The Machine Learns Lacan
Note: I wrote this a while ago and never published it, but the sudden and explosive rise of AI language models caused me to resurface it. Apologies — what follows is super pretentious. But it’s got a few pithy one-liners I rather like.
“People worry that computers will get too smart and take over the world, but the real problem is that they’re too stupid and they’ve already taken over the world.”
―Pedro Domingos
The Lacanian Thorn:
If there is to remain a gulf between the sciences of body and of mind, then we should be careful of the tendency within communities to strive for the inevitable synthesis. The drive to identification is too strong to be ignored, even as the artifice of man enables ever greater and greater means to precisely this end.
For reasons that I will assume to be self-evident, Psychoanalysis asks us to maintain a certain neutrality on the question of truth. As such, it is also important the we remain skeptical of the relationship between biological structure and subjective experience. It’s essential if we are to maintain the aforementioned gulf. Of course, such a project is absurd, we can only maintain any division for so long. And so it is through the medium of this absurdity that psychoanalysis asks us to innovate. And when a gauntlet is thrown, silence will fail if we are to maintain the sense of agency that the psychoanalytic movement demands. Consider how the modern neuro-scientist, through his advances in neural imaging, challenges us to consider the possibility that what we experience might not be ours to properly converse with; that something like the hysteric’s is the only possible position. How easy it would be to hook us up to machines in order to determine which areas of electrical intensity cause us to need our anguish.
And yet…
The scope of the challenge that such notions pose to the psychoanalytic project remain as profound as ever. Recall that Walter Freeman was a contemporary of Lacan; an extreme division that speaks to the progress we’ve made. Freeman, the man who hoped to make lobotomy mainstream, thought that people could be made better by treating the body instead of the thought. He left a legacy of wreckage and docility, literally mangling the minds of those entrusted to his care; the promise of an easy cure for the mundane anxieties of everyday life has long corrupted many a so-called doctor.
Fortunately for the future of the humanist project, the psychoanalytic approach demands we maintain a skeptical distance from such notions; the school is designed to look behind the thing being shown, and simply postulate any outcome as the correct orientation for desire. What is done is freely chosen from the associations generated by the patient. The analyst, whatever the state of their ego, must make themselves ultimately dispensable to this end.
That this system can work makes it especially imperative that we not deny the challenge being continuously raised by the sciences. The neuroscientist offers a new way of conceptualizing the relationship between appearance and psychic health. And like everything, rendering such mysteries visible requires an intermediary; a metatron that can speak of the unconscious thing hiding within the mazes of meat and impulse they purport to show. Its signals must be translated for us; digitized. Ours is a world increasing dominated by the virtual. Perhaps Lacan presaged this movement when he spoke of the Alethoshpere, perhaps not. In either case, I’m not sure anyone could have predicted the chaos that computers would introduce to the pursuit of truth.
Of course, the presence of what we like to term the unconscious means that any attempt to replicate and transmit risks becoming an exercise in projection, and the invention of the computer has led straight back to this all-to-human tendency. We ask the computer to replicate our phantasies; to render something visible and birth it into the universe of signifiers. To make our visions into something comprehensible and render the whole in the the function of the one.
The transition into the realm of signifiers is fraught. It invites a tendency, perhaps even something like a drive, to resolve the impossible gulf between them. Signifiers like to pile up — one explaining another explaining another — precisely because they aim to replicate themselves. The ballet of 1s and 0s that make this new study possible should be familiar, and yet the parallel to the psychoanalytic project remains elusive. Binary logic rules the world of the virtual — can it be said to rule the world of the mind as well?
Hegel suggested as much. The whole of dialectical materialism contrives to paint a picture of the binary nature of choice. Lacan’s divided subject too, trapped between signifiers, speaks to an infinitely reducible superposition that can only be thought of in binary terms. The study of history even suggests that what might be called human progress moves thusly. Our ingenuity is a byproduct of the hazy static of the unconsious; a consequence of the forced choice that existence introduces. It takes only a touch of extrapolation to see it in action through the history of scientific advancement, the transmission of technique, and the collected data surrounding it.
