The word creativity is strangely hard to define.
What makes a work "creative?" The Stanford Encyclopedia of Philosophy states that to be creative, a the created item must be both new and valuable. The "valuable" part already skates out over thin ice, because it immediately raises the question of "valuable to whom?" I've seen works of art -- out of respect to the artists, and so as not to get Art Snobbery Bombs lobbed in my general direction, I won't provide specific examples -- that looked to me like the product of finger paints in the hands of a below-average second-grader, and yet which made it into prominent museums (and were valued in the hundreds of thousands of dollars).
The article itself touches on this problem, with a quote from philosopher Dustin Stokes:
Knowing that something is valuable or to be valued does not by itself reveal why or how that thing is. By analogy, being told that a carburetor is useful provides no explanatory insight into the nature of a carburetor: how it works and what it does.
This is a little disingenuous, though. The difference is that any sufficiently motivated person could learn the science of how an engine works and find out for themselves why a carburetor is necessary, and afterward, we'd all agree on the explanation -- while I doubt any amount of analysis would be sufficient to get me to appreciate a piece of art that I simply don't think is very good, or (worse) to get a dozen randomly-chosen people to agree on how good it is.
Margaret Boden has an additional insight into creativity; in her opinion, truly creative works are also surprising. The Stanford article has this to say about Boden's claim:
In this kind of case, the creative result is so surprising that it prompts observers to marvel, “But how could that possibly happen?” Boden calls this transformational creativity because it cannot happen within a pre-existing conceptual space; the creator has to transform the conceptual space itself, by altering its constitutive rules or constraints. Schoenberg crafted atonal music, Boden says, “by dropping the home-key constraint”, the rule that a piece of music must begin and end in the same key. Lobachevsky and other mathematicians developed non-Euclidean geometry by dropping Euclid’s fifth axiom. Kekulé discovered the ring-structure of the benzene molecule by negating the constraint that a molecule must follow an open curve. In such cases, Boden is fond of saying that the result was “downright impossible” within the previous conceptual space.
This has an immediate resonance for me, because I've had the experience as a writer of feeling like a story or character was transformed almost without any conscious volition on my part; in Boden's terms, something happened that was outside the conceptual space of the original story. The most striking example is the character of Marig Kastella from The Chains of Orion (the third book of the Arc of the Oracles trilogy). Initially, he was simply the main character's boyfriend, and there mostly to be a hesitant, insecure, questioning foil to astronaut Kallman Dorn's brash and adventurous personality. But Marig took off in an entirely different direction, and in the last third of the book kind of took over the story. As a result his character arc diverged wildly from what I had envisioned, and he remains to this day one of my very favorite characters I've written.
If I actually did write him, you know? Because it feels like Marig was already out there somewhere, and I didn't create him, I got to know him -- and in the process he revealed himself to be a far deeper, richer, and more powerful person than I'd thought at first.
The reason this topic comes up is some research out of Aalto University in Finland that appeared this week in the journal ACM Transactions on the Human-Robot Interaction. The researchers took an AI that had been programmed to produce art -- in this case, to reproduce a piece of human-created art, but the test subjects weren't told that -- and then asked the volunteers to rate how creative the product was. In all three cases, the subjects were told that the piece had been created by AI. The volunteers were placed in one of three groups:
- Group 1 saw only the result -- the finished art piece;
- Group 2 saw the lines appearing on the page, but not the robot creating it; and
- Group 3 saw the robot itself making the drawing.
Even though the resulting art pieces were all identical -- and, as I said, the design itself had been created by a human being, and the robot was simply generating a copy -- group 1 rated the result as the least creative, and group 3 as the most.
Evidently, if we witness something's production, we're more likely to consider the act creative -- regardless of the quality of the product. If the producer appears to have agency, that's all it takes.
The problem here is that deciding whether something is "really creative" (or any of the interminable sub-arguments over whether certain music, art, or writing is "good") all inevitably involve a subjective element that -- philosophy encyclopedias notwithstanding -- cannot be expunged. The AI experiment at Aalto University highlights that it doesn't take much to change our opinion about whether something is or is not creativity.
Now, bear in mind that I'm not considering here the topic of ethics in artificial intelligence; I've already ranted at length about the problems with techbros ripping off actual human artists, musicians, and writers to train their AI models, and how this will exacerbate the fact that most of us creative types are already making three-fifths of fuck-all in the way of income from our work. But what this highlights is that we humans can't even come to consensus on whether something actually is creativity. It's a little like the Turing Test; if all we have is the output to judge by, there's never going to be agreement about what we're looking at.
So while the researchers were careful to make it obvious (well, after the fact, anyhow) that what their robot was doing was not creative, but was a replica of someone else's work, there's no reason why AI systems couldn't already be producing art, music, and writing that appears to be creative by the Stanford's criteria of being new, valuable, and surprising.
At which point we better figure out exactly what we want our culture's creative landscape to look like -- and fast.
No comments:
Post a Comment