June 6, 2022
No, AI Can’t Produce Art (but It Can Write)
Based on discussions and what I read, a lot of people are afraid AI, artificial intelligence, will “steal” their jobs – what the hell happened to the “one day we’ll have robots and we won’t need to work” dream of early sci-fi? But I digress… Writers don’t seem to be an exception. A great number of them seem to be worried about AI replacing them. So, can AI produce art?
In this post I argue that no, AI can’t produce art – for reasons we’ll explore. But AI can certainly write. It can already now, and I’m certain it will become even better.
Where’s the difference, then, you might ask.
The difference, to a large extent, revolves around matters central to the writer-or-artist distinction. More insidiously, perhaps, it’s about conditioning us into patterns that can have far-reaching (and unpleasant) consequences.
What Is a Writing AI?
Or – do I dare? – who is an AI writer?
Aspects of consciousness don’t really enter the “AI can’t produce art” discussion at present, but they might at some point in the – probably near – future. In other words, there will soon come a time when we’ll be asking ourselves whether AI “counts” as consciousness.
It thinks therefore it is?
In any case, this is far beyond the scope of this post. Still, I’d point out that it can be part of the statement including AI producing art: AI would be conscious if and only if AI could produce art. But I’ll let you ponder on that.
Instead, let’s focus on definitions: A writing AI is a program (computer/technology/whatever non-human) that can create original texts, well enough not to be immediately obvious that is made by something non-human.
Some Problematic Aspects
Already, there are several problems with this definition. Does a silly program like my poem shuffler count as that? I’d say no, because though the order of the lines is original (and random), the base text is not. Similar programs, like the random text generator – or even the more sophisticated Word Journey – still don’t fulfill the requirement of originality.
An AI would need to go far beyond creating coherent (or coherent-like) sentences. It would of course need to be able to make sentences like “Colorless tables are on sale today”, it would also need to “understand” (whatever that means in the context) the semantics and pragmatics behind “Colorless green ideas sleep furiously” (and why it can’t be used), but it would also need to create a story – a narrative progressing from point A to point C passing through point B – and make it appealing.
That’s a huge task.
It also reveals several other problems:
- appealing by whose standards?
- how can it know what is appealing?
- how do we define “original texts”?
- who decides (and by what standards) if it’s made by something non-human?
It’s a Turing test on steroids!
AI Can’t Produce Art, but It Can Write Really, Really Well
Before we go to the crux of the matter – why, in my opinion, AI can’t and never will produce art – let’s fully acknowledge something: AI can write really, really well – even now. It’s not science fiction, it’s not a concept; it’s reality.
I totally envision a very near future in which AI will be able to write novels in the blink of an eye, and people won’t be able to tell that it wasn’t written by a human.
So, if AI can write really well, why can’t AI produce art?
AI Can’t Produce Art (then Again, Can We?)
The reason why AI can write spectacularly well is simple. It’s also the same reason it addresses the issues I mentioned earlier, regarding being appealing, original, or having points of reference.
You see, AI is a master emulator. It’s made (and it then teaches itself) to emulate like there’s no tomorrow. A writing AI is the very embodiment of cargo cult writing. It’s programmed to absorb an impossible amount of text, perhaps from incredibly diverse sources, and then mindlessly (quite literally) Frankenstein it into coherent, original-looking text.
And it works fantastically well. You wanna know why?
Because we do it too!
We emulate each other constantly. Indeed, most of the novels (films, songs, etc.) surrounding us are emulations to begin with. And the more constrained they are – likely as a result of genre or marketability – the likelier they are to be both mediocre as art and a result of emulation. Indeed, a writing AI would do a better job at writing, because it would be better at emulating – as a result of its much higher ability to access source texts and shuffle things.
AI can’t produce art, and neither can we – 99% of us, 99% of the time.
But it’s that 1% that makes all the difference.
From Emulation to the “A-ha!” Moment
99% of a novel is basically an emulation of preexisting patterns. It’s little more than an amalgam of the author’s own thoughts as well as others’. It’s the collective sum – personal as well as sociocultural – that has led the author to that point, in that moment.
But there’s a 1% (and if we’re pessimistic, I could add: 1% of the novel of 1% of the writers) that is genuinely original – in ways that aren’t easily expressible, precisely because of their inherent nature.
Shakespeare coined almost 2000 words and plenty of phrases – such as “it’s Greek to me” or “too much of a good thing”. Arguably, perhaps he simply recorded a pattern he’d seen or heard elsewhere, but we shouldn’t underplay his contributions. The key element here is breaking the rules in order for something to fit an artistic framework.
As a writer, do you remember when you came up with a word that just “felt right” even if it didn’t exist? That’s a tiny morsel of art, but it’s art. It’s creating something out of nothing.
An AI can’t do that, because – even if it has self-learning capabilities – it doesn’t have a recourse to the disconnected (non-sense-making?) thinking patterns humans do. An AI, quite literally, is programmed to follow rules. Even the act of breaking a rule would have to be programmed.
Could we have an AI that learned on its own to break the rules? I guess that would take us back to the consciousness debate. If (and only if) an AI could do thatThere is a very tricky aspect in all this: Exactly how could we tell if an AI really was conscious, when we cannot determine that for humans?, then yes, an AI could create art.
AI Can’t Produce Art – and that’s Very Dangerous
The entire discussion is problematic, I hope this much is evident from the post. There are simply too many things assumed, too many aspects that aren’t easy to define – if they are definable at all. There’s certainly a degree of anthropocentrism in this approach, and a sense of chasing our own tail.