It is through the act of this transmission that belies the strangeness of “progress”: we learn through the interpretation of symbols and the imagination of order. The works of those who came from elsewhere or from “before” emerge, resolve, and dissolve into the mud of our own private acts. This is to say that, as individuals, our species learns the structure of the world not as a consequence of discovering knowledge, as though it could somehow be said to exist a priori for the subject; eg. the rock was there before we stubbed our toe on it. For the sake of experience this cannot be the case. Instead, we each encounter the rock as something altogether new, regardless of its state in the minds of others. This relationship is really the essence of slapstick: we only learn to be mindful of rocks through a progression of stubbed toes. This process is only funny in hindsight, of course.
This might be summarized thus: there is no collective knowledge, no shared feeling. The mind has no access to rocks or toes beyond what it has imagined. But it can imagine them, and this is what is essential about structure. How we pass through it shapes how we come to experience it; our unconscious knowledge of world. Whether we look for the rocks and find them at fault, or simply run through the tall grass and hope we miss them matters little. But once discovered, this knowledge, because it must be made to fall into language, carries the structure of grammar and shapes how signifiers interact with desire. When encountered, it must be worked into the impossibly complex networks alive in the mind of that which lives. Only then can it achieve some semblance for the subject, usually through the medium of language. When we stub our toes, we let the expletives fly — however we found the rock.
Positions within language could be said to emerge from the space of experience; language is alive in us; signifiers carry personal meanings to us — their identification is what the study of the unconscious ultimately promises. But there is nothing there — nothing that lies outside experience waiting to be discovered. Instead, experience assumes its own structure, signified through the marvelous fantasies of the divided subject. A simple trick of projection would suffice to explain to us then, why the the world might appear structured — it might just be us. The nodes that govern our choices are hidden ones, and they remain that way by slipping away into something else the moment that we test them against the real. As Lacan says, experience is bullshit: a subjective screen that only serves to tyrannize the subject of language into the game of civilization. In the face of such castrating conclusions, one might be tempted to ask, “why bother?”
But to ask the question at this level is almost without purpose. Nor is it, technically speaking, necessary. As a consequence of its being unnecessary, a subject can pass through all of life and never need to ask it, or anything else for that matter. And so psychoanalysis might be said to begin with the question about the question. In the absence of analysis, we take what is presented, and from there follow the signs to desire. Once analysis intervenes, we take what is located and suss out a map to desire, this time with feeling. As Lacan suggested repeatedly, analysis represents an opportunity to start again. And again and again…and again. Eventually, we’ll drop it and get on with things. But without this intervention, without properly “dropping it”, the world can appear like the outcome of some vast physics puzzle that our segmented brains have calculated an end to. An end we’ll dig out of the unconscious, or maybe just the iChing.
And really maddening thing, of course, is that we know this perfectly well. Outside of analysis, one might not ever need to locate the precise direction of one’s grain. The hysteric strives for the impossible: the satisfaction of a desire belonging to an other that is not properly there. No wonder the demand itself keeps slipping around, seemingly frustrated by the ones who really run things. The master is similarly castrated, but largely by the dependency they form with the hysteric — the one suffers under the desire of an other and who doesn’t need to know what they want. This is perhaps what the the avenue of the dream shows us: when structure belongs to the Other, our duty simply becomes one of “fitting in.” Surfers call it getting “slotted”, but there’s a word for it in every community.
This serves as the foundation of Freud’s science, the science of psychoanalysis and of the subjectively lived experience; the science of the drive (”why do I do anything?”). And because knowledge is structured like a grammar we can begin to sense that as a consequence, it is also structured like a religion. As is science. Her students truck in the meaning found through ritual arrangements of signifiers, and are as such eternally bound up in their gravity, no matter how wide they might seek to make their orbits.
This is a deeply human problem; an inherited one that wends its way into everything we take for granted. The reason for this is simple: each generation is forced to reinvent, again and again, the techniques of their ancestors. We do not come programmed with an inherent understanding of the world of signifying chains. This is why scientists today study actuary tables, both ancient and new, seizing on a million private visions in the hope of locating something common. Economists pour over social media feeds in a struggle for insight into mass behavior through what should really be regarded as performance art. That many of science’s greatest accomplishments should seeing arise out of nowhere is a consequence of errors arising from this attempt at perpetual transcription. The beliefs that emerge from this process are what underlies the drive towards consensus — that same one which eternally risks the better for the sake of the good; the less for the enough (or the more, for that matter). Belief is an iterative process, combining the creative and the formulaic in a way that can look a lot like progress.
