On the artificial artist
In discussions of chance, it’s sometimes remarked that if a hundred million monkeys were tapping away at an equal number of typewriters for an incredibly long time, one of them would, by pure coincidence, eventually reproduce the collected works of William Shakespeare.
There are, however, several problems with this conclusion.
One is that that these monkeys wouldn’t actually produce a truly random output, but rather conform to patterns of motion and action that inhere in their own physiology, which arguably could never reproduce said oeuvre. Another, and a bit more complex counter-argument, relates to what formally characterizes a truly random output, and how this fatally contrasts with the rule-governed character of the Shakespearian works as a set of abstract objects. I.e. they’re formally speaking not a random pattern, but a highly ordered set, and as per the second law of thermodynamics, it’s entropy rather than order which increases over time (so a random output over time will tend towards less order and complexity).
Furthermore, there’s an additional line of argument pertaining to the nature of an artwork as a relational product of actual intentionality with regard to both genesis and reception which undermines the viability of genuine artificial creative agency even if we disregard the aforementioned problems relating to the generation of original artworks and similar output.
(AI artwork generated from the keywords “post apocalyptic horse wind fire stairwell”)
The arguments from this discussion can be transposed to issues of creativity and rational agency of hypothetical artificial persons in the AI discourse. Arguably, the rule-governed behaviour of such entities could not even in principle generate the abstract objects associated with complex human creativity, but at best only imitate actual artefacts of this kind. This post will explore this category of arguments in relation to a criticism of creative artificial agency, and evaluate which possible conclusions they could allow in relation to actual artificial subjectivity as such.
The first problem above is rather straightforward. Since millions of monkeys don’t actually function as random output generators, they’re rather going to produce effects in line with their own nature. An effect is going to be proportional to its cause (as per the PSR or the principle of causality), so unless we ascribe to the monkey-playwright something akin to independent rational agency, his output will be circumscribed by the patterns inherent to his nature.
And as the same basic principles of cause and effect holds in the case of artificial agents, we’re not going to be able to readily attribute randomly emergent artistic output to such causes either. You can of course feed the machine actual bits and pieces of that sort of output, and implement rules for putting them together in relation to a random seed for a little bit of variety, which is precisely what you see in connection to the famous AI-generated editorials or oil paintings, but this is merely the non-intentional rearrangement of already existent material (or indirect human rearrangement).
Of course, you can try to argue that this is all that creativity really is, and that there’s no genuine deep creative agency anywhere among us sentient animals, but if you bite that bullet, there’s no position left to defend.
To even in principle support genuine creative output by an AI, we’re going to have to ascribe it something along the lines of rational subjectivity.
The second line of reasoning I initially mentioned further compounds the above problem, and relates to the hypothetical products of truly random output. Since random output as such by definition is not rule-governed (not even to an extent, e.g. mostly random against a stabilizing deterministic base), there’s a fundamental problem with the notion that it could engender a highly ordered set such as the Shakespearean works. In other words, what we call random output implies a definite, highly specific, type of pattern (cf. the image of cosmic background radiation), a discernible form if we bring in Aristotelian terminology, and this form or pattern is simply not the same sort of entity as a rule-governed output. The idea that something mysterious designated as random could generate any sort of coherent discernible pattern whatsoever is basically a holdover from David Hume’s radical denial of the principle of causality, i.e. the (incoherent) position that any effect can be produced from any cause, or no cause, whatsoever. This also implies the reception of randomness as a sort of non-cause from which in principle can emerge whichever effect you like.
But the third problem I alluded to in the beginning is all the more wicked for the proponent of genuine artificial creativity.
The key to this issue is that an artwork, or a narrative, or any other sort of conduit of aesthetic experiences, is not simply and finally a specific, circumscribed object, complete unto itself as some sort of Cartesian substance.
An artwork is necessarily a complex, relational reality which must be intentionally apprehended. In that regard, Shakespeare’s narratives really do not exist outside of such a relational network of intentions. They’re not a simple object, separable from this context of intersubjective meaning, which implies that they can’t be dislodged from an actually living and conscious context of causation that brings the narratives about as comprehensible aesthetic experiences.
In other words, if a hypothetical AI equipped with a random number generator, and in a cosmos entirely devoid of human or alien intelligence, managed to produce the exact set of markings that we interpret as the Shakespearian corpus, it wouldn’t actually be the Shakespearian corpus. It would just be a pattern of markings with no actual meaning attached to them, just like Lewis Carrol’s “jabberwocky” or “slithy toves”, since our words are actually distinct from the propositions we use them to express.
There is a line of argument that runs somewhat counter to this critique, however, which contends that words are not entirely arbitrary in relation to the propositions expressed. The reasoning is that the words at some deep ontological level to a certain extent reflect the character of the propositions in question (meaning that we humans will tend towards choosing words of a certain character to fit the propositions we desire to express). Still, such an argument will at best only show that an AI lacking all forms of intentionality could stumble on certain patterns of markings, or phonemes, which has a particular affinity towards being intentionally connected to certain propositions.
So we still don’t actually have the Shakespearian corpus produced by such an AI and in that sort of context, but merely a pattern which to a conscious mind has some sort of inclination towards being associated with certain propositions, yet without possessing any kind of necessary connection to the narratives in question.
