Brief Thoughts and Echoes on Arguments against Image AIs
It is interesting that looking back at the (very few) pieces that I have written, and the (much more numerous) thoughts I have internally debated on, they all, in some form, related to this topic, the topic of Image AIs, Artificial Intelligences mimicking the abilities and purposes of artists.
I am a computer scientist by profession, and a hobbyist artist by interest. How would I wish to have wonderous inventions situated within the intersections of my familiar fields, but as is the current state of these image generating AI models, I cannot support them at the slightest, because I do not consider these models ethical.
There is a lot to think about, between the application of these models, the underlying theoretical components, their societal impact, and my own philosophy on art, artifical intelligence, and the power of technological advances. I do not think I can adequately put them into words fully at this time, but to briefly summarize my opinions and stances to my best ability:
I believe that current image AI models are immoral, exploitative.
I refrain from using these models, and while I will not ask everyone else to do the same, I urge you to, and I ask you to please refrain from using anything from me, by me, or about me as input to these models.
I have a long-standing habit of not expressing constructed personal opinions on controversial contemporary issues. The reasons behind this boils down to selfishness: That I as an individual can’t bring about anything positive by complaining on the internet. However, as a researcher and professional on (a discipline tangent to) artificial intelligence, I am confident that this is a discussion and debate that I can in due time contribute to. Thus I would like to voice my opinion, even as unrefined as it currently is.
What inspired me to write all this and make such a statement is in part the increased exposure of these image AIs on twitter and across the internet, and most importantly a youtube video titled The End of Art: An Argument Against Image AIs, by Steven Zapata Art, recommended to me by an artist friend of mine. What follows is (and part of what above was) mostly shallow echoes of the lengthy but thoughtful arguments made within this video, with some personal opinions mixed in, so I would recommend watching this video from start to finish, if you are at all interested in art ethics philosophy.
The arguments presented by this video essay, summarized, interpreted by me, boils down to:
- The dataset these models are trained on are highly likely to be unethical, raising concerns regarding copyright, privacy, and fair use.
- Despite them being advertised as such, these models are unlikely to remain just mere tools for artists, and instead aim to act as their replacement.
- AI models, given the aforementioned potentially unethical assets, have an unfair advantage against human artists.
- To relegate art-making to machines, and eventually relinquishing the industry to them, is a suicidal act of berefting oneself and humanity of the joy and accomplishment found in it.
Ethical Concerns
On 1. The dataset that these image AI models train on, at least the ones that are published, are commonly datasets obtained from crawling and scouring the internet, including copyrighted and private content. While the nature of these datasets, on the surface, are non-profit and for research purposes, and are thus exempt from copyright legalities, their participation in for-profit image AI models are questionable at best. There are such datasets that are funded and operated by the very for-profit organization that train image AI models using them, and such act is unlikely to be attributed to anything other than ill faith.
On a technical level, this is my main gripe with the current status quo of image AIs. If there is an image AI model trained transparently and reputably on ethical data sources, using only copyright-free content, or acquiring permission from artists or properly compensating them for inclusion of their copyrighted work, then I would have no technical reason to oppose it on a macroscopic level. However, I am still very much against it individually, on a philosophical level, which I might discuss later.
Replacing instead of Assisting
On 2. While most image AI models are advertised as tools assisting artists, and while many people in good faith believes in that, I am pessimistic that they are intended to replace artists instead of helping them. This is evident by simply reading these companies’ statement with some critical thinking, and retrospecting over precedences of technology marketing themselves as tools.
The video essay elaborates this in great detail, and as someone completely untrained in art and only partake based on fleeting, unsystematic interest, I do not have much to add to this argument.
Unfair Advantage against Human Artists
On 3. Despite the analogy that can be made between a human artist browsing the internet for reference, and an image AI scouring the internet for training data, these two acts are fundamentally different. Humans are incapable of precise reproductions of artworks, and in the process of approximately doing so, be it via tracing or referencing, put effort in the such acts and learns experience of some form, which is in of it self applaudable. Quoting the video, ‘If you can perfectly trace a painting, you earned it.’ Machines lack such imperfections and can precisely digitally copy artworks, without putting much thought into it, if the process behind these models are even capable of being called ‘thought’.
This once again relates to my personal distrust and aversion of machine learning and artificial intelligence, as opposed to my interest, both philosophically and professionally, in classical (i.e. non-machine-learning) algorithms. You can probably check my previous blog posts (some of them not yet translated to English, apologies) for thoughts on that, so I will not repeat.
Relinquishing Enjoyment to Machines
On 4. There is a fundamental difference, again, between how machines replaced assembly line workers by automating things, and how machines intend to replace artists using these image AIs. Putting parts together, sewing cloth, cooking food, these can relatively be considered dumb labor, and most of those replaced from assembly lines will have no quarrel working another job paying similarly. Art-making, however, is an act that its workers - artists - do out of their passion and enjoyment, at least in part. Arguments can be made, and in most circumstances should be made, for automating dumb labors and benefiting both labor and production efficiency, but to within any reason support these image AI models in an act of berefting yourself, and ultimately the entire society, of the joy and accomplishment to create artistically as an artist and art-maker, is grim, foolish, and selfish.
What is different between previous arguments and this one is that there is a component within that does not depend on the ethical concerns of current models. The core philosophy underlying these arguments against image AIs is extremely humanistic, based on the principles that humans and humanity are prime, more important than supernatural entities and concepts, which for the sake of this discussion includes machines, tools, and technology. The core components of my philosophy are likewise deeply humanistic, that there is value in the creative unpredictablility and potential of human nature, and that there is in fact such things as the souls, of artists, of individuals, of humanity as a whole.
The video essay addresses such, admittably wishful, thoughts, as arguments for ignoring image AIs. This belief of mine does not stem from this discussion on this topic, but instead is born of a hope that there is indeed something that makes us humans, makes us unique. I construe this as an argument of mine actively against image AIs, as I do indeed value the humanistic component of a piece of art over other, more technical aspects.
As a storyteller myself, I am consistently in need of artworks and portrayals of characters and creatures I would like to present to my audience, some of them ideally very specific and not always available on the internet without commissioning. Despite this, I still find comfort in knowing that the artworks I use is drawn by a human being, with thoughts and passion poured into it, even if they do not meet the specifications of my requirements.
But after all, value is a subjective matter - much like my philosophy on humanity, soul, and true randomness - which really holds only if you believe in it. The truth is that many people don’t, or don’t care enough.
Oct 22nd, 2022. West Lafayette, Indiana.
