We need to induce a temporary amnesia and create false memories in our Markov chains. Let’s erroneously rewire our neural networks to simulate exotic psychic anomalies and then wait for the weird sounds to flow out. If we are going to use genetic algorithms to evolve music, let’s mess with their DNA and breed mutant hybrid genres. We may even need to find a way of expressing a degree of doubt in our algorithms—something at odds with the mechanical determinism of Boolean Logic. And what would happen if artificial intelligence systems could access their own glitchy fallibilities and be programmed to embrace their quirks creatively? Rather than error correcting their bugs, perhaps these fallible systems could make use of them? What about the algorithmic equivalent of a clouded judgement—one that leads to a creative discovery? As it stands the coming musical singularity promises us no more than consumerist re-enactments of dead composers and deceased pop stars, benign AI musicbots and algo-curators, dull location-based musical concierges and mood-detected playlists—musical material as meaningful and interesting as artificially generated weather.
— From Cal Abyme’s journal, June 2029
Up until the early twenty-first century artificially intelligent music went nowhere. Stuck on a lonely and confused path, it obsessed endlessly over mimicking popular music from the past using weak AI and flimsy machine learning systems. Its success was judged only by its ability to imitate music made by humans; its ability to meticulously copy celebrated musical styles. Ironically the banality of AI music grew in direct proportion to the “cleverness” and “sophistication” of its algorithms. By 2018, connectionist networks running reinforcement learning algorithms were composing the some of dullest music in human history.
But by 2030, breakthroughs were beginning to be made, as Cal Abyme was busy applying non-Boolean logic to bit-shifted RNA on his DNA computer. No one was quite ready for the music; it sounded nothing like anything anyone had heard before, or was expecting to hear. Some listeners argued that the very definition of music needed to be changed due to the psychological impact of its weird emergent complexity; others said that its sonic properties might be used as a weapon of control, following the successful application of the psycho-perceptual physicality of its frequencies in mind entrainment experiments1. It was not long before Abyme’s programs began to proliferate throughout the web, spawning many new strange and fantastic genres of music—not just a few, but thousands of them.
Almost every journalistic account of artificial intelligence made during the dark ages of AI regurgitates the numbingly familiar story of a machine fooling humans into thinking that the music it created was composed by another human. But while his scientist peers held fast to Alan Turing’s famous thought experiment as an unparalleled model for how to validate the progress of their AI work, Cal Abyme, alone, knew otherwise. Suspicious from an early age, the young Cal was often heard to speak wryly of “Turing’s Trojan Horse”, and had stopped playing the imitation game in his own research a long time ago. It was Abyme’s conviction that Turing had embedded the Turing Test into the history of AI as a sneaky joke. But to what purpose? Was it a clever decoy, aimed to avert attention from the true potential of what creative AI could be? What had Turing been up to? The question vexed Cal, and he knew that the real answers in AI lay beyond Turing’s clever distraction.
The First Song
Abyme’s first song is a strangely haunted piece of music. Its weirdly-styled vocalisations layered over a beguiling tangle of detuned chord changes, sound like a Dadaist hallucination. Its mathematically twisted patchwork of samples mesh into a deranged and ghostly music: a soundtrack for a spectral droid, immune to the arrow of time, riding westward to Westworld2, an empty saloon bar haunted with cybernetic apparitions; the sound of a cranky melancholic old cyborg questioning it own ontology, a delusional ghost in the machine, with voices in its head and a sense of doubt.
Musicologists all agreed that the first song had confounded their expectations of what creative intelligence could be, precisely because its weirdness implied a sense of intentionality. Up until then, a perceived lack of real intentionality was one of the criticisms that had been levelled against artificial intelligence and machine creativity in general. In the late twentieth century, a philosopher called John Searl argued against the possibility of truly intelligent machines, no matter how intelligent they might seem, due to their inherent lack of intention. Searl, in his argument, revels in the fact that standard computers can only manipulate symbols (syntax) but cannot attach any meaning to them (semantics). For Searl, the difference between intentionality and the simulation of intentionality is consciousness (and creativity) itself. And “without consciousness”, retorts Searl, “a series of functional interactions remain empty and meaningless to the computer even when they look meaningful and creative to us”.
