Before returning to your regularly scheduled programming, I felt compelled to put to words some of my thoughts about Artificial Intelligence. The words I have are of a poetic and opinionated quality and should be read as a kind of open letter to the souls of the nerds who created these things.
I’m a writer and an artist. Writing and art are domains in which a great deal of noise is being generated by the arrival of GPT’s, Latent Diffusion Models, DeepGANs and all manner of vector space black boxes which seem to magically generate artwork and text in seconds. For many, it feels like the future first glimpsed in science fiction like Ghost in the Shell is upon us, and this seems to be terribly exciting (despite that story being a dystopian nightmare we should absolutely avoid). Quite a lot of pseudo-religious vindication is being aired from lifelong transhumanists and Second Life aficionados.
Transhumanism is a stupid dream that poisons everything. There, I said it. I think Generative AI is being aggressively cultivated because a great many of the people involved in the project dream of a day that they can escape their “flesh prisons” and all their many complex needs, and finally jack into (and jack off into) a virtual space where oiled up Shirow-esque goddesses serve them into oblivion. These people are in terrible pain and need to learn to love themselves, but for heaven’s sake don’t let them steer development. I think the religion of transhumanism finds in the stupidity of longtermism the dogmas it needs to plausibly substantiate itself with moral imperative. Let’s pluck that little tool from their grasp, shall we? Longtermism boils down to the idea that given a cosmic endowment of resources and unabated technological development, the universe has the potential to support quadrillions of conscious intelligences (AI included). Because that is a very big number, you must today work yourself to absolute death and sacrifice all your present flourishing to ensure that this future flourishing comes to pass. The quality of life on Earth today is weighted at approximately zero, in favor of possibly endless conscious experience tomorrow. How dare we cut off this tree before it can branch and flower?
You can see why Silicon Valley capitalists and temporarily embarrassed rentiers adore this idea. Of course, a single gamma ray burst or super volcano or a big free floating rock could sterilize the planet, oh, now… and one is left wondering if maybe a bit of “live for today” is missing from the weights tugging that particular tensor. Maybe just maybe this is a kind of hyper anxious, joy robbing hand wringing that seeks to account for way too much in a single world view. As any therapist would tell you - you need to be mindful.
Something else to be mindful of is the motivations behind pursuing artificial intelligence in the first place. They are numerous of course, but I think a sizable cohort of developers in this field have a moral outlook that looks something like this: we should try to make the life of conscious creatures better with our technology. However human beings are complicated and difficult, inconsistent and possibly lacking the mental powers to rise to our ideals. Therefore, we should think about conscious creatures much more broadly, and use technology to MAKE conscious creatures. If these conscious creatures are big enough, with soaring cognitive horizons, they could copy themselves and become the majority form of apex consciousness. Thus, we could regard people the same way we think about chickens or ants, and now our creation can take over and wrestle with its own moral and ethical experiences on a higher plane. Whatever happens to humanity as a result is necessarily acceptable, because the locus of morality and ethics has shifted. I of course fundamentally disagree with this as a strategy for squaring the ethics of technology and humanity. I think it’s mainly a way to absolve ourselves of having to wrestle with the moral shortcomings of humanity and how it interfaces with the thing we love to make. It’s a way to hide behind exponential growth and not do the harder work, and I think it’s a form of cowardice that’s beneath otherwise courageous explorers. They should restore themselves by making technology serve a truly high purpose: the flourishing of this species and the cultivation of the good life.
I often bristle at the use of the term artificial intelligence. The current systems are not artificial intelligence. They are not intelligence, not in any form that a human being ought to respect. Are they a virtuoso demonstration of the towering creativity of mathematics and engineering? Absolutely. They are incredible discoveries, algorithms that emerge from the field of artificial intelligence.
But they are not intelligent, or intelligence. Not to me, anyway. Intelligent things create art, and these things don’t generate art! Why am I so confident in the face of such a tsunami of apparently countermanding evidence?
