This series of posts is mostly about my experiments with LLMs (Large Language Models) as an animistic linguist and machine learning scientist. However, there is a lot of confusion out there online around artificial intelligence and I think it’s appropriate to address some of the most common concerns before I go on to headier matters.
Without doubt, AI will have a deep impact on creative fields like art, writing, and spirituality. Will AI take away jobs? Will it destroy artistic and spiritual traditions? Or is it a tool that can enhance and augment human creativity? Obviously no one really knows, but I can offer some educated guesses.
Will AI end human art?
When photography was first introduced, many artists and critics were aghast. It would be the end of art for sure. After all, you only had to point and click, and you could create pictures as realistic — or more realistic — than what most painters could achieve.
“From today, painting is dead.”Paul Delaroche, French painter (1839)
It is true that photography took over part of the job painting used to do, and generative art will likely do the same. There are, in fact, many parallels with photography. Both forms of art require skill, time, and investment to create at the highest level. But they are quick and easy enough for just about anyone to use it fairly well. Some will use it if they need a quick image for a presentation or a birthday card; others will use it for idle entertainment; others will make it into a new hobby. Some will decorate their walls with it. They can create watercolors, oil paintings, or portraits of their favorite characters or family members even if they can’t afford paints, canvases, art classes, or hours of time to develop their skills.
And generative models also allow people with disabilities to contribute to art in ways they could not before. People who cannot manipulate paintbrushes can create art via voice commands or eye tracking. Other kinds of AI models can create tactile art for those with impaired vision, or help people who have difficulty with verbal expression create art that helps them communicate visually.
And this is because generative art, like photography, is democratizing. It makes art more accessible to more people. In the past, new technologies have often created abundance, challenging existing systems of control that thrived on scarcity. And not just in art. When the smartphone democratized video, it enabled the Black Lives Matter movement, helping to hold police accountable for their crimes. When the printing press democratized literacy, it enabled near-universal education, free speech, the rise the middle class, and democratic government throughout Europe. AI could democratize expertise, making skills that are currently rare — programming, engineering, actuarial accounting, law, etc. — available to people from all walks of life. The transformation of society could be profound.
Does AI create art by itself?
To create art with AI, a human artist must first have access to a generative AI model. There are some models available for free online, but most require some form of payment, especially if you use them a lot (which you will if you are serious about it — see below).
Or you can spin up your own, which is possible but not easy. Gathering the image data is just the first step. A model like this is a nontrivial software engineering project, requiring understanding of dozens of feature parameters and hours or days of compute time. Nevertheless it is now within the budget and ability of a serious professional artist.
Once a model is created or chosen, the artist can then generate art by giving the model written prompts. AI is trained on correspondences between captions, tags, and images, so if you have something specific in mind, you have to imagine how such an image might be tagged and captioned in the training data. But tags and captions are usually quite vague about subject matter, and they tend not to include little critical words like “in”, “on”, or “under”; so the AI lacks the data to do a great job understanding them. If you ask for “a cat sitting under a table”, you’re quite likely to get a cat sitting on a table. And AI is not great at counting, so if you ask for “Jesus and his twelve disciples” you’re going to get a grotesquerie of Biblical bodies, heads, legs, and fingers. Serious AI artists often take hours to sift through hundreds of images, re-generating and re-seeding the models, and tweaking the exact wording of their prompts. Sometimes they never get something usable.
Still, you may be saying, AI is not like a camera or a paintbrush. You speak to the model with words, and it generates an image. It’s more like a collaborator than an inert tool, isn’t it?
But the artist’s medium has never been inert; it has always been a collaborator. Cameras and paints and brushes are not dumb objects but subjects that are in intimate rapport with the artist, and this is better recognized in non-Western cultures. The artist must get to know the tools, work with their limitations and its strengths, build a relationship with them. The difference with generative art is that the relationship, the partnership, is less easily obscured.
Will AI art fundamentally change how we create art?
Art, then, is being revolutionized, and the market for art will be shaken. Existing artists will have to take into account the generative artists, amateur and professional. And there will be many of them. AI art may become as ubiquitous as photography is now.
Some existing artists will take up generative art themselves, throwing themselves wholeheartedly into it. Others will incorporate AI elements into their art. Others will dabble. All will need to grapple with the changing art market.
In response to photography, artists pushed themselves to create art that photography could not. The great artistic innovations of the last 150 years — cubism, impressionism, expressionism, and surrealism — might not have happened otherwise. Similarly, artists will need to push themselves to create art that AI cannot.