But intentionality was only a small part of the story. The non-human eccentricity of Cal Abyme’s music was one thing, but the strange effects it had on its listeners’ thoughts and behaviours was entirely something else.
June 21, 2031
The driverless hoverpod eased slowly towards the parking bay. Its chromium chassis was alive with scintillating, curved reflections of the lake. Cal often thought of these flashing diamond points as coded messengers; with a little effort he could flatten out the patterns in his mind’s eye and perceive them as a complex two-dimensional cellular automata, each glimmer the phase-state of a wave calculating its vectors through space. A computation!
Beeeeep. “Delivery for Cal Abyme, please accept?” asked the automated voice system, in a monotone voice. As the hoverpod eased into the parkbay to offload its cargo, Cal eased out of his daydream. A small packet flew from the hoverpod and through a delivery chute into the lab; inside it, a small glass phial containing a spec of biomass.
Entranced by the Rybo-Glyphs
Cal brushed aside the colloidal screen and gesture-initiated the volumetric display. The hovering holographic symbols whirled so fast that their traced forms created a visual music of their own. Cal checked the dancing rybo-glyphs, each one representing tens of thousands of parallel DNA computations. The volumetric refresh rate had trouble keeping pace with the programs’ mutation values so that the blur of its shifting r-glyphs began to develop a stroboscopic life of its own; the pulsing symbols conveyed their own asemic logic in the persistence of his vision, instructions transmitted from an alien dimension.
Cal began voxel-sculpting the r-glyphs into intricate configurations with minimal sweeping gestures. Each sweep of the volumetric would cause the r-glyphs to spin. The centrifugal force would generate a spherical harmonic to finely tune each r-glyph’s data set. The tactile feedback from this new generation of volumetric was so precise that Cal could actually feel radiating pockets of air fanning his fingers as the r-glyphs span. Sniggering to himself, he thought of the old days when computer users clumsily interacted with data using mice and keyboards. In those days, finding a “data-match” was like the proverbial blind man looking for his needle; no need for a haystack. Sometimes Cal would allow r-glyphs to roam freely in the larger pools. Algorithmic agents made secret liaisons with one another, and who could tell exactly what data was being shared between genes? Cal inspected the freshly-delivered glass phial with a eyeglass. It contained his entire life’s work. He estimated that the full-stop-sized piece of biomass contained 20 trillion zeribytes, about the same as all of the combined data generated on planet Earth between 1946 and 2024.
It was 6.23 p.m. and the Sun was falling into the rarefied sky and onto the horizon. It was the one moment that always stole Cal’s attention. The Earth’s convection currents made the Sun appear to melt with a dripping Gaussian blur. The melting disc was a sickly and luscious orange. Cal called this “apocalypso orange”, because it reminded him of nuclear detonation tests. Staring at the molten glow, he thought of the dying Sun diving towards its slow demise five billion years from now. And today—right now—apocalypso orange bounced off steel and glass, a thousand tiny dying stars enumerated in the windows of the hazy city’s skyline. A terrestrial constellation of his melancholy; a star map of his loneliness. His mind wandered to Zoe Aurora, and he heard a faint whisper of her voice in the back of his mind. He hadn’t heard a thing from her in nearly four months. He wondered what she might be doing right now. A large passenger plane banked in from the south on its way to land at the city airport. The jarring sound of its two engines’ differing frequencies created a third wavering drone. An Airbus 1400, he thought, recognising jets by the sound of their engines alone. He watched its fuselage recede into the distance until it had become a tiny black dot on the horizon. A full-stop to a fading day; and a full-stop to end his relationship with Zoe. Finally the tiny black point vanished into the darkening sky and the orange glow of the sun evaporated. He enjoyed testing the limits of his perception in this way.