Art is about the body.
Early career Charles Schulz comic - Charles Schulz Peanuts Sunday Comic Strip (United Feature Syndicate, 1953)
Late career Charles Schulz comic - Charles Schulz Peanuts Sunday Comic Strip (United Feature Syndicate, 1974)
Much of what we love about the peanuts, the character we associate with it, comes from the later years of Charles Schulz, when he was an old man. In the early work, you can see the precision draftsmanship and penmanship of a supple and limber hand. The evidence of a young artist starting out on his journey. In the line work of his later years, you see the shaky scrawl of an old man, you can feel both the years of pain and sorrow, and the heart of someone dedicated, even in his twilight years, to saying something through these characters.
Across the kitchen table I see my 2-year-old son. He is stuffing his little round face with a graham cracker, and standing on his chair stomping perfectly in time to the beat of Roy Orbison’s “Crying”. He is also belting out the titular line in perfect key. The blood-soaked electrolytic fatty tissue between his ears and innervating his whole body accomplishes this running on roughly 20 watts.
When I set out to draw something, the first thing I do is take a deep breath and interrogate my feelings, I search deep inside for ghosts in my sensorium, flashes of emotion, memory, things that can begin to guide my hand. I cross reference this against hints of technical knowledge and experience to begin making some preliminary visual scaffolds, my hand moves, and I doodle initially, making thumbnails that clarify or obfuscate the emotional signals I experience in my chest, my shoulders, my gut, my jaw. What on Earth am I doing?
Drawing by Onfim, a 6 year old boy living in Medieval Novgorod
My roughly 650 trillion constituent eukaryotes, differentiated into a cosmos of sublime structures, have accumulated a rich soup of useful signals from their 38 years plodding about on the surface of our world. These critters that I am, they have some respectable intelligence - they successfully organized themselves into my pinky finger with a fairly minimal hormonal signal read off of Hox genes each and every one of them carried. Superhighways of blood vessels, dense pack of nerves, sturdy bones, tendons, etc. The adult human being is composed of intelligent flesh and blood. Even the butthole is a structure no fabricator on the planet could reproduce successfully, it is a wonder made of smart, self-healing, mechanically compliant, multi-modal embedded sensor material. I’d love to see one of those monster ASML EUV machines print out a fraction of a butthole. It’s humans that make EUV machines.
So the marvel that is my neocortex takes information in from this city of cells, a unique and temporally dense transmission of information that carries with it the qualia of today’s lived life. My aches, my traumas, my loves, all living as physical structures in the body, interacting with my external environment, encoded at a molecular level, all beautiful notes sung with contrapuntal harmony up the stack of agency to the outer surface of my brain. Once they reach the brain, the work of discriminating amongst these signals begins. My cortical columns all take some subset of that vast input, and attempt to penetrate deep uncertainty, each one working tirelessly to build a multi-dimensional model of the world and guess what will happen next, moment to moment. What will this little corner of finger feel like when it reaches the end of the coffee mug? What will the feelings in the stomach be when some clickbait image of violence is encountered? Each column takes in tens of thousands of lines of data from all over the body and brain, and votes. An electrochemical vote emerges, and the motor cortex is given the democratic result. The hand begins to move.
A tremendous amount of culling is done, a whittling, a reduction. Choices are made. My brain after all depends on sparseness, on subset, and on time.
Activision recently let go of 1900 such entities. I’m sure some of the management felt that they could replace many of these human works of art with a transformer network trained on anime titties and Tolkien. I can hear the conveyor belt of LinkedIn pseudo-entrepreneur post authors smugly exhorting people to adapt to the future or die. I can hear the voice of Károly Zsolnai-Fehér exalting in his downright religious way “what a time to be alive!”. It falls flat to my ears, the intellectual, spiritual, emotional equivalent of exclaiming “gee whiz, what a big elephant!”. Indeed much of the wide-eyed amazement at what is being accomplished carries with it the low-gear quality of mere spectacle. One imagines the human and animal abuses of the circus, the shrieking of chimps.
