Should artificial intelligences win art competitions?

Vice reports that the Colorado State Fair’s fine art competition has been won by a person called Jason Allen, who submitted this AI-generated piece of art.

I do quite like the picture, but, as you might expect, it’s created a bit of uproar amongst some artists of a more traditional nature. One of the more viral tweets was:

TL;DR — Someone entered an art competition with an AI-generated piece and won the first prize.

Yeah that's pretty fucking shitty. pic.twitter.com/vjn1IdJcsL

— Genel Jumalon ✈️ Nan Desu Kan (@GenelJumalon) August 30, 2022

Whilst the results of a single art competition may not be of world-changing significance, to me it provides a fascinating exercise in “what do things even mean?”, with a bit of “what do we actually want for the future?” mixed in. It’s not impossible to imagine a day where AIs simply become better at winning art competitions in general than humans.

Note here a distinction between “winning art competitions” and “producing valuable art”. The former is easy to measure and hence optimise for.

Nonetheless, we’re not there yet, at least not in an unsupervised fashion. An argument in favour of the human author of the AI piece deserving credit is that he did have to go to some effort to create it.

For a start, he apparently developed and iterated a load of prompts for the AI to work from. These kind of systems usually work on the basis of you typing in some kind of text description to work from, for example “darth vader mowing the lawn“

Perhaps there’s an art to doing that. Certainly you can see a lot of much less pleasing looking AI generated art on Twitter from people typing probably less well thought out generation phrases in! Certainly some people seem to think there’s intrinsic value in good prompts – there’s already a marketplace for buying and selling the best of these phrases called PromptBase, with an associated occupation of “prompt engineer”.

Allen then curated the best of what the AI produced which presumably requires a certain level of artistic understanding. And then did whatever was necessary to transform them into a physical picture, including doing a bit of processing in Photoshop. I’m certainly not saying this is the same as someone, for instance, painting a picture in the conventional style, but it’s not nothing.

Perhaps there’s special power at present in the combination of human and machine, in a similar way to the game of chess, where the combined effort of a human and a computer is often thought to be particularly effective, leading to the introduction of “cyborg chess” competitions.

In any case, is an art competition really about rewarding effort? We don’t typically base art competitions on how much human effort went into the painting (as far as I know – disclaimer: it’s not a world I’m involved in!). We often can’t even know the level of effort. But when we do, I doubt anyone disagrees that an artist that tries desperately hard but just isn’t very good is a thing that can exist. Likewise, seemingly naturally talented artists are not looked down upon because good art is “easier” for them to produce, quite the opposite.

So if an “artist” finds a less labour intensive way to produce art, is that really a problem?

In some of today’s competitions in other domains it surely would be seen as such. The Olympics do not allow participants to take certain drugs that would improve their performance in a way that makes winning the competition easier. I assume they’d also not allow athletes to cybernetically enhance themselves with AI-enabled robotic limbs before running the 100m sprint. But I doubt there’s any thoughts of banning the use of AI to support the athlete’s training regime.

An obvious quick solution to this specific problem would be that competitions just have a separate “AI generated” section. After all, chess now has specific “cyborg chess” competitions where humans and AIs team up to seek victory, whilst explicitly maintaining the default that you cannot bring a computer to the game in a conventional tournament. But even there I can imagine complicated grey areas.

What if Allen had generated the above image via AI, then picked up an IRL paintbrush and created a painting of what the AI had produced? Hence actualizing a picture that’s based on whatever the AI “imagined”, for want of a better word. I suppose conceptually this might be akin to an artist redrawing another artist’s painting. How would we feel about an artist who repainted a copy of the Mona Lisa winning a competition?

Or is the human who generated the picture less relevant? Does the AI deserve some or most of the credit? To be fair Allen did make it clear he’d used the AI in question, which is called Midjourney, although it’s not obvious that the audience would have understood what exactly that meant.

