Artificial intelligence is all the rage these days. Case in point: while I was watching football this past weekend, there were two television commercials in heavy circulation during the games that featured AI avatars—Siri and Watson—having life-like conversations with actors.
As you may know, I have a few opinions about the prospects and limitations of AI. Recently, I had an email chat with novelist and philosopher of science Victoria Alexander about AI, art, and chance. Alexander’s work focuses on the uses of chance in nature and in fiction and the changing conceptions of chance in science, religion, and art. What follows has been lightly edited for clarity.
One of the talks you give for the New York Council for the Humanities is entitled, “What can Art teach us about Artificial Intelligence?” So, what can art teach us about artificial intelligence?
I ask, “What can Art teach us about Artificial Intelligence?” because I’m aware that general audiences—consumers, movie-goers, administrators—tend to assume that artists can’t teach scientists how to do their business. At best, an artist is seen as a kind of household angel, a well-to-do mother in Victorian England, providing spiritual guidance, emotional support and inspiration, but living apart from the world where things of practical import get done.
I rather think that the art part of human nature is the smart part. Without creativity, without associative thinking that can leap from one fact to another odd fact like a dreamer, no new ideas would come into being, no new questions asked or answered.
Good scientists know this. Many of them are artists themselves. It’s the rest of the population, and even artists themselves, who often seem to be unable to see the true value of artistic contributions. I’ve worked for 20 years facilitating art-science projects. It makes me miserable to see an artist turning himself into a handmaiden of science, using his skill to illustrate (or interpret) scientific concepts. C’mon man, I want to say, don’t make your art about science, use your art to do science.
What can art teach us about artificial intelligence? If science is well-defined terms, quantitative analyses, and rigid algorithms, it is an automaton. If art is evolving, and adapting and making use of fortuitous coincidence, art is what makes matter alive. Art is intelligence; whereas science is information retrieval. We need to understand art if we are going to imitate it with Artificial Intelligence.
What specifically art can teach depends upon the field. To give an example from my own area, biosemiotics, we are trying to understand how and why biological cells seem able to respond, as if intelligently, to environments, not just react. We’ve discovered that even chemicals can have contextualized meanings that can change. Standard cell biology doesn’t have the vocabulary to discuss these kinds of processes, so we borrow from the arts, and with this language, and only with this language, are we able to define the problem and try to figure out how to solve it.
As a Public Scholar defending the honor and worth of the Humanities in this shopping mall we call society, I hope to impress upon others that art is not just nice, it’s not merely a luxury or a distraction; it is an essential part of what makes us capable of intelligent behavior.
One of the not-quite-dead horses I like to beat around here is that evocations of the term science are often saddled in scientistic terms, by which I mean that the term science is a multiplicity disguised as a unity. As a monolith, science is near impossible to define with the precision that scientists demand within their own specific disciplines.
First, let me say that there is a lot of bad science that I don’t consider “science”, poorly conducted experiments, biased statistical studies that present false conclusions, or studies that are intentionally deceptive (product Z has not been shown to cause cancer in mice under certain conditions, so it’s safe). There is a lot of bad art that is either too rigid and has nothing new to say, or is too loose and has no language and so can’t be considered “art”. But that’s not what we’re talking about. You want to know if Science and Art in theory, not practice, can be defined.
Any species of thing is hard to define. Everything is at least a little fluid. At the risk of plunging this discussion into obscurity, I will draw upon emergence theory. All stabilities in nature (like categories, species, even matter) are dynamically stable. There is no such thing as static stability. Everything we know to be predictable derives its predictability from lower level dynamical interactions, constant change, randomness. Shades of Heraclitus. But I do not fall into the mire of thinking that, because all is in flux, we cannot define things as stable entities. Emergent properties are as real as things get, the only kind of things we have. An emergent is not disguised as a unity; it functions as a unity—because of its multiplicity and not despite it.
Let me put it this way, if the definition of science were too rigid, the first anomaly that came along would put an end to it. It exists and continues because it is flexible; science by nature is adaptable.
So I do make distinctions. This is art, that is science. That is not art, that is not science. I also conflate the two. You cannot have science without art, and there is no art without science. Science is a habit of thinking that forms over time. It is reliable insofar as it is constantly tested against the world. These experiments usually strengthen the habit and sometimes they destroy it, allowing a new habit to form. The art in science occurs at the moment the new habit begins to form. A Kuhnian paradigm shift. Let me stress that the artistic moment is not the moment of destruction; it is the moment of the emergence of the new habit.