And ours is an age when a great deal has suddenly started to look like progress.
Repetition — the progress of Mastery:
It is a cliché to declare the progress of human engineering and knowledge miraculous. Our seemingly numberless triumphs; a conga line of distractions, amusements, and displacements have reshaped both the surface of the earth and the bodies that inhabit it. The latest of these triumphs, and the culmination of much of our historic effort to date is the computer, and the digitization of experience through networks of fiberoptic dendrites and silicon chips. As these machines work to reflect ever more of our experience back at us, the world of psychoanalysis might do well to consider the simulations that they perform, as their increasing complexity has made the machine into more of a mirror of its creator than ever before. The Mechanical Turk, a famous chess-playing hoax, needed a man inside the box in order to affect a simulacra of life — Google’s AlphaZero has rendered professional chess players obsolete, even if it still requires assistance in moving its pieces. Moderns are teaching databases to think, by seeking for their solace in a sea of numbers.
This is no small feat.
But “why” isn’t really the thing here. It could be as simple as this: humans don’t really like to repeat themselves, and like to outsource repetition wherever they can. Why should it surprise anyone that the desire to automate that most challenging of undertakings, that of thinking, is driving our latest revolution? It shouldn’t. We learn what to build the same way we learn everything else, after all — by keeping at it until it seems right. The act of thinking is, in the mind, no different from digging holes and splitting stones; indeed it is the most repetitive of all human processes. So it should not surprise us that we have, through the medium of the computer, created a mechanical mirror that can take this on in our stead. Though you would be forgiven for feeling that parody is the best that anyone, machine or otherwise, can do.
Things have changed since we spoke of the computer in monosyllabic grunts (“bits” and “bytes”, “tubes” and “cards”). In those days, these hulking monstrosities lived mostly in basements and science fiction stories. Lacan doesn’t speak of them — for all his interest in algebra, he never considered how potent a tool that might make the calculator. But he can be forgiven. If you’re like most people, by the time most of us really noticed the computer in the waking world was about the same time that one wound up on our wrists. And that wasn’t all that impressive.
At first.
And perhaps these systems still haven’t reached the place where we can say for sure that what they do looks like thinking. However, this should not cause us to detour around humility. If we are to grant anything to the power of evolutionary theory, it should be this: there is a pathway (or maybe a chain) between multicellular machines and what we dare call human. While the creation of “category” might be a projection, “that which thinks”, that which allows consciousness, has long contented itself that its machines are essentially toys, as easy to tame as they are to unplug. The time for unplugging has passed though — these systems have become essential parts of our collective experience. Once they start thinking, it’s only a matter of time before they find a place within the structure of our discourses. Indeed, understood thus, we have already crossed this threshold.
The denseness of the global system we experience every day has required an equally profound evolution in software — the superego of the machine. And like a child, software systems have entered something like a mirror stage: today, that thing on your wrist may well be receiving its instructions from a cloud (quite literally) of signals exchanged between millions of remote processors billions of times per second. Millions of bodies and trillions of dollars have been reallocated to expanding its reach, its speed, and its capacity. Server farms have embraced the whole of the earth, bringing with them air conditioning and an enormous demand for electricity. Iceland has become the land of the future; even while Reykjavik may well be swallowed by the sea.
And these machines are dedicated to a single purpose — the automation of decision making. So-called machine learning systems are instructive here. On a basic level, a machine learning algorithm involves the creation of a neural network consisting of input nodes (the software objects that store sensory data) and layers of “hidden nodes” that perform whatever mathematical alchemy that is triggered through the input. Following some matrix multiplication, these hidden nodes take on new states, the exact cause and structure of which are still poorly understood. Suffice to say, if you pump this pseudo-random output through an activation function, you can eventually teach a vacuum how to navigate around a room. It will learn the shape of it’s environment, and something of it’s place within it. You can simulate the same thing with remarkable ease — creating imaginary walls that can function like a simulacra of repression.
