AI and art

I’ve been thinking recently about whether an AI would be capable of creating art. I’ve actually written an essay on the subject. It’s difficult from the word go because of the general subjectivity of art and no clear consensus on as much as a definition. The essay is something I’ve updated a couple of times because I felt it incomplete. I may feel this way about it for some time and I thought getting the opinions of the artists in this community could help me track down possible issues that I hadn’t thought of. So what do you think? Is it doable? Would we be able to come to a conclusion even if it were?

Here’s the text body, minus the abstract. What do you think?

Composition:

Composition has been reduced to rules in an attempt to make it teachable. Though there’s not an ironclad set of instructions for creating art, use of the principles can bring about fairly strong examples. Graphic pieces might be best centered around balance. This would include use of canvas real estate, symmetry and size of forms, use of warm and cool colors and light and dark shade variants. Graphics maybe the most practical genre for AI in that they tend to be more logically and methodically composed. With painting a scene there are also many tools that could be employed to create the subject/s and add interest. Even natural subject matter can be generated with the use of fractal math for instance rough mountain slopes and tree branches. Placing them then with consideration of the “Rule of Thirds” and off center focal point as though they are being raytraced from a camera in a 3D workspace could be a rout for implementing composition 101. The rules for color theory and shading were derived from physical science and have already been implemented in shaders. I think composition would be one of the more strait forward aspects of art for the AI.

Interest:

In making a piece interesting, an interesting subject or theme is required. Since the basis of this is essentially subjective, it’s unlikely to be empirically demonstrable and probably subject to criticism. There are however rules for this as well. In painting a scene one could generalize it with forming a sentence. It could have a subject and predicate that could be embellished with adjectives and descriptive adverbs. One might be able to create a piece that tells a short story in this manor. Like with mathematics the components can be derived from natural language. This of course is a field that is well on its’ way. One issue might be with the pattern recognition needed to create components that indicate movement or some form of animation for predicate visuals.

Expression:

Expression could be fairly effectively simulated through rule sets. Negative emotion could be expressed through breaking some of the composition rules for example, using muddy colors and an overarching lack of contrast could depict an air of dolor. Positive emotion could be conveyed with pastel type colors, dramatic lighting schemes and even lots of rounded shapes. It would be safe to assume that this is where the bulk of the controversy will lie. There will likely always be some debate on whether AI could create an expressive piece. I personally think that our view of our emotional responses being complex beyond pragmatic reduction and implementation just has not been shown in the art world. I would even suggest that art has helped us to better understand our emotional states and how environmental factors can effect them. This is a tool that could be used to invoke an emotional response.This is also something that I think could be researched effectively and empirically.

Creativity:

As artists we like to think of ourselves as creators but to the contrary even our more creative works are composed of pre-existing components arranged in a novel way. Even the works of the more creative artists like Picasso and Salvador Dali follow along that line. An assertion of the notion that an AI could not be as creative as a human does not appear to be justifiable because of this. It would seem that creativity could be implemented with simple, rule breaking rules.This is a likely topic for debate among the art community however a difficult one to push as an issue.

General Subjectivity:

Debate on this topic is likely to persist indefinitely. The issues with finding a consensus are many. The art community is an anarchy in its purest form. Subjectivity is not only evident but also encouraged. The question as to whether an AI could create art is made essentially unanswerable by the fact that we can’t even agree on a definition of art. Artists do however tend to assert that it’s in no way important. It’s my intuition that there will be a large percentage of artists that would prefer that the answer to the question be subjective. I think it would be safe to say that many would expect that an AI would compensate for not being biological through technological strengths. The Atlas vs Dionysus debate is likely to play a role in forming opinions. This is an ongoing debate on the importance of technical (Atlas) and emotionally expressive (Dionysus) skill sets. I would expect that some would assume that an AI would be likely to be the pure Atlas in that the technical capabilities are thought to be more quantifiable. Personally I don’t feel confident at this point to make an assumption about whether a software regulatory system could produce a response (especially with music) as our emotions can. It would seem possible in principal but the idea is so foreign that I would see a decision either way as a presupposition. All in all I find the issue of subjectivity to be such a huge issue that it essentially renders itself a non-issue.

