As a philosopher, I tend to be concerned with meta-aesthetic matters, such as trying to define “art” or sorting out whether an AI can create “true” art. David Pogue has addressed the subject of AI and art more pragmatically by focusing on concerns about the economic impact of AI on art and artists. That is, the question of whether AI will be taking the jobs of artists. The impact of AI on art is certainly worthy of philosophical and pragmatic consideration, thus this series of brief essays of which this is the first.
As Pogue noted, AI (broadly construed)is already producing original paintings and music. It should be noted that these computer programs are relatively limited. For example, Amper is a guided system for assembling samples rather than an AI Mozart or Prince. To use an analogy, it is like an exoskeleton for music: it does the hard work but must be guided by a human operator. It is, however,capable of producing high quality results—those worried about digitized junk music will be pleasantly surprised.
As researchers and companies improve the technology, AI will continue to expand its reach into the realm of art and will also increase the variety and quality of its creations. As such, concerns about AI and art are not the stuff of science-fiction, but a real-world concern right now. And, like climate change, something that will only increase.
From a philosophical standpoint, a critical question is whether AI created works are art. The obvious problem with sorting this out is that there is, as far as I know, no necessary and sufficient definition of “art” that would allow a decisive and objective answer. As it now stands, the question can only be answered within the context of a specific theory. That is, the specific question is whether AI art is art under this or that theory. To use an analogy, being a work of art is rather like being a sin.Whether something is a sin or not is a matter of a specific religion. That is,the specific question of whether an action or thought is a sin is whether it is a sin in this or that religion (or interpretation of the religion). This is distinct from the question of whether it truly is a sin. Answering that would require determining which religion has it right (and it might be none of them—there might be no sin at all). As such, I cannot answer whether AI art is art with certainty until I know which, if any, theory of art has it right (if any). That said, it is possible to muddle about with the usual cobbled together messes of existing theories.
One broad distinction between theories that is especially relevant to AI art is between theories that focus primarily on the work and theories that focus primarily on the creator. In general terms, the first sort of approach involves art requiring certain properties in the work. The other sort of approach involves requiring that the work be created in a certain way by a certain sort of being. These approaches will be discussed in upcoming essays.
From a pragmatic standpoint, the key concern about AI and art is the impact of AI on the economics of art. This includes the general concern about machines taking jobs from humans and the impact of automation on the economic value of products. To illustrate, if someone who needs music for a Youtube video or TV show can simply be guided through its creation by an AI, then they have little reason to hire human musicians to create it for them—unless the humans are cheaper. As another example, if an AI can create thousands of original digital images with each click of the mouse, then what impact will this have on the value of visual art? While it is tempting to think that machines will not be able to match the creative abilities of humans, this seems to be mere wishful thinking—if programs are not already “taking jobs” from human artists, they soon will be. While these are practical matters, they also raise philosophical concerns and will be addressed as well in upcoming essays.
My only thought so far is: in what way is this different from computing, or automation in general, replacing other forms of work? Everything from horseshoes and Fords, through graphics and animations that were once drawn by hand, to modern Watson-type applications in health, finance, general advice like Siri and Alexa?
Is there some categorical difference to be drawn? If so, where?
Good question.
There is the usual automatic assumption that creating art is somehow distinct from manufacturing a car; that art requires a soul and feeling while making a car is just a matter of following a blueprint. But this is something that must be argued rather than simply assumed. After all, creating works that seem to be art usually involves following rules about perspective, timing and so on. Even when a creator breaks rules, that could also be made into a rule for breaking the other rule. So, my view is that there is not a distinction in kind between the arts and the crafts. But, there are certainly good arguments against my view.
I like to joke that one of the main reasons that my wife and I have been together for so long is that we each accept our place in the household, and divide up the responsibilities equitably. In general, she is responsible for all the small, meaningless, day-to-day issues like where we should live, where the kids go to school, when and if we should buy a new car, renovate our kitchen, paint the house. She’s also in charge of the budget, saving for retirement, investments – all the little stuff.