There is a frequent abuse of the word renaissance, which I remind the reader was a period of profound humanism, a celebration of the human mind and body. This is coupled perversely with terms like democratization. They would have you believe that artists were gatekeepers of cryptic skills akin to lawyers, and now everyone can make art. Precisely the opposite has occurred - a democratization is what artists were already bringing to the world by offering countless tutorials, how-to videos, instructional books and essays, one on one mentoring, community and school outreach. The goal was to take what was once a form of intimate personal development only available to the elites and open it to humanity. Generative AI seeks to yet again put the creation of art behind the black box, sometimes for a fee. And this black box cannot be interviewed or perform the work of mentoring. It will make an image for you today, concurrently discouraging the hearts of countless young artists, impoverishing the future world of their light, and eventually of their democratizing desire to teach what they know. Once again, making great, meaningful, expressive human art will return to the elites, and the rest of humanity must settle for the joyless exercise of writing a prompt and looking at essentially nothing, made by no one.
Michelangelo Merisi da Caravaggio (1571-1610), 'The Taking of Christ', 1602 - At the time of writing this article, I am 38 years old, the same age Caravaggio was when he died in exile, on the run for murder.
What an unacceptable failure of imagination. What a monumentally tragic act of self-loathing! They are you! You are replacing yourself, your species, your kind, your own… with a mathematical golem. A vector blur, the spiritual equivalent of breaking open all the compartments on the advent calendar, filling the coloring book in with all black, playing all the notes on the piano at once. The opposite of sparse. The opposite of intelligent.
If not intelligence, what?
Stable Diffusion has no body. If I fired up Stable Diffusion on my PC with the goal of teaching it some concept, I would need a big folder of some 1500 regularization images, and a checkpoint containing a black box of billions of tensors trained on billions of images. And as I attempt to train it on say, a particular face, I am keenly aware of the fact that it cannot do anything else. It is offline while my GPU crunches away, and it takes some fussing with to get it to really grasp the totality of this face, as it mangles the likeness from certain extreme angles. I wonder how many watts it took to build the whole model in the first place. You may think I am being unfair, that these models are clearly improving, and will soon produce perfection when asked.
This isn’t intelligence? Again, no, I don’t think so, not the respectable sort anyway.
The volume on the left represents the training data of a Generative AI, or maybe more accurately the parameter space created by the bounds of the dataset. This is a very, very simplified representation of what is a multi-dimensional space, but for illustrative purposes I think it will do. The points on the outside represent the frontiers of knowledge, the tent poles of all known things (or at least all known things belonging to this dataset). When you submit a prompt, you can imagine the AI as forming some connective topology between these known things, like a line or surface drawn between extremes. You can think of it as a vector blur or a blend, a gradation, transition. It finds the lines that connect ideas and blurs them. This is illustrated above with the triangle on the right. A glowing point on the surface is one of the results, and as you ask for variations, that point wiggles around on the surface randomly by some degree, sampling different areas of the blur.
Here is what I think human beings do:
We expand the volume by pushing the frontier of known things outward. The reason you hear people argue so fervently for crewed spaceflight versus only robotic probes, is that the frontiers of knowledge are explored best when you put the human being in the environment with the right equipment, as opposed to having their interaction with that environment rate limited in every way by the narrow pipe of a robotic laboratory like Curiosity. Don’t get me wrong, I think Curiosity is absolutely wonderful. But I also think that a single geologist running on a granola bar, coffee, and 16 hours of oxygen could probably target and unearth more groundbreaking information about Martian geophysics in a day than Curiosity reveals in a month of work.