Then again, can art not be anthropocentric?
I’m not sure about that – I’m not even sure art requires an audience. I’m certainly not able to properly define what is the threshold separating an original text from one that isn’t, or all that.
There is, however one thing I’m certain of: Being surrounded by AI that can’t produce art is very dangerous. The reason isn’t in the AI itself (despite its being considered akin to witchcraft by many people, an AI is a very logical, man-made thing that does what it’s told).
And that’s the problem!
The Warning from Google Photos
If you use Google Photos, you might have noticed how Google suggests “Fix Lighting” for several of your photos. The thing is, it can’t tell apart a photo that is underexposed by accident from one that is underexposed deliberately, for artistic reasons.
I know about it because I’m an experienced photographer and I know what I had in mind when I took the photo. I also have enough experience to guess if a photo by another photographer might have been underexposed deliberately or not.
But people not as experienced with photography do not. And so, they are conditioned by Google to have a certain level of reference in regard to what constitutes “a good photo”. Do that enough many times, and you’ve just created a new paradigm (of technical normalcy, conflated with artistic value) where photos “are not supposed to be” underexposed.
Writing on the (Facebook) Wall
Similarly in texts, we have been conditioned for years to think in certain patterns. “Use hashtags”, “be short and sharp”, and my personal favorite (not), “Avoid passive voice” – oh, yeah? Says who? Developers of grammar-checking plugins, enamored with the idea such things can be quantified?
I mean, how do you even begin to express the following in anything other than passive voice?
We might characterise these peculiar hybrids of hyperbole as a series of immense Gothic bridges thrown between the universe of discourse and the forces of history
Punter, David. The Romantic Unconscious: A Study in Narcissism and Patriarchy. New
York: New York University Press, 1990. p27
How should we satisfy passive voice demonizers? “…of immense Gothic bridges that we threw between…”?
The truth is, we have conditioned our writing for likes, comments, and – above all – to influence “the algorithm” and get more visibility.
An AI that can write “really well” (=it can emulate humans) is only likely to make matters worse, precisely because it will be much harder to realize the slow, insidious slip into mediocrity. AI is a fantastic tool, as long as you know how to use it.
Things are already bad when it comes to having access to high-quality art. We don’t even need AI, we do a great job ourselves creating an ocean of mediocrity around us, production-lining novels with little to no meaning. An AI will make it even worse because it will build on such patterns (because that’s all it can do), and thus further condition us to conflating art with industrialized “art”.
5 Comments
Comments are closed for posts older than 90 days
I think they should concentrate the AI on things humans do very badly (such as medical expert systems – doctors have a very poor track record bringing all the symptoms together and figuring out the obvious, only partly because they get so little time to do it with each patient), and which AI might do better simply because of access to such a huge database.
Weather and sports reports from an AI are already indistinguishable from the, as you so correctly note, mediocre reporting of the humans trying to make a baseball score seem original.
And algorithms produce some fascinating visual ‘art’ – by making calculations of elaborate formulas like the Mandelbrot set quickly.
But ‘Art’? Even most humans don’t produce Art.
Medical records is indeed an excellent application, precisely because humans (doctors in particular?) seem to be too scatterbrained – or outright not caring – about details and history that could be critical. Been there, pretty frustrating.
I’d definitely vote for AI handling medical records and at least suggesting diagnostics. My late father died in hospital after six weeks during which time I missed just one day of visiting. On a daily basis, I witnessed nurses and doctors – always someone new – skimming the first page of the medical chart and basically continuing the status quo. It took over a week of nagging on my part to get them to xray Dad’s hip. Yup. Broken. Dah… -cough- Frankly, I don’t think even present day level AI could have done worse.
Getting to the meat of your post though, I believe that 1% applies across the board, not just to writers. Painters and composers etc have to make a living too, so the ‘successful’ ones are often as derivative as the bulk of writers. In fact, it’s gotten to the point where I never buy Indie books that have thousands of reviews. Not because I believe the reviews aren’t real, but because experience has shown that what I consider to be ‘good’ is generally not what the average reader considers to be good. So I tend to search out the writers with more modest popularity. I’m very rarely disappointed because those writers are usually the ones who push the boundaries, who try for something new, who refuse to create cookie cutter stories.
So…could AI create anything resembling that 1%? Nope. No matter how sophisticated AI/neural networks become, they will always lack the chemical soup that gives humans their ability to ‘feel’. All of our processing is a combination of electrical impulses and chemicals that do their alchemy at the level of the synapse. Short of creating AI with wetware of some sort, and simulating the functioning of billions of synapses all working at once, there’s no way an AI can rival us when it comes to creativity.
I once read that the human brain is the equivalent of 17 billions computers. I can’t remember how the author of that article arrived at that precise number, but I laughed at the thought of needing a planet spanning computer just to approximate one human being. 🙂
Sadly what you say about the future of creativity is probably true. Just as most people are happy eating a hamburger on a synthetic bun rather than French cuisine, so most people – assuming they read at all – are happy with pulp fiction. That target audience won’t care who or what produces their hamburger so long as it’s quick and cheap and gives them the flavour they’ve become conditioned to want. 🙁
Many thanks for your elaborate, thought-provoking comment!
-giggles- sorry! It was rather long. I have a thing about AI. 😉