You write that “art is intelligence” and that it’s an “essential part of what makes us capable of intelligent behavior”. At the risk of making the same blunder, can you define art? Or in other words, with our anthropologist hats on, what social function or functions does art perform? Surely, it’s much more than just associative thinking.
A lot of art, like science, involves algorithmic thinking, automated responses, information retrieval, counting, measuring and modeling. Without these habits or conventions, these rules we have discovered, no new thing can emerge.
Associative thinking is key. It is also called abduction. We tend to understand the world in terms of the habits we have evolved over our lifetimes and with the structures that we have inherited that make our habits possible. These are our representations of the world and also our self-representations. We tend to recognize that which we have the capacity to recognize. So this leads to the question, How do we ever learn something new? First, there is our very important tendency to generalize and to ignore differences—every similar thing is somewhat unique by virtue of changing contexts, if nothing else. So our responses have to be a little flexible. This flexibility makes mistakes possible. We can mis-recognize X, something unknown, as Y, something known, because X and Y are structurally similar. Now we are able to interact with X, re-purposing old habits and structures. But this interaction may not lead to the same kind of outcome as an interaction with Y would have done. If so, the habit of recognizing Y will not be reconfirmed. This may destroy the habit altogether if that outcome was essential for survival, or it may lead to another habit, an adaptation.
As a biosemiotician, I study these kinds of processes in biological cell interactions. Cells, like artists, can make generalizations and mistakes, and if these mistakes turn out to be good for the cell, they are self-affirming and habit forming. (There are unhealthy habits that are self-affirming too, for a time.) The ability to maintain habits, evolve, and self-perpetuate is what distinguishes a living system from a non-living system.
A living system, then, is one that practices both science and art. This process of associating disparate things to create new meaningful behaviors (new habits) is the process of semiosis. So, art does not have to be more than just associative responses (poesis) against the background of habit (semiosis). That’s a lot.
The social function of art is the way it helps us overcome our bad habits of thought, self-affirming practices that have become disconnected from reality. Biases, ideologies, consumption habits, and institutions are built upon useful generalizations that at some point can start to do ourselves and our world harm.
In “Chance, Nature’s Practical Jokes, and the ‘Non-Utilitarian Delights’ of Butterfly Mimicry”, you use the example of the faux grub-bored holes in the dead-leaf butterfly’s wings to argue, echoing Vladimir Nabokov, that the superfluous detail of this mimicry suggests that its morphology was not shaped by a gradual process of natural selection, but rather, by “pure chance”. You also write, though, that “Nabokov saw evidence of a kind of protointelligence in evolutionary processes”, that he “associated nature’s use of chance with interpretation”.
Does your “though” above indicate that you see chance events and interpretation as opposed here; whereas mimicry shaped by natural selection depends upon the interpretations of predators? If so, yes but…
Darwinism wants to have function (reproductive fitness) as the main driver of evolution, but it doesn’t touch upon the source of the structure, to be selected or not. It supposes new structures are just the product of thermodynamics, normal disordering tendencies, ever so slight random shufflings of the deck. If what is “new” in the world is barely distinguishable from what was “old”, then natural selection is seen as the all powerful, slow but steady, driver of really new creation. There is something very prosaic about this that Nabokov did not like.
Engineers work like this, making small improvements bit by bit, improving efficiency, more than totally re-designing. Artists (and inventors) re-conceive known tools and find utterly different ways to use them. They have to be a lot more interpretative, seeing things in ways that others (who are following the rules very closely, making things better and better but not essentially different) can’t see.
If the grub-bored holes appeared by chance, without the help of natural selection, then this might be proof that other structures also appear suddenly. These other structures (unlike the grub-bored holes) could be useful and, if so, then could be selected. (Hopeful Monsters.) Natural selection in such cases would only explain how these new creatures proliferated, not how they were shaped. It may be that Nabokov’s “non-utilitarian delights” help us see the real role of natural selection more clearly. A popularizer, not a creator.
Is Nabokov’s “pure chance” theory somewhat less interesting than the idea that one butterfly species can be driven by selection to look like another? People often say, it can’t be “just” chance, associating chance with meaninglessness or the absence of explanation.