I, on the other hand, have much greater responsibilities – like “What is art?” So with that experience, I feel uniquely qualified to answer your question about AI.
When I was in college, I took a few courses in music theory. One of the ongoing homework assignments we had was to take a simple melody and add three additional voices to it, to create a four-part harmony. We began with a certain set of rules – we had to stay within a tonal center, for example; we had to avoid parallel fifths and octaves, and we learned about certain “cadences”, pleasing chord sequences and resolutions.
This was an intro class; none of us really knew much about music – but we all were able to produce nice sounding harmonies if we followed the rules.
Where did the rules come from? Observation. Musicians and musicologists analyzed thousands of pieces of music from the 17th and 18th centuries, and derived the rules from the established norms of the day.
New rules were added – chord inversions, modulations, drifting away from the tonal center and coming back again – and with the new rules came greater complexity and variation – but they were still rules. It’s kind of like a logic tree – you start with one concept (chord), then you have two or three choices of where to go next – then depending on what choice you make, you have more choices – etc, etc – and each choice takes you in a certain direction and gives you several forks in the road.
It’s not hard to imagine that even a very limited AI engine can be taught how to do that. So when does it stop being a set of rules and variations, and start becoming art?
Take a little time and watch this video:
https://youtu.be/eRkgK4jfi6M
At about 5:00, when Collier is talking to the professional musician about harmonies, his discussion of theory gets incredibly complex – but it’s still a set of rules with a couple of added factors – rules about various innovations, and rules about how to break the rules.
But the conversation shifts from theory to emotion, to “feeling”, to “shared experience”, and that’s where the AI will fail, in my opinion. Human experience is ongoing, it’s fresh, always new, and there are no rules about how a person should or will “feel” in any given situation. So when an artist (broadly defined – musicians included) has a spontaneous moment of inspiration and connects in a very deep, visceral way to the viewer or listener – that becomes “art”. This can also lead to an intellectual process that combines emotion with theory, and the artist will produce something totally new – that has never been done before -that will make some people laugh or cry, and others start having deep discussions about theory and practice and history and methodologies.
At about 11:340 in the video, after talking to Herbie Hancock about some complex structures and innovations, Collier begins talking about the emotional impact of a particular variation – how once you get to a particular place in the structure there builds this tension – an inevitability that he calls “glorious”.
Hancock calls it a “pull”; and Collier says,
“I love talking about these sorts of things emotionally because that’s a “feeling” – I know that feeling in my life”.
Later on, Hancock is talking about an experience he had playing with Miles Davis. He had been in a creative rut for a long time – and as he was playing, Davis suggested to him that he leave off the bottom notes. Hancock mis-heard him, and thought he said “butter notes”. He tried to figure out what Davis was talking about – and that serendipitous error not only got him out of his rut, but led to a burst of creativity that he said garnered him more applause than he had ever had before.
So here’s the thing – whether it’s Herbie Hancock or John Coltrane or Mozart or Bernstein – they all learn the rules and then somehow are able to make a deep human connection with themselves and with others; it’s based on emotion, on feeling, on humor, and on life experience. And once they do that, the rules change!, and those who come after these artists learn a whole new, expanded set of rules.
No matter how complex an AI system is, by definition it will never have human experience. So any “art” created by AI can only be created from within the rules, the rules about breaking the rules, the rules learned by human innovation, and a sort of structured randomness. I’ll never say “never”, but I think it will be a long time before Big Blue can say, “That chord structure somehow brings out the same feeling in me as when my father left when I was eight”.
…continued on next post
This takes us back to that fundamental Cartesian problem. Our good dead friend Descartes argued that machines could be made to perform very impressive tasks (he even anticipated talking machines), but they could never have minds–because for him the mind is an incorporeal thinking substance while a material machine would be matter that can never think. Hume argued against him on this point and this battle has infected sci-fi, philosophy and AI theory ever since.