The End of Specialisation: E-Golems
By 2029 the days of scientific specialisation had come to an end. The arrival of electronic-golems allowed scientists to study many hundreds of paths of inquiry simultaneously with sophisticated automation. An e-golem would be given a specific problem to solve. Multitudinous datasets could be run through in seconds, each with an e-golem in command. Cal sent off nearly twenty thousand e-golems in May 2028 alone, just to figure out the best case scenarios for False Memory Markov chains. E-Golems could even meet and compare their data. After picosecond-long debates they presented the best fit scenarios to the master e-golem, who would present a definitive report to their human master. E-golems took seconds to solve calculations that it had taken months to do on the old computer architectures. It was through an E-golem search that Cal found a way to bitshift DNA. From then on, he never looked at the old problems of AI ever again.
One of the trickiest parts of DNA programming was writing anonymous enzyme functions. But the abstractions of DNA bit-shifting reminded him of Brainfuck and this kept him in good stead. Brainfuck was one of the most obscure esoteric programming languages ever written. In 2015 Cal won a competition for creating a Turing-complete cellular automaton he called “Conway’s Wager” which he wrote entirely in Brainfuck.
Cal gesture activated the lab spotlights. They cast a stretched constructivist shadow across the far wall. The lab’s reductionist interior looked as if it had just jumped from the screen of a ray-traced architectural render (Ambient Occlusion switched on). Cal preferred blank pictureless walls. Pictures on walls would become familiar over time, and that familiarity rendered them invisible. So why hang them in the first place? The only visible colours in the room were a spectra of book spines neatly stacked in piles on the concrete floor and lining shelves. Placed on one pile of books was a geometric diagram of interlocking triangles, the Shri Yantra, a souvenir he brought back from India. “The Shri Yantra diagram is a Machine for Meditation”, he had read in a book on non-dualism. The bookshelves held an immense collection on mathematics, philosophy, cybernetics and many esoteric subjects. There were about a hundred on programming alone, including ancient tomes on computer vision, recurrent neural networks and agent-based systems. At the top of one pile was Speculations on the Fourth Dimension: Selected Writings of Charles H. Hinton. Its psychedelic cover design, a lattice of blue and red aberrating polygons—reminiscent of a Victor Vasarely painting—matched the Shri Yantra’s design.
Cal was wearing his black and white patterned jumper, a lattice of white triangles on black representing the famous Rule 110: Cellular Automata. Cleanly shaven, with sharp blue eyes and tight-cropped hair, his appearance leant more closely towards that of the new-age modern than the stereotypical scruffy, unkempt hacker. And he was looking good considering the tough time he’d had with the break-up; he’d resolved to keep his appearance sharp, as if to broadcast to the external world that everything was OK, even if it wasn’t. In fact, the emotional pressure of the break-up was actually driving him in his work: it was something to react against; it was like a feeling for revenge, not revenge against Zoe but revenge against his own feelings. The break-up created an edge, and it gave him a feeling of necessity that he hadn’t experienced in years. Zoe, gone from the apartment had left a vacuum for sure. But he could use that vacuum and fill it with his ideas and work; he could fill it with molecule-shaped numbers.
Cal’s Mind Wanders Through Computing History
Entranced by the flickering superimpositions on the volumetric, Cal’s mind began to drift again. It was if all the breakthroughs in computer history seemed to focus in on him at that point. Delusions unchecked, he imagined all those breakthroughs—from Babbage and Lovelace to Turing and Von Neumann—as a topological landscape in which this moment in time, his time, was a peak that cast an impenetrable shadow over all the others.