If you’d humor me, I want to posit an alternative hypothesis to explain, at least in part, the delightfully superfluous detail of the dead-leaf butterfly’s wings. Is it possible that the female dead-leaf butterfly, in the course of choosing a mate, might find faux grub-bored holes more beautiful, thus the mate who possesses them more desirable?
Any reproductive advantage of a new form might cause it to increase in number, but the new form itself might not have been gradually shaped by selection. It might have emerged suddenly, by chance, and then found a use.
Sexual selection is often used to explain forms that don’t seem functional on their own, like overgrown tails that are more an encumbrance than not. But peacock tails might be caused by gene reduplication instead, or some other mechanism of orthogenesis. A gene segment is sometimes copied and doubled and this usually results in more and more of the same, longer tails, bigger skulls, shorter tails, etc. Gene reduplication can go unchecked as long as resources are plentiful. If this is true for peacock tails, then selection is not necessary to explain how they came to exist.
Or it may be that sexual selection is actually being driven by an increase in pheromones that happen to coincide with greater tail or horn growth or whatever. Many forms are linked genetically; the same gene can control different processes in different contexts. So maybe it’s not what the ladies see so much as what they smell. (I think pheromones on butterflies are excreted from the wing area and a new wing design could mean different hormone levels, I don’t know.)
I’ve read studies that have tested sexual selection, trying to isolate a physical difference to be sure it is not some other variable getting female attention. For example in one study, researchers put little hats on canary heads to see if the ladies preferred them. The study claimed they did, but they had graduate students watching the birds and making subjective judgements about what the ladies were feeling. Not scientific enough for me.
Nevertheless, I am open to the idea that sexual selection has gradually shaped many forms, especially “non-utilitarian delights”.
In effect, a semiotic choice reinforces and accelerates a chance mutation. Or in other words, to what extent do interpreting organisms play an active, even conscious role in shaping their own being?
I do believe that animals, all living forms, even bacteria, make semiotic choices, and this is a driver of evolution. But making the scientific argument that we, much less other simpler animals, can make choices is very difficult. What we call free will, making choices—not just acting according to what our biology says, our chemistry says—comes down to how well an individual is able to make his own luck.
To go back to some things I wrote earlier, the habits of interaction that an organism uses to negotiate with its world are its semiotic constraints—it can only interact with what is meaningful/useful to it. To interact with the world in a new way, the organism has to stumble upon a chance similarity that seems close enough to something known, but is really very different. An organism that has interacted more has developed greater semiotic complexity and is therefore more likely to stumble upon new ways of being.
To bring this point down to Earth, people who speak several languages, work in multiple areas, have traveled widely, have studied the minutiae of some topic, or have read a lot of poetry tend to be lucky in stumbling on new ideas. Also people who are superstitious or crazy tend to connect things that should not be connected; they tend to get lucky too.
There are other good arguments that animals do play an active role in their evolution. An animal’s tendencies can affect the probabilities of genetic changes occurring in or being passed on to its offspring. The sciences of epigenetics, symbiogenesis, and lateral gene transfer give ample evidence of this. Robert Reid’s Biological Emergences discusses this as a new kind of Lamarckianism.
I love your definition of the “artistic moment” as “the emergence of a new habit”. Smartphone use is becoming increasingly notorious for the ways in which it retrenches habit, and often, of the bad kind.
Our connectedness means that selection could cause a new form to spread very rapidly. So a new political ideology might spread from a single tweet. Unfortunately, our interconnectedness tends to wash out novel ideas rather than promote them. We are becoming very homogeneous, not more complex.
Can you envision a situation in which a smartphone could provoke “artistic moments” rather than inhibit them?
We need more niches, groups cut off from the crowd, to provide the space for novelty to emerge. We need decentralization. We need diversity. Can a smartphone help with that?
If we are talking about whether or not computer code can be artistic: A hacker who turns spyware on itself and broadcasts the results? A search algorithm that points you to what you’ve missed instead of what you expect? A computer virus that transforms itself based on interactions, starts doing things its designers had not envisioned? An app that helps everyone start using secret languages?
Sean Miller holds a PhD in English from the University of London. His book, Strung Together: The Cultural Currency of String Theory as a Scientific Imaginary, is available from the University of Michigan Press. He’s the co-founder ofReaderly, an edtech company based in Portland, Oregon. Readerly makes an app that hones critical reading skills through interactive gameplay.