As noted above Allen had to provide a prompt to the AI to generate the image. How do we feel about someone who texted their friend a request to paint a picture winning a competition with something that was created by the afore-mentioned friend, albeit somewhat constrained and selected by the winner?

Although most of us don’t attribute much in the way of personhood to AIs at this point. So if the AI does deserve credit, then does that mean the creator(s) of the AI are due some? What if Allen had produced his work by first creating the AI himself and then using it to generate an image? Which is not what happened, but I’m sure there’s people out there who could make a stab at that kind of process.

How deep should we go? Most AIs are built by processing vast datasets, many of which are effectively generated from the mass output of humans. Is there a way in which you and I, and the other billions of humans potentially feeding the machine, deserve some kudos if the AI’s output was potentially affected by our toil?

In an edition of the Galaxy Brain newsletter that discusses a different art-generating AI called DALL-E, a user of that system called Andy Baio, is quoted as saying:

DALL-E is trained on the creative work of countless artists, and so there’s a legitimate argument to be made that it is essentially laundering human creativity in some way for commercial product

As a potential counterargument to going too far down that vortex, perhaps Midjourney is just a particularly fancy tool. We don’t usually give credit to the maker of the paint or canvas an artwork is created via, no matter how much ingenuity and flair went into its creation.

The competition category that Allen’s picture won was called “Digital Arts / Digitally-Manipulated Photography” after all. Presumably the other entrants also relied on some form of computer manipulation to create their pieces.

I’m sure no-one would quibble about the usage of the (once?) ubiquitous Photoshop digital editing software. But these days Photoshop proudly proclaims its use of AI – “Now the world’s most advanced AI application for creatives” was one of their marketing messages a couple of years go, promising to let artists get from the picture on the left to the picture on the right in “just a few clicks”.

Now this is certainly not the same as you typing in “picture of a sheep” and it producing a farmyard scene from nowhere. But it is using AI to allow some people to produce art they would not have had the ability to do in the past.

Even remaining in the domain of art, this issue goes way beyond art competition glory. It’s also important remember that being an artist is actually a paid job for many people. And probably one with particularly high barriers to entry in the first place, let alone to get into a position of a well-paid and reliable income. We all know the trope of the starving artist.

The original Vice article also mentioned the controversy that was generated when the author of the Galaxy Brain newsletter I mentioned above Charlie Warzel, chose to illustrate one edition with output he generated from the same Midjourney AI.

This generated a storm of upset and critique that a big commercial publication like The Atlantic was cheaping out and using self-generated AI art rather than paying an illustration. Warzel wrote a followup edition of the newsletter reflecting on the mass, often vitriol and hate-filled, feedback he got. In reality apparently the way his work is structured he would never have had the opportunity to hire an illustrator in the first place, so in this case it was more a substitute for using stock photography. Nonetheless he concluded that he probably wouldn’t do the same thing again.

i promoted this myself. the avenue i have to illustrate my newsletter is a getty account and i was momentarily tired of using the same image of Jones. i was playing around w/ midjourney as i’ve seen other newsletter writers do & decided to try this. i don’t think i will again.

— Charlie Warzel (@cwarzel) August 10, 2022

That said, whilst there are frequent, often doom-laden, articles about technological automation taking over jobs that people are currently paid for, it’s rare that we see a successful effort to actually stop that happening. Sometimes it’s actually seen as a good thing if the jobs are considered to be mundane, hazardous, or – if feeling cynical – in categories that could be uncharitably summarised as “the sort of thing tech elites think that only working class people do”.

Businesses under capitalism have of course constantly embraced whatever technology is available to reduce the human effort needed to perform various tasks as soon as it reaches a certain degree of affordability. That’s simply because it usually ends up being much more profitable for the business owners, or at least seen as such. This can of course be good if it makes people’s lives better. But in reality, the resulting human cost is usually shunted off to wider society to deal with, or in less progressive systems borne by the affected individuals themselves.