Some quick notes on the irrelevance of Turing-like tests for differentiating the creations of humans and machines.
Let’s forgo the cruft and dive right into the question: Is there a fundamental difference between the artistic creations of a human and those of a machine? We can hardly finish stating the question before somebody has approached with long stride, solution in hand. Inevitably, their solution will take the form of a Turing-like test: we take some subjects, and ask them to distinguish between images (or texts, films, or what not) created by humans, and those created by machines. If our subjects are unable to meaningfully distinguish between the two then the question is settled, as no practical difference remains. There would then be no difference between AI and human art, or so the story goes.
It is a strange story however, stranger still in that it declares itself as a solution to the problem. As a solution, we should expect it to be able to resolve the problem, to serve as a persuasive tool. It should at least have the minimal power to change some minds. A solution to a question which sways neither the inclined nor disinclined has no effect, and is hardly a solution.
We can imagine the response to such a test as easily as we can imagine such a test being conducted. If the test failed at showing indistinguishability, those disinclined to maintain a difference would state that the test was premature, that the models used were outdated, and lacking in parameters. If the test succeeded, those inclined to maintain a difference would note that the test was methodologically flawed, that the artworks used were outdated, and lacking in originality (a poor sample of human ingenuity). Indeed, objections could be made regardless of how well designed the test is. It seems almost by the nature of the question that the conclusion of any test would leave an open question in the air from which an objection to the test can grow and thrive. This would not be the case for all questions of course: a regular consumer of cured salami may drop the habit upon seeing a well designed study linking it to cancer. But in our case, there is no conceivable scenario where the individual who maintains the difference drops it, all because of a test.
But the tester is unperturbed, and may take the above further if so desired: If the test was rejected by the disinclined, they must have definite reasons for doing so (sample bias, naive parametric assumptions, etc). To not have definite reasons here would be unreasonable. Given their reasons for rejection, we may run another test which addresses said issues; a test that is bigger, bolder, and hence more statistically valid. Undoubtedly, the disinclined would reject this test as well. Reasons could again be given–there is never a lack of reasons to reject a statistical test, and acceptance only indicates a will too exhausted to find faults. So again, another test, another rejection, more reasons, and so on, ad infinitum. “Now, wouldn’t the disinclined look a bit foolish after a while, with their constant nitpicking?” The tester could then pull out a comparison to water diviners whose beliefs, while not entirely refuted, have begun to look a bit ridiculous under the weight of growing evidence.
The case above bears resemblance to the man who, not sure whether fool’s gold is equivalent to real gold, runs an indistinguishability test, and upon receiving a successful result runs out into the street in his pajamas, holding bundles of pyrite and shouting: “Hoozah! They are the same, and now I am rich!” We laugh at such a person, regardless of how well his test was set up, for we know that pyrite and gold are different, as they differ in their constitutions . The test showed that they were indistinguishable. It did not show that they are the same.
It it likewise in our case; they are different, and they differ via their constitutions. In the case of pyrite and gold, this fact just points to another kind of test, a chemical one that reveals the molecular compositions of the objects in question. In other words, they must both be broken down in a kind of analysis, into simpler parts which can be compared so conclusions can be made. But if we were to take any artistic creation and place it under a microscope, we would not find its parts to be simpler, but to be infinitely more complex. Each part, for it to be an actual part, and not a mere abstraction, must be constituted by the same web of values, purposes, and beliefs that our simplest gestures take meaning upon. If not, then it is in some way disconnected from life, and thus exists only as a possibility, an abstraction, a figment. And so it is that within even the smallest piece of text is contained the entire world.
Such grandiosity may seem absurd
To put this testing business behind us we should finally ask: is the difference between machine art and human art something definite enough to be tested or even given a clear delineation? If we mean art in its actuality (and not as high abstraction) then this is plainly false–at least for the time being. The statement must be restricted to the present, unfortunately. A shared web of beliefs is not a static edifice. It changes little by little, day by day, resembling only dimly what is was some time ago. For the time being, art still forms a legitimate part of the world, and because of that it remains linked in the web, in touch with actuality. But values can be changed, and they can be changed in such a way that we untether art from this web, leaving it to float in the vacuum of space like a jettisoned astronaut.
Thus, in the end, the difference between human and machine art is something that can potentially be settled by a definite test, but only in the case where the role of art in life has been reduced to a nullity. By then of course, the test would say nothing. So the real question is not “is there a difference?” but “should we maintain a difference?” Whether or not there is a difference is our collective choice to make, and if we see value in this difference, it is our fight to maintain it.
(Thumbnail image: Roger Bacon, conducting a scientific test. From the Middle Temple Library)