values it.

The real problem is that most people just do not care about art that much, whether it is painting or music or live theater or whatever. 

Artists and Tools

Prompt: "robot artist holding paintbrush"
We often use the word "artist" to mean two overlapping but distinctly different things. 

An "artist" is someone who makes creative work, driven by their own desire, "the muse", and/or some kind of inspiration. Artists make what they want and/or what they are compelled to create, as opposed to "entertainers" (who make what people want) or those doing "work for hire" (making things they are paid to make, but would not otherwise). In this context, the word can apply to anyone who "makes stuff": musicians, filmmakers, actors, poets, writers, and painters. We also use "artist" as a kind of catch-all for those who operate in visual media: painters, drawers, sculptors, illustrators, and so forth. 

I think it is important to remember the word has this dual meaning, as it can make things both blurrier and clearer.

I consider myself an artist: I am a former professional musician (I did it for the money) turned diligent amateur (I do it for the love). I create more work more frequently than many professionals, but I put little effort into "monetizing" my art, because I know the cold, hard truth: Most people just don't care that much about music anymore.

Computer-based tools and computer assistance are available for all artistic endeavors. If you are a creative person writing, painting, drawing, making music -- any of the traditional "fine arts", there are few places where machines aren't providing some kind of critical support. Even the work in that New York Times article was digitally photographed, cleaned up, and edited. 

AI has already won. It is here to stay, like steam hammers and drum machines. Like steam hammers, AI art generators will replace some human drudgery at the cost of the jobs of those humans. Like drum machines, AI art will help clarify the value and difference that humans making things with their hands provides and simultaneously open paths to new and different kinds of art. 

We can bemoan the loss of some kind of special human quality, or the value of art, or even just the cash being paid to artists. But the existence of AI art does not preclude any of that. You can have both. 

The reason we're afraid of AI art isn't the machine. We're afraid of AI art because we already know what we'll do with it, and what it says about what we really think about artists and creativity.