If our mind is material rather than some special substance, then having non-organic thinking and feeling machines would seem to be fundamentally an engineering problem. That is, how do you get silicon to do what carbon does? Now, it might turn out that there is something unique about the organic that precludes anything non-organic from thinking and feeling. But this is a question for empirical science.
To your other, more mundane point about artists and jobs – there really is something to that, but I’d argue that there’s a semantic difference there. Someone working as a professional artist is really someone who is well-versed in visual media; understands composition, form, positive & negative space, implied line, color, harmonics … and is able to very naturally produce very high-level artwork that is heavy on formula, and light on spark or spontaneity or innovation.
I had this discussion with students in an aerial photography class. We were flying drones that have incredible stabilization and way-finding mechanics, and are equipped with very, very sophisticated digital cameras. It is possible to set the cameras to “automatic” with some set variables (“cinematic”, “high contrast”, “High Dynamic Range”, etc) and almost without thinking, produce what I call “Instant Awesome”. The combination of perfect exposure and a very new point of view leads to a whole new genre of unexplored art, where nearly everything is an innovation.
But I cautioned the students that technology is like a Siren – it will lure you in and ultimately kill you if you let it. Drones are proliferating – they are getting more and more sophisticated and more and more accessible – and it will not be long before this “Instant Awesome” becomes passe, and the really great imagery being produced today becomes the stock catalog of tomorrow. It won’t be long before a drone can be programmed with some kind of AI function, and we’ll be able to send it out to take a morning’s worth of awesome pictures and return before we finish breakfast.
And yes, for those who color between the lines, their jobs will be threatened. Not that coloring within the lines is a bad thing – there’s a pretty wide range of variation that will keep these people ahead of the curve for quite a while – but ultimately the technology will catch up. But there will be a small handful who stay ahead of the curve, and are able to understand the shortcomings of even the most brilliant technology.
Think of the difference between Monet, for example, and Bob Ross.
Good point. As you said, a person could just set out a fleet of camera bots and then go through the results looking for the best shots. As you noted, chance will result in some amazing pictures. The same could also be done with painting bots and even writing bots.
DH, I think we’re agreeing on the potential of computing to replace humans in the “craft” part of art – and all arts require competence in techniques that constitute the craft.This has happened, is happening, and will continue to happen.
The idea that art is somehow disconnected from anything else, as higher, unavailable to machines, ineffable, part-mystical, though, is a view I have moved away from.
The difference between Monet and Bob Ross is that, as you say, Bob Ross followed the rules, and painted between the lines – having learned the rules for where the lines should be. And Monet broke rules. But many painters broke rules. We remember Monet because he broke the rules at a time when fashion had stagnated, did it with some talented others, and caught public attention. Many other artists also broke rules, no doubt some in brilliant ways, but did not catch wide attention. And so their works are forgotten, or lost.
Not least in my change of view was the realisation that most art – the vast, vast majority – simply doesn’t work. We may think psychology and sociology have terrible replication and validation, but they are paragons of effectiveness and consistency compared to art. Most art leaves me unmoved. Most art leaves most people unmoved, Some random sights, sounds, smells can be evocative to me, as can some works of art, especially those I associate with specific experiences in my life – the song that was playing everywhere the week my mother died, or the songs I listened to with friends or alone as a teenager, for example; but that’s not because of any artistic merit. Pick a hundred people at random, and put them in front of a hundred pieces of art that they haven’t seen before made by humans and a hundred made by computer. Get their honest reactions to each, having primed them not to stretch. Perhaps measure objectively, with biometrics or fMRI. Hmmm, that should be a project for someone.
Also, art goes in and out of fashion. J.S. Bach’s music was effectively erased for over a century. If art has intrinsic merit, how can he have been considered a) a top composer and innovator b) a mechanistic “mathematician”, not worth playing and c) one of the greatest handful of composers of all time in different eras?
(BTW I will be moving around over Christmas so may be slow to respond to any follow-ups.)
The — usually anonymous — computer programmer as Auteur?