We can perhaps most easily cash this out in relation to the basic Platonic/Aristotelian ontology, according to which the immaterial forms, of which our conscious intentions are a type, are non-identical to the matter of physical bodies. This also implies that the form of e.g. a triangle can be arbitrarily associated to any symbol or set of symbols in a certain context. So we can for instance play a game where “ABCDE” means triangle, and thus intentionally and consciously take “ABCDE” to refer to the actual triangle as an immaterial form. This intentional “pointing towards” is absent in the example with the AI in the empty cosmos. The ideas do not reduce to their physical manifestations.
(the old railway station house a stone’s throw down the road. by Mikael Lundberg)
When I grew up, my parents had a summer house up in the forest outside of a town called Kalix up north. This house was fenced with young pine and spruce trunks, and when they aged, some sort of parasite or decomposer made these intricate patterns beneath the loosened bark. And while you and I could reverse-engineer these inherently meaningless patterns into the basis of a language, we first need to intentionally and consciously ascribe meaning to them. Aesthetic experiences, or symbolic conduits thereof, are similar. You can’t randomly generate them with no relationship to the actual context of meaning and interpretation.
So what’s the take-home from all of this? Well, my point is that artificial creativity and actual artificial artistry inevitably presumes actual conscious (and I would add, rational) agency. And this means that also this discussion finally boils down to whether or not strong AI in the sense of genuine artificial subjectivity is possible or not. I’ve criticized this notion elsewhere.
The mythical role of creative artificial agency
My friend and once-supervisor Mikael Stenmark coined the concept of “redemptive scientism” back in the mid-90s. Mikael has been an avid critic of the irrational excesses of instrumental reason for the last thirty years, and he used this particular notion to delineate and examine the conviction that modern natural science, i.e. the objective description of physical reality (considered a set of inert objects devoid of meaning and intentionality), can somehow bring us salvation, perfection and ultimate fulfilment.
I don’t think many people explicitly and consciously subscribe to this belief. All the same, I’d argue that it’s a hugely influential theme of the set of worldviews that make up secular modernity. There’s a clear connection here to critical theory’s emphasis on the ubiquitous societal infiltration of exchange value, and how this relatively impoverished abstraction ultimately derived from instrumental rationality and displaces essential and relational thisness and rootedness. A tendency towards hope and trust in artificial agency on a framework of redemptive scientism is at heart a manifestation of the subsumption of spirituality and religious intention.
Of their integration beneath capitalism’s value form and its implied trust in the graces and fruits of instrumental rationality.
(another AI-generated artwork, the keywords I used this time were “capitalism steel brutalism cybernetic brain-link poverty concrete”.
the irony inherent in the fact that this is a wonderfully apt illustration of the entire discussion is truly painful)
+++
It’s like the creative artificial agent in some sense is the personified manifestation of the hopes and aspirations inherent to the myth of Faustian progress, the conceptual substructure of the modern West and industrial civilization in general.
The creative AI is in this way modernity coming full circle, reinstating the god of the machine. In this context, it’s not really important that, as John often interjects, “it [artificial intelligence] doesn’t fucking work!” It’s a myth of immanent salvation being revivified with violent fervor at the end-stage of technological civilization. An apocalyptic ritual of renewal towards the resurgence of faith in the decaying technological system.
And this is really why the entire system of digital surveillance is rather palatable to the majority. Why my well-educated acquaintances breathe sighs of relief at their loss of political franchise and their concession of power to the administrators of information capitalism. The world is incomprehensible to them. But the myth isn’t.
The ascension of the machine god is progress. It’s enticing, inspirational and makes possible something akin to friendly AI demigods in a culture devastated by secularization. It allows our frustrated human longing of connection and real relation to at least virtually experience encountering a new being, both as creators and inferiors, the latter part of which opens the possibility to partially cede the heavy burden of existential and political responsibility (together with agency).
It’s the beautiful end of the West, the spectacular finale of the modern project.
You can find meanness in the least of creatures, but when God made man the devil was at his elbow. A creature that can do anything. Make a machine. And a machine to make the machine. And evil that can run itself a thousand years, no need to tend it. You believe that?
I dont know.
Believe that.(Cormac McCarthy, Blood Meridian, or the Evening Redness in the West)
Great stuff. Came here via the link you left in the bad cat's comments.
It strikes me that the question of whether a machine can possess agency, and therefore subjectivity, is intimately related to the question of where these qualities originate in the first place. I can't help but wonder if the relatively relaxed attitude towards numerical engines in Japanese science fiction is due to the animist assumptions from the Shintō underpinnings of their culture. In that context everything - the rivers, the mountains - is in some sense conscious, and this in itself is morally neutral. By contrast, the Christian assumption that the soul and body are separate things necessarily implies bodies without souls - therefore the mountains and rivers are lifeless. Should a machine become intelligent, it would then of necessity be possessed, either by an angel or a demon. Thus we tend to view AI with either hope or terror.
You wrote, "An artwork is necessarily a complex, relational reality which must be intentionally apprehended." I love speaking to a live audience (my job requires some public speaking, and I used to teach). But I hate speaking over a telephone for an interview. I need that complex, relational reality that you describe.
It makes the atomizing experience of Netflix, etc., that much more pernicious. It is an unexpected form of "divide and conquer."
Thanks for posting.