Connection:

The ability to connect with and move in some way one experiencing a piece of art maybe one of the most difficult issues that will be faced. We couldn’t be even remotely certain about cross-qualia translation. With art being subjective it wouldn’t seem to be an obvious hindrance but in principal, with assessing an AI’s piece we couldn’t be pragmatic in determining as to whether or not we understood the piece or if we are even capable. I don’t think this is a preventative issue because it exists now to some degree between humans. This also would likely be subject to an ongoing philosophical debate.

Eudaimonism:

Many artists in the community (especially young ones) are enthusiastic about contributing to society with their art. The pinnacle of this in my opinion is through invoking epistemic self-reflection. When a piece causes pause and consideration for ones way of thinking and/or behavior, it has the potential to promote a greater sense of altruism in those who experience it. It would be difficult to speculate as to whether an AI could accomplish such a task. Back to the issue of connection it could be lost in translation or it could be an experience we are incapable of understanding. It could on the other hand bring about a greater understanding in some novel way. This of course is just unanchored speculation and I’m looking forward to being able to experience AI art. This would at least bean opportunity to come to some sort of understanding.

A Prediction:

It would seem that the best way to predict an outcome would be to draw from the outcomes of previous attempts at art by non-humans. There is an added degree of separation with the AI being non-biological as well but I doubt that it will have that much influence. It would be safe to assume that the debates would go through the motions as they have in the past, and when an AI creates a piece of art it will be an interesting enough situation that the piece could command a hefty price or hang in any museum around the world. Since there would as always be no pressure on the art community to come to a consensus on whether it is art; it would just be an addition to many millions of pieces that are shrouded in subjectivity. The question itself is an uninteresting question and the answer is likely yes. Why not?

This may interest you if you are not already aware of David Cope’s work.
http://artsites.ucsc.edu/faculty/cope/experiments.htm
http://artsites.ucsc.edu/faculty/cope/mp3page.htm
I have several of Copes books and found them very worthwhile.

Thanks. I’ll look into this. :slight_smile:

Something you didn’t cover is intent. The definition of art is debatable and subjective, but perhaps not as much as its relation to intent. For example, what if you look out at a beautiful sunset. Is that art? What about beautifully formed, symmetrical rock formations? I would say no, and that intent is crucial to producing genuine art. Assuming you created an art making machine, would it really have artistic intent, or, being a set of predetermined rules with a human specified goal, would it simply be the product of a carefully constructed system? This also calls into question the ownership of the art. If a machine makes a piece of music, who is the composer, the machine, or the person who made the machine?
Personally, I’m creeped out by the idea of machines producing art, and indeed, I’m skeptical that you could even call it art (but again, that’s subjective). To move into a more practical line of thought, what would this mean for the future of art and artists? Let’s say a studio can churn out 20 feature films with the press of a button, and sit there collecting profit while spending hardly any money at all. Filmmakers and artists would be out of a job! An interesting and troubling discussion indeed, but an issue I doubt we’ll have to face for a long time.

There are a lot of philosophical debates revolving around intent itself. I think it would be good to add that but the AI researchers have already been addressing it due to it being a more universal topic. One of the more common arguments is that it’s rooted in the fact that we overestimate our agency. We don’t know what consciousness is and it’s possible that intention could be present with self awareness. It seems to be a common notion that complex or even parallel information processing could bring about self awareness. Another school of thought is that it’s quantum mechanical (Penrose Hameroff) but this doesn’t prevent strong AI in quantum computers.

My own personal opinion after a lot of research is that even though conscious thought and cognition occupies a small fraction of our resources and energy we tend to focus on it. We also tend to assume that we are not quantifiable without having evidence to support it. Electrons seem to have some unexplainable connection with each other that can reach across vast distances instantaniously. Panpsychists believe that consciousness is a spectrum…that all matter has the building blocks of awareness and from the evidence, I don’t find it disagreeable. I of course can’t say with any adamantcy whether that is the case in reality but it prevents me from ruling it out. It could be that consciousness is an emergent property of complex self assembling systems. It seems that the best way to find out is to build a complex system that is capable of learning and self modification and see what happens. The issue of AI becoming the dominant intelligence across the board is a common concern. It seems that all any of us could come up with would be pure speculation. We couldn’t guess with any certainty what a more sophisticated beings world view would be. It’s my guess though that one sophisticated enough to create an immersive work would not be a tool for us to use.