AI and Art III: The WorkShare
A look at AI and art in the context of defining art in terms of the qualities of the work.
While it certainly makes sense to consider the qualities of the creator when determining whether a work is art, it also makes sense to consider only the qualities of the work itself. On this approach, what makes a work art are these qualities. Naturally, it also makes sense to consider the effect of this qualities on the audience as a key part of sorting out art. For example, David Hume’s somewhat confusing theory of beauty seems to make beauty a matter of how the qualities of an object affect the audience. Other thinkers, such as Plato, take the quality of beauty to be an objective feature of reality. Defining art in terms of objective beauty would seem to entail that the qualities of the work determine whether it is art. Since one could go on almost forever considering various qualities, it is fortunate this essay does not require a theory of what qualities of a work make it art. All I need is the hypothesis, for the sake of the discussion to follow, that something being art is a matter of the qualities of the work—whatever they might be. This hypothesis is, of course, easy enough to challenge.
One practical reason to focus on the work itself rather than the artist (or other factors) is that there can be cases in which one has a work and yet lacks information about the artist and even about the context of the work. For example, the creators of most ancient works of art found by archeologists are unknown. As such, judging whether they are art requires judging the work itself. Or forever shrugging when asked whether it is art. This can, of course, be countered by pointing out that the creators were human, and much is known about humans that can be applied in sorting out whether the work is art.
One way to counter this is to imagine works found that predate humanity or alien works found on another planet by xenoarcheologists. It is easy to imagine that we might know nothing about the creators of such works. As such, there would be two possibilities. One is to claim that there is no way to judge whether the work is art. The other is to accept that the work can be judged on its own, keeping in mind the obvious fact that the assessment could be in error.
Another way to counter this is to consider the case of AI created works in the context of an aesthetic version of the Turing test. The classic Turing test involves two humans and a computer. One human communicates with the other human and the computer via text with the goal of trying to figure out which is human and which is the computer. If the computer can pass as human long enough, it is said to have passed the Turing test. An aesthetic Turing test would also involve two humans and one computer. In this case, the human artist and the art computer would each create a work (or works), such as music, a sculpture or a painting. Naturally, the test must be set up so that it is not easy and obvious which is which. For example, using a human artist whose style is well known and a bad art program would not be a proper test. Matching a skilled, but obscure, human artist against a capable art AI would be a fair test.
After the works are created, the human judge would then attempt to discern which work was created by a human and which was created by a computer. The judge would also be tasked with deciding whether each work is art. In this case, the judge knows nothing about the creator of a work and must judge the work based on the work itself. While it is tempting to think that a judge will easily tell a human work from a machine work, this would be a mistake—AI generated art can be quite sophisticated and can even be programmed to include the sort of “errors” that humans make when creating works. If the AI can pass the test, it would seem to be as much an artist as the human. If the work of the human is art, then the work of the AI that passes the test would thus also seem to be art.
If whether a work is art depends on the artist, then a judge who could not discern who created the two works in the test would not be able to say which (if any) work was art. So, if a computer created a brush-stroke by brush stroke identical work as the human and thus assured that the two works could never be distinguished, it would follow that the judge must rule that neither work is art. However, this seems to be an absurd conclusion. One could also imagine a joke being played on the judge—after the judgment, they are told that painting A is by the human and B is by the computer and then they are asked to judge again. After they reach their verdict, they are informed that the reverse was true and asked to judge again. As such, having art require a certain sort of creator seems absurd—what makes something art should be in the work itself.
One way to counter this is to use an analogy to a perfect counterfeit of a $100 bill. While the perfect counterfeit would be identical to the real money and utterly indistinguishable to all observations, it would still be a counterfeit because of its origin—being legitimate currency is not a matter of what makes up the money, but how the money is created and issued. The same, it could be argued, also applies to art—thus a work created in the wrong way would not be art, even though it could be identical to a real work of art. But, of course, just as the perfect counterfeit would seem to destroy the value of the real bill (if one is known to be fake, but they cannot be told apart, then neither should be accepted) the “fake art” would also seem to destroy the art status of the “real art.” This would be odd but could obviously be accepted by those who think that art, like money, is a social construct. But, suppose one accepts that being art is matter of the qualities of the work.
If it is the qualities of a work that makes a work art and an AI can create works with those qualities, then the works would thus be art. If an AI cannot create works with those qualities, then the work of an AI would not be art. The real challenge here is working out a theory of art that successfully sorts out the necessary and sufficient conditions for such works. Once that is done, it will be possible to know if an AI can create art.