It is easier than ever before to produce visual media, but what does that mean?
Since my previous piece two years ago, artificially generated material from LLMs (Large Language Models – the core of all modern generative AI) has only become more visible. To the point that it’s difficult to avoid an advert shilling for the newest ways to turn armies of expensive graphics cards into money. In that light I believe it’s time to take a second look at the relationship between AI and creativity.
As an artist, or someone who requires art, what do these services mean for you? It is undeniable that these services can provide a quick fix if you need ‘content‘ for a project you are working on; Almost like magic you can have something polished within mere seconds. If you need to impress and have few resources to hand then it can be incredibly tempting to reach for an AI tool.
However I would argue that in most cases these generative outputs are likely to hinder your project, especially if not used very carefully. The presence of this material will, ironically, only make your project more forgettable off putting and bland. You may also find in many cases that automating the creative process runs counter to the notion of creating in the first place. Finally, the hidden costs of AI might be prohibitive to you – depending on your priorities.
For the sake of focus I’m going to lead with the visual arts, but these arguments are still generally applicable to the other mediums.
Can LLMs make art?
Quoting a previous version of myself:
As with most areas of philosophy the definition of creativity is hotly contested. Though there is an emerging consensus that for something to be creative it needs to satisfy two conditions. Firstly it must be ‘novel’ – meaning a new combination of elements, original in some form. Secondly it must be ‘valuable’, though value in this context could be replaced with ‘exemplary’ or ‘notable’. It must have something to make that originality of wider interest – though this has been contested. Whilst there have been studies on AI and creativity, the limiting factor in current research is in-domain knowledge (to allow researchers to assess creativity within a medium), and valuation of results (the ability to judge the creative worth of the output). The complexity of the task of evaluation cannot be stressed enough – and alongside definitions of creativity, these tend to rely on factors like emotional response, motivations, lived experiences, shifting tastes, values and more. This is a running theme for researchers of creativity, alongside aspects of self expression, expression of of ideas, and a blending of them between the conscious and unconscious mind.
What I’m getting at is that theory of creativity tends to be connected to theory of mind and theory of consciousness. I propose that creativity has a direct causal relationship with consciousness. It is my personal belief that to be able to create in the way we understand true creation, one has to have a conscious mind.
This is why if you should choose to use AI services to create your own work it won’t be bringing any new ideas to the party. The lack of ability to generate ideas is why the only ‘value’ AI can bring to a work is brought by you and the context you bring to it.
Having said that, I do believe it is possible to derive value from AI generated work. As LLMs are trained on data siphoned from a biased society, the results are indelibly coloured with the strata that already exist in society. LLMs are a great resource for highlighting bigotry and various assumed truths – especially as western culture has a tendency to dominate the online sphere and is disproportionately represented in AI output. In this sense it can be used as a mirror, to look at ourselves. Having said that it is incredibly unlikely an AI image on its own will have enough artistic weight to be considered to have ‘true’ value.
Originality, and lack thereof
Something you’ve likely already noticed is that it is exceedingly easy to make painfully mediocre material using LLM’s, whether that’s midjourney, DALL-E, Stable Diffusion, etc. Facebook (for those who still use it) has become a ghost town dominated by boomer-bait ‘fake’ memes pumped out by pages dedicated to disseminating artificial images, and other social media sites aren’t far behind. You’ll notice that this AI slop tends to have a distinct look. To pick a random example:
Quite often the output has a hyper-glossy lighting treatment and has a tendency to look like an advert. The easiest way to understand why is to think of AI art as the result of taking ridiculously huge quantities of text and images, chucking them in a blender, and reconstituting that into output like the above. Fundamentally, an AI algorithm is unable to create or innovate – it is a very well designed mimic. It takes a request and pieces together a vast ocean of visual works to create a kind of hyper detailed exquisite corpse – a mish mash of concepts squashed together to make an impressive approximation of ‘content’.
Many argue that there is no difference between an AI algorithm using references to create art and a person using references to make art – but in my opinion this is a dishonest comparison. A human takes inference from reference material to create work that exists in dialogue with the reference. Through context and concept the reference material is used as a point of inspiration to make an entirely new work. In most cases, all of the actual mark making of the final piece is derived from the artist’s own hand. In contrast, the algorithm simply recycles the data of the original work into a new configuration – there is no interpretation because an LLM only has relational connections made during training, and no ability to create anything but a predictive guess at what the desired outcome of a text prompt should be.
This lack of context, awareness, ability to create a dialogue, imagination in short, is what makes AI content simultaneously so arresting and so forgettable. It recycles what has been popular before to predict what would be popular now.
Notice how Suhail focusses on details but there is no context or meaning here. Sure, there is an obligation for some art to be ‘beautiful’. But for art to truly speak to someone it usually needs to be imbued with some kind of meaning. And I think this is where AI highlights something important.
The Medium is the Message
The advent of AI art is not the first time a new medium has impacted the creative community. The advent of the printing press re-contextualised calligraphy – no longer pure communications technology but an expressive medium. The camera re-contextualised painting; taking it from a medium for stuffy portraiture to, once again, a tool for self expression. De-prioritising realism. How does AI art re-contextualise art? Or any other application it is turned to?