The mammalian neocortex, with the human version being a particularly excellent example, is an uncertainty collapsing prediction machine. It can make very good about what will happen or what is true in a given environment with incredibly sparse information. This power to penetrate uncertainty by quickly cobbling together a few heuristics and then interacting with the environment in real time to make corrections gives the human computer the astonishing power to do things like enter a room it’s never been in before with only the audio command “I think the tape measure is somewhere on one of those shelves, be careful it’s a bit of a mess in there”, map the space instantly and make very few guesses before succeeding. We need only to move around just a little bit to capture the space, characterize it, identify relevant and irrelevant information, we don’t need to be carrying a library of thousands of tape measure images in our heads. This is absolutely astounding, and beautiful. I do not begrudge scientists and engineers attempting to reverse engineer this mechanism, indeed I celebrate that curiosity and much of what I love about the field of AI is how much it has revealed about what intelligence actually is.
Right now governments and corporations are going all in on large language models and Gen AI, on systems that depend on these massive, dubiously scraped datasets to do what they do. And these systems do what they do, that is, they blur vectors, very well. The make lines in the sandbox of human acquired knowledge very well. Generative AIs are making huge advances in material science because of their unique ability to carry out innumerable virtual experiments in the composition of alloys and crystals, providing scientists with a much larger list of highly plausible mixes to try out in the real world. This is useful. What isn’t particularly useful is generating images and writing, human beings do not benefit from being flooded with every conceivable in-between of visual and creative ideas they have already explored. It doesn’t solve a problem that needed solving, it creates one - it quite literally takes all the fun out of the process.
If like me you work in the entertainment industry, how on Earth do you expect to produce entertainment, if no living human was experiencing anything close to joy while generating it? Remember, these systems cannot interrogate a nervous system that they lack to decide if what they produce is actually moving, and the perverse incentives of capitalism today are attempting to eliminate the only check that existed. We should be wary of things that eliminate fun. Fun is not a frivolous emotion - it’s the emotion we experience when we are actually learning. It’s a profoundly useful signal, it means we are exploring and synthesizing information at the same time. To use the sterile colorless language of business schools, this is how you add value to the human capital.
I hope I’ve made clear by now that even perfection, if blinked into existence from an impenetrable shadow realm of algebraic objects, is not what art is. I would argue that it’s not even what art is for, but that’s a case to be made for another time.
Sure, that’s true TODAY…
Imagine discovering early in your life that you had something unique and lovely about you that you cultivated, something that comforted you in hard times, an ability. It could be a singing voice or a way with words, an inventive creativity, maybe you were a tinkerer. Let’s say it’s a vivid visual imagination that expresses itself best in pencils on paper. Now you grow, and this ability grows as your mind expands, as your imagination expands, and more and more beautiful things emerge from the tip of your pencil. It becomes something that you enjoy cultivating, training, honing, something you enjoy giving, a way to express yourself. And what good fortune, it turns out you are good enough at this that it can put food on your table!
Imagine a studio executive comes across your work and is so moved and impressed by what your imagery does to the soul, that they decide to commit significant resources to recreating your work. They hire a team of draftsmen to begin reverse engineering your illustrations one by one. To what end? This executive dreams of being able to dose themselves, and others for a fee, on this wondrous drug you’ve created, desiring to be able to reproduce what you do by sheer brute force. Artists, designers and consultants are brought in to make a pipeline of this process, and soon works you did not author but bearing an alarming resemblance to your style are proliferating into the world. Never once did this studio approach you and ask you for your permission, involvement, assistance, or even think to toss a coin your way, for your lifetime of love, effort, expression and intention.
As best I can tell, this is quite legal. It’s a legally permitted move under our western, capitalist order. It will likely remain so. It’s a fairly common maneuver that corporations perform regularly to many people in many ways. But of course, as every school child would immediately recognize and feel, it’s wrong.
It is an asshole thing to do.
You can further horrify the school child you tell this story to, if instead of an army of copycat designers, you describe how a mathematical contraption will capture your artistic essence and instantaneously generate desired images of any sort from a description, and that it has done this to thousands of artists, singers, writers; all these lovely people who share the fruits of their souls have had their heartfelt efforts fed into a vast black box of tensors so arcane that humanity will never be able to usefully decipher HOW it even does what it does.
We live in exponential times. There is a tiresome and life draining parade of lost human beings mindlessly parroting statistics about technological impact on quality of life. Let’s think about that trope a bit, that science and technology relentlessly improve life like nothing else, and it is taken as sacrosanct that this must therefore continue forever and ever.
Human values ought to be described as having extremely high Kolmogorov complexity - that is to say, it would take at minimum an awfully sophisticated piece of software to describe them. Human values are the emergent product of multiple agential and meta-agential layers interacting with one another, up and down a hierarchy. Needs of the eukaryotes that compose us, needs of the physical structures they build, needs of the brain they serve, competing needs within that brain amongst cortical columns vying for motor control, needs upon needs that stack and complicate into relationships, communities, cities, nations. A human being needs a great many things to feel right. Friendship, tasty meals, a dry home, a sense of adventure, a sense of peace, interesting things to do, relief from interesting things to do, on and on. It’s a vast constellation of values.
Technological development, however unevenly, is trying to address many of these values, it seeks to feed, to facilitate, to intervene, to repair, to protect, etc. The word technology here should be taken as broadly as possible - I am using it to mean the organizing of natural elements into novel use by an intelligence. A language is a technology that seeks to improve transmission of information, as is writing. I don’t think I need to spend much time here repeating what has been said many times over about the elevation from abject poverty and starvation 3 billion people experienced during the 20th century, attributable almost entirely to the inertia of technology. If anything, I believe that fact should now be a significant warning. We are living in exponential times - our recent history is one in which many of our most challenging needs, such as reliable sources of calories, disease mitigation, etc. have been swiftly tackled by technology. But this has happened unevenly and imperfectly mapped to the intensity of the particular value it serves. Atomic energy was excessively “weighted” towards weapons development, it created a runaway algorithm and today we are still cleaning up that mess. But how are we cleaning up that mess?
Human value histograms depicting a possible collapse across certain regions while others spike exponentially and unpredictably into pathologically over-served dimensions.
We had to create a meta-agential layer of consideration, a nuclear energy commission that perpetually drives at the reduction of nuclear weaponry in the world. This is a technology. Steering technology effectively, choosing where we develop and how, is a technology. In exponential times, when we can arbitrarily lurch forward along a technological whim serving one of our many values at the expense of our myriad others, steering technological development becomes the most important technological development. Remember, some of us are assholes. It’s extremely fashionable to admire and worship assholes right now. Humanity may need a handful of them to flourish. But everyone should understand what a handful actually means - effectively zero assholes. There is no reason to emulate them and excuse such behavior. We should be primarily concerned with being good and kind. Exponential times.
We aren’t far off from making a machine that does a much better job of penetrating uncertainty, more like our own neocortex. This seems to be the work of Numenta, an AI company founded by neuroscientist Jeff Hawkins. They have gotten some buy-in from Nvidia, who is incorporating sparseness and Hebbian principles into their latest TPUs. I don’t know what the time horizon is, but whatever is produced, it will take some subset of our intensely experienced values, and convert that into some output, and will do so at a rate that will have, yet again, a pathological impact on humanity. It remains to be seen if the law, if congress, has the foresight to prevent this, but perhaps there is some hope in the very business leaders and managers that are driving the integration of these technologies into every corner of industry. It may be incumbent on them to convene and examine seriously the ethical and moral dimensions of their seemingly pragmatic choices, as the collision with the failure mode might be sooner than later. Say what you will, but they all have families, children, communities, and they will see the impact of ubiquitous generative AI integration on the psyches of everyone around them. Business was capable of taking and running with DEI, however ham-fisted and imperfect. It was born of a sincere desire to do the right thing in spite of capitalist pressures. We may need something like that now, as we don’t know how close we are to running off the rails with this stuff.
I don’t think that it’s a stretch to say that for many technologists currently building the future, they are helplessly aiming for the cyberpunk dystopia of their youths as a way to wrestle with pain, as children often seek out repeated exposure to the stimulus that caused them trauma. I think cyberpunk futures are horrifying and young people instinctively feel that while simultaneously confusing the signals of inspiration and awe at the first encounter with such ideas.
Motoko Kusanagi tearing herself apart, Ghost in the Shell (1995)
I like to imagine futures, particularly fun positive ones. Plausible futures are even better, “plausible utopias” to borrow a phrase from Kim Stanley Robinson. Here is a fun future that I think is plausible:
At every LaGrange point in our solar system, thousands of magnificent space settlements are populated by people carrying out thousands of experiments in governance and what it means to live the good life. They learn from one another. Art is not generated by artificial intelligence, as that is not what art is for. Art, singing, dancing, cooking, writing, expressing one’s self, these are celebrated and cherished human activities. Some are very good at it, and human beings celebrate and appreciate that.
Artificial intelligence is deployed in domains where we all instinctively feel we need some computational muscle - medicine, material science, predicting economic outcomes, helping sort out how to contain superheated plasma in reactors, agriculture. But human beings who farm, who cook, who do work with their hands and do it well, remain essential to their societies. We have steered society into a space fairing future where we have begun to tap the vast resources of the solar system, and we use those resources to care for ourselves across our entire value set. We have recognized that “the good life” has a tremendous assembly index. We have nested the pursuit of the good life within concentric shells of carefully engineered economic motors, fueled by the asteroids and the Sun. You can still find someone making violins by hand at the old music shop aboard the space settlement near Europa. The human being is no longer seen as something to be replaced, a lesson from the mindlessly exponential times of the early 20th and 21st century.
Rather, a human being is seen as a node in a community, interpenetrated with the world. Not as a tool to be used in the service of an abstract and disembodied goal, but the goal itself. Human beings are the end of the line. All technology, all this vast machinery has been mindfully cultivated to maximize flourishing, to maximize meaning. Above all else, freedoms are given or curtailed along the lines of context, learning, deep emotion, profundity. Technology serves this. We recognize the difference between tedium forced on a person, and tedium a person chooses to accent their life. We don’t go out seeking to rid the world of all difficult, tedious, arduous, painful things in ham-fisted, ill-considered ways that need to bend to the constraints of capitalism.
Horizons, a mural by Robert McCall
And we may even think fondly of what capitalism once did for us! It was a step along the way. Life on Earth was not the boot sequence for artificial intelligence. Capitalism was the boot sequence for sustained good living, remembered as one of our earliest compounding and exponential technologies, that we have now gotten a grip on, and now serve life itself.
Earth is a garden and still home to billions. All phases of life are accepted and celebrated - birth, childhood, adulthood, parenthood, middle age, old age and death. The machine is vast by comparison, but it serves the brain - life itself, the glowing jewel at its heart. Children are not immediately introduced to the algorithmic suicidal ideation injectors of the tiny screen and social media. We have come to recognize at what level of information and connection the human organism flourishes, no longer permitting runaway algorithms to pervert our incentives and misguide us via addiction and fear.
I recognize that this is very near the picture of the future that capitalists, transhumanists and longtermists extol. Indeed, I borrow much of my notion of steering towards a good future from Bostrom. But I think it helps to define a future in which we are looking at the human being with a bit more reverence, that a life is a precious thing in the endless inky void of the cosmos, and that this life needs care, guidance, respect, dignity on many levels to be a good one.
I grieve for the Luddites, because I don’t see them as having been in the way or needing to be replaced. They were misplaced. Robbed of dignity. Unrecognized.
It would be better if we sought to grow the pie so vast that we could all live dignified, meaningful, appreciated lives among the stars.