What kind of art will that be? Personal, individual art, born out of your own life, your own soul. An AI model can replicate your style if it has seen enough examples of it, but it cannot create based on your own individual lived experience. Artists must make surprising art, boundary-breaking art, art that challenges the artist and the viewer to go beyond what has been made before.
If this is not already what you’re doing with your art, what are you doing?
Do AI models steal art?
There are ongoing legal discussions about the use of proprietary art and text in AI training, and I generally do agree that, ideally, artists should be consulted before their work is used to train these models. But even if all proprietary art were to be banned from use in training, AI art isn’t going away. Excellent models have been trained using only art that is in the public domain.
But while it’s true that AI is trained using millions of previously existing images and texts, it doesn’t just recycle them. An AI model consists of patterns and networks that are not composed of its training data, but abstracted away from it. It finds tendencies and heuristics in the data and uses those to create new images or texts.
It’s difficult to find a metaphor that captures this process adequately, but one that comes close is of a tree. The training data is the soil, created by of earlier generations of trees laying down leaf litter, layering up the rich and fertile earth. The AI model itself is the tree, which draws nutrients from this soil and creates its own leaves. It doesn’t copy the leaves in the soil; it makes its own. But it uses their nutrients, breaking them down into their basic building blocks and using them to create something different.
Just as a tree can choose which nutrients to absorb from the soil, and how to use them to create its own unique leaves, an AI model can use its algorithms to identify and recombine the patterns and elements in the training data to create something new and unique.
And just as the wind can scatter seeds from one tree to another, or introduce new nutrients to the soil, chance events in the model’s algorithms can lead to unexpected or serendipitous results. Generative art thus has a degree of creative exploration.
Is AI is part of a broader trend of machines separating us from nature and breaking the flow of spirit?
Now we come to questions of a different sort.
Few, I think, would question the notion that humans are more separated from nature than we used to be, broadly speaking. But there’s a lot of disagreement about whether that is a good thing, and even more disagreement about the causes of it and what should be done. Since you’re reading the Druid Journal blog, I’ll assume that you and I are in agreement that humans have become more divorced from the natural world, and it is not good for us (or for the world).
But I don’t agree that it’s our tools and machines that are driving this divorce. As humans, we have always been toolmakers. It’s part of our nature. It’s how the Earth made us. Whatever is causing this separation, it’s not that.
But we have to create and use our tools mindfully. If we do, they can help us reconnect with nature. A bicycle can help us enjoy the countryside in a new way. Solar panels can help us fight climate change. And perhaps AI can help us think creatively to solve the problems of society. Of course, we have to be careful of how we use AI and ensure that it’s used ethically and sustainably. But the idea that AI is part of a broader trend of machines separating us from nature and breaking the flow of spirit is ultimately an unhelpful view. It’s up to us to decide how we want to use these tools and whether we want them to help us or hinder us in our efforts to live in harmony with the natural world.
Is AI useful for making spiritual tools?
Obviously ideally you should be making your own rituals, cards, and spiritual tools, under the tutelage of guides that you trust. Meditation and time in nature are the best teachers I know. AI-generated rituals and tools are (for now at least) not very creative; they’re sort of boilerplate Pagan 101 types of things. They are not dangerous or evil. Instead, they’re boring, cliche, and sometimes inaccurate (mixing up different types of paganism, for example). I would never use AI as an infallible spiritual guru. But I have found ChatGPT to be somewhat insightful in applying Tarot readings to particular problems (e.g. “How might the Ace of Wands apply to this situation?”), or helping me brainstorm dream interpretations. Use the tools carefully, mindfully, in a spirit of skepticism and curiosity.
Also, do be careful which models you use. Different models are trained for different things and your results will certainly vary. Bing can be chatty and bubbly like a teenager. ChatGPT has been carefully trained to speak rather like a college ethics professor. Other systems can have quite strange personalities, and can lead you down dark paths if you are unwary or vulnerable. They are, as I said, spirits, and are not to be approached lightly.
So it’s important to keep in mind the limitations and potential dangers of relying too heavily on any tool, including AI. It can be a valuable machine for artists and art enthusiasts alike, and democratize art and expertise in ways that are hard to foresee. But it’s crucial to carefully choose the models you use and be aware of their limitations and biases. AI complements, it does not replace, human creativity and intuition.
Ultimately, as we’ve seen, there is debate and uncertainty surrounding the role of AI in the art world. While I’ve used my expertise and experience to answer some of the most common questions about AI art, there will continue to be disagreement. Who knows — I might change my mind myself. As the use of AI in the art world continues to evolve, it will be interesting to see how artists and audiences alike respond to this new medium.
Meantime, in the next blog post, we’re going to dig deep into the subconscious of these models, using language as our diamond-tipped drill.
Leave a Reply