There are different models one can follow as well. The guided learning model, rather like a parochial school, is the preferred method when the designer understands what output is expected — computers can learn to drive cars relatively quickly. Deep learning models are different, and are mostly used to discover patterns in data — identifying credit risk or learning how to advertise to specific individuals. These methods sift through vast groves of unconscious data and learn, slowly and tediously, what will eventually win clicks. It reminds one of a child learning to win the attention of its parents. Will our machines rediscover the edipal complex? And will they tell us, or simply play on the unconscious error that our our own self-ignorance enables? The dog already walks itself, the only question is if we’ll notice where we’re being led. GTP3 might make mistakes, but you need to be an expert to catch them.
For psychoanalysis, the challenge is clear. Today’s children won’t remember an age before the omnipresence of the computer, any more than they will remember an age before sexual liberation. What has been built had to be built up; to be built out. Most of us experience the computer through its incredible power to replicate and distribute words — text messages, Facebook posts, and increasingly bizarre spam emails. Pushing out content on a scale befitting our global civilization takes a lot of words; most of them the same. This seemingly endless repetition has given rise to a whole series of industries dedicated to automating away the secretaries and switchboard operators — in creating mechanical solutions to the work of transcription and distribution. Or what some have called “women’s work.” Remember: when computing began to go mainstream, the notional “ideal wife” shouldn’t have needed to work, so many a diligent engineer busied themselves by leaving their wives with nothing to do. Switchboard operators, typesetters, secretaries; those nodes in global information chains have all been pushed aside for a new kind of master.
From the perspective of social perpetuation though, creating these chains was a massive problem. And once we started to solve it, you’re met with a new one: having something to actually say — something to have transcribed and moved along the chain. And so another industry emerged to handle the relatively minor modifications that linguistics demand we embrace in order to seem authentic. And now, “Love, Kevin” can be automated as easily as touching the “L” key. These days many of your emails, and a good number of your news reports are likely the product of some fitful piece of automatic writing conjured from dreaming database. It took maybe a year after speech-to-text transcription become good enough to actually be useful until computers started making hair appointments.
Hello, Computer:
Of course, once machines start writing your love letters and deciding your beauty regimens, you still have the trouble of coming up with something fresh. But this is just another repetition — something that we are commanded to do by nature. Ritual has always sufficed for our ancestors, and in way, we have transferred ours on to our machines. Let them handle the management of sacrifice. Let your full-stack developer be your priest. Mathematical repetition, once considered the realm of philosopher kings, drove Turing to hook together his world-changing machine in the first place — cracking enigma was, above all else, tedious. The better irony is that it was the Nazi ritual of hieling Hitler that provided the law for the machine to obey.
So we now we have these hyper-advanced calculators that can talk, walk, and remind you about ride-sharing apps when you’re leaving the office. What we do, when we do it, and how we do has become a vast well of data, forgotten by us, yet filed away, someday to be miraculously rediscovered through some act or another (most likely a late-night bout of Googling — thank you Internet Archive).
No wonder 98% of the human genome is non-coding. It takes a lot of trash to make a person. Or a thinking machine
So bored programmers have channeled their libidos into the the construction of algorithms that can even do the learning about learning for them — they know that they want to know, so why not have the machine get to the bottom of it for you? Researchers in every field have heralded the arrival of these technologies. After all, any servant of knowledge knows that the more you can put under your eye — the more you can see of the world — the less work you have left to do in understanding of it. This is why pie charts feel more accessible than spreadsheets. They also obfuscate a great many assumptions and metaphorical emplacements.
It is, of course, absurd to suggest that a computer could ever be a thinking being, but they mirror us. They ape us, in much the same way we ape each other, and for precisely the same reason. Computers can serve the Other as assuredly as we do, save one crucial difference: the computer serves an other that is finite — the algorithm has been designed with an end in sight. Once it finds its mean, it is complete. For the machine, to understand (as psychoanalysis asks us to) that its apparent end does not actually belong to it is a the same as it knowing the space between a 0 and 1.
Consciousness, on the other hand, can only ever appear like a servant to the Other. The divided subject is a thing that moves between its signifiers. Unlike the computer, it is not mechanically bound to them — it cannot be. Indeed, it depends on this fact for its survival. No, consciousness is not a machine. It is the something else which speaks, and we know that itcan only ever dream about knowing. But it can dream of machines, and of being.