In a similar way I think it re-contextualises the very act of creation in the first place. It refocuses the creator’s act from ‘how’ to ‘why’. If you simply require something as opposed to nothing, and the actual material itself doesn’t matter then sure, it may fit your purposes. It simply allows you to create more – as long as the ‘more’ looks an awful lot like something you’ve already seen. It will always trend toward the average – the most common avatar of the concept you’ve requested, the archetype. It is an engine for derivative, ‘good enough’ work. This means that, in the creative sphere, the primary distinguishing factor – especially as the quality of AI outputs improve, will be intent and imagination. Why are you making this? What are you trying to achieve or convey? What are you trying to say? Is detail required to say it?
Heyheymomo is a creator of high quality eyebleach – aggressively wholesome webcomics. They are simple both in content and execution. They effectively convey incredibly human situations. In short they are infused with a sense of humanity that can only be created from a perspective of understanding humanity. It communicates on an instinctive level. In Spanish term word “Duende” is used to denote art that has life – a kind of soul imbued to it. Art with duende provokes an emotional response in the audience. Art that is alive. By comparison generative ‘art’ is is inherently dead, zombified even.
So can I use AI to help me make art?
Until an algorithm understands the qualia (i.e. the innate nature) of existence, it will not be able to impart true imagination to work – and in that sense never truly be able to ‘create’ anything. If the aim of what you are making is to say something new or specific to you, using AI will always be a guiding force steering you toward mediocrity. Of course, you can always use AI material as a starting off point, but still know that that starting point will always come from an average, rather than extraordinary place.
This might be acceptable for your purposes, though it will be difficult to use the master’s tools to demolish the master’s house. And considering what LLMs make is essentially a vertical slice of peak neoliberal culture, any direction you’re likely to want to go in is probably away from the centre. Unless you like beige, of course. I recently had to condense a piece of writing down from 5k+ to under 750 words and out of curiosity passed it through a chat bot. So much of the essential meaning of the piece was lost, and so individualised did it portray the problem, that in the end I rejected it entirely in favour of redrafting the piece myself. The scanner sees darkly, and if you choose to take on this vision, so too will it alter your perception.
So yes, if it’s a useful tool for you, then feel free to use it. But remember a real reference will always be more useful than something an LLM sticky taped together – and will give you greater cultural context because you will know where it came from.
The Cost of Entry
Aside from reasons around art theory and nebulous notions of creativity, there are other reasons you may decide against using AI to automate your creative process. Or parts thereof. The most widely covered aspect being the debate around intellectual property and theft. You may have certain feelings around whether ‘Intellectual Property’ should even exist in the first place – in a perfect world I certainly wouldn’t argue in its favor. However in a world of IP hoarding mega monopolies and asymmetric enforcement of those rights, copyright law may be the only line of defence against the wholesale stealing of innumerable artists’ work. As mentioned before – these services analyse petabytes of data, largely taken wholesale from the internet without obtaining consent from their creators. In fact, because of the way AI works, any output it generates is ultimately a kind of plagiarism.
Aside from this, the environmental cost of generative AI is staggering. Creating a new version of a model requires the equivalent power of a small nation to train – around 11twh and expected to rise significantly before 2026. On top of which, the power required to run these data centres adds even more power requirement, and many of the datacentres running the hardware use evaporative cooling – placing huge demands for water on the areas surrounding them. Some estimates suggest as little as 5 queries processed by GPT-3 can consume half a litre of water, and even suggest its’ successor (GPT-4) can consume the same amount for generating as few as 100 words of response
There’s also the human cost of the AI ghost workers who are employed en masse ensure the proper functioning of these service in what are poorly compensated effectively digital sweatshops. AI output can be unreliable and reproduce darker elements in the training data. People are employed en masse to act as a human crumple zone to ensure the output is sanitised – leaving these precariously employed people responsible if something goes wrong.
What Now?
I hope I’ve managed to convey the complex relation the AI economy has with the arts. If you still so choose to employ the outputs of these systems then hopefully you do so with full awareness of context.
I would like to leave you with one final thought though.
We live in an attention economy. Part of the reason the lure of AI is so strong is because it helps wrung out creatives produce more in a shorter time span – which is a demand made by our social media driven culture. If you don’t create often and in a certain way you lose visibility because you aren’t creating what the algorithm is looking for. As suggested in an article by Jason Koebler, AI is being used to brute force social media algorithms. To chase trends and spike visibility with whatever is popular at this exact point in time. And that does have power – it can be, and is, exploited.
There is a careful balance to be struck between achieving a goal and hitting a target. In the modern era the target increasingly seems to simply be ‘more’. But creating without intention, without taking pause to reflect on who you are, what you’re doing, and why, it can become empty. Especially if you stop enjoying it. Take it from someone who’s been there (and still is, to some extent). Don’t turn your passion into a job. Don’t kill your interest for the sake of the grind. Don’t destroy yourself in pursuit of an unattainable goal. And above all think. Make your creations out of passion and enjoyment. Create because if you don’t you’ll go mad. Create from the heart – however you do that, employing AI tools or not.
But beware.
He who fights with monsters might take care lest he thereby become a monster. And if you gaze for long into an abyss, the abyss gazes also into you.
Don’t lose yourself along the way.
My previous piece on AI can be found on my website https://gaffen.co.uk at https://gaffen.co.uk/posts/ai-ip-theft-and-the-death-of-creativity/. It delves further into ideas of creativity and machine consciousness and has a heavier focus on philosophy.