His mind traced the lineage: Babbage’s Analytical Engine, the brass steampunk deity of the industrial revolution that paved the way for acoustic delay line storage; a thousand digits stored in the time it took for electrical pulses to traverse a five-foot mercury tube. Mercury, the archetypal alchemical substrate, symbol of of Hermes the messenger, could not have been a more apt substance. And then sand, transformed into silicon to free the flow of information on the tiniest of scales; a billion calculations made on a single grain. The twentieth century had been the technological equivalent of the great Cambrian explosion, with an endlessly proliferating ecology of software and hardware. And now, with Cal’s Neuromorphic DNA System, a circle had been closed. Biological material had jumped from being a life-bearing data container to a data processor.
I believe that in about fifty years’ time it will be possible, to programme computers to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.
— Alan Turing, “Computing Machinery and Intelligence”, Mind (1950).
Turing’s paper had become part of popular folklore, almost mythical. Scientists had quoted it without ever reading it. While other programmers were invoking Turing the demigod as often as they invoked their recursive functions, Cal had been trying to decrypt its hidden secrets. Why had Turing embedded his trojan horse—the Turing Test—into the history of artificial intelligence? Cal knew it was meant as a riddle, a provocation, a zen koan, for someone in the future to decrypt; for the right person at the right time. And perhaps that person was Cal Abyme.
Entranced by the Computer’s Shifting Symbols
The patterns of light from the volumetric display refracted through Cal’s cornea, and some was reflected back off its inner surface. A thin film of iridescence, a form of optical interference—light calculating its own trajectory. The patterns he saw were eerily familiar, like scattered impressionist aftermaths of the reaction-diffusion equations he’d studied so many times before. They reminded him of the computational morphogenesis drawings that Turing made shortly before his death. The condensation of dots seemed to contain a secret language of their own, the facsimile of a self-describing Borgesian God-script3. If both sets of patterns—Cal’s and Turing’s—were layered on top of one another, their geometries would be in complete agreement. It was all starting to make much more sense.
DNA Computing and Neuromorphic Systems
Cal scanned the histogram for mutations, tell-tale signs that program functions had morphed into syntactically correct program statements. Each scan, taking milliseconds to compute, amounted to a compile duration of over 5000 years in standard evolutionary time. He was looking for a few mutants that survived twenty or more generations, so he ran a-survival-of-the-fittest heuristic on selected generations and pushed the best ones to the genetic repository. Most mutations were biochemical, fewer were caused by quantum perturbations and by far the least common type was effected by cosmic background radiation bombardment. This last route created some of the best mutations for extremely compressed and efficient program functions.
Cal thought of DNA as a ticker tape of different organic molecules, a miraculous microscopic Turing Machine, with digital storage and a processor combined. For billions of years it had been nature’s most efficient machine. The implications of read/write errors in classical computing were disastrous, but with DNA computers they were a gift, only making the programs more efficient through their evolution. Most living things—including humans—contain enough meaningless code to render any ordinary solid-state program non-compilable. But living DNA thrives on those errors, they increase the chance of random variation leading to increased adaptability. Cal knew that his software too would benefit from this throw of the genetic dice. Even better, the molecular structures of the DNA strands were structurally isomorphic to the new non-Boolean logic circuits he’d been developing. The building blocks of life already contained the right structures for the dense non-deterministic flows of logic he needed to run his Artificially Intelligent DNA programs.
Non-Boolean Logic/God Rubrik
There had been two major impasses in AI up until 2028. The first was the limitation of discrete state processing in the old classic silicon-based computers—however continuous they seemed, they could only process data one step at a time. But artificial intelligence really did need a brain, and the brain worked with multiple neurons firing continuously to create an emergence better known as consciousness. In 2027, biological computing superseded traditional silicon hardware. DNA architectures with real neural networks finally become commonplace. They had solved the problem of continuous flow with massively cellular processing. But there was still a second problem, and that was the shackles on data flow caused by antiquated Boolean logic. It was this problem that had wielded a maniacal grip on Cal’s thinking in recent months. He’d spent monkish nights, engulfed in silence, thinking about how to transcend the limits of logic. Only a new dawn would free him from the dense mental labyrinths of multiple shades of true and false. He wasn’t a stranger to machinations of logic. In his teens he’d made a unified God rubric of strict object-oriented equality. The grid-like matrix contained rows and columns that checked the equality between different kinds of esoteric datatypes. It looked like a half-finished game of Go: dots signified strict equality and empty gaps no equality. Cal was intrigued by its pattern; he always felt that the patterns of dots and spaces was a signifier of something greater.
“Error and unpredictability may be essential to our ability to think”, wrote George Boole. Cal thought the quote ironic, considering Boole had been the ironmonger of cast-iron logic.
Emergence of The First Song
The decision to opt out of the monthly hackathon had been provident. It would give Cal an opportunity to run the latest version of his system. He had always used computing as a way of avoiding people; not that he was socially inept or shy, far from it. Cal just preferred conversations with himself in the dream-time of computational disembodiment; floating in numerical space, alone.
There was a slight pause in the flicker rate of the volumetric—then a freeze—a crash? No, the RNA pre-processor had bailed out due to a dead-end heuristic search. “Absence of a signal should never be taken as a signal”. It was sometimes good to think like a machine and not like a fallible human, he thought.
The symbols appeared back on the screen, multiplying in thicker and faster proliferations. Then something happened—an unexpected glyph spin, a promiscuous genetic liaison? He then realised it had been a symbiogenetic instantiation with an entirely new mutation. A newly generated song was ready to be heard: “The First Song”.
As he began listening to its quivering frequencies, intelligible geometrical shapes began to form in his mind, like wakeful hypnagogic visions. The frequencies generated a visual music of spinning polytopes, hypercubes and contracting hyperplanes. “Geometric solids constructed from sound”, he thought. Where were these images coming from? The symbiogenesis algorithms had generated music, and now that music had the means to program the visual brain itself?
In fact, the symbiogenetic instantiation had caused an evolutionary mutation to be embedded into the music so that it became crammed with brain-hijacking zombie binaurals. All of our perceptual pathways are affected by what we hear, and now Cal had stumbled upon a way to mainline mutant frequencies into the auditory cortex so that it could commandeer all the other perceptual circuits in the brain at the same time. Cal recalled W.S. Burroughs quip that “language is a virus”; now his AI music could was the sonic equivalent. And if the first song could subliminally engineer the listener’s mood and behaviour by colonising the auditory cortex, could it perhaps be used to control people? And if this first song was so powerful, what about the next generation, and the generation after that? he thought. He stood frozen, envisaging endless self-devouring circuits of zombie symbiont music radiating low level binaurals through electronic networks.
Turing’s Trojan Coda
“And what of the Turing Test?” he thought, as an orange arc of sun appeared to slowly raytrace the eastern horizon. “It’s counterproductive—it only ends up telling us more about the way humans think than the ways machines do”, he mused. He then glanced at some old notes of his on Turing’s paper: “The only way by which one could be sure that a machine thinks would be to be the machine and to feel oneself thinking”. Paradoxical and intractable, he thought, imagining himself as the mind of his own DNA machine. But his own mind was a DNA machine! He was convinced Turing had wanted us to know that we could never really prove a machine could be considered intelligent, because the very concept of intelligence was riddled with endless self-reflexive contradictions. Turing had intended his test as a provocation to make us think about thinking. His paper itself was a machine for thinking itself.
If “any sufficiently advanced technology is indistinguishable from magic”, Cal thought, then any sufficiently advanced intelligence must be indistinguishable from magic too!
It was time to run another batch on the volumetric for a new song—.
(1) Mind entrainment (also brainwave or neural entrainment) describes the synchronization of brainwave frequencies with the periodicity of external auditory or visual stimuli such as binaural beats or stroboscopic light. Entrainment technologies (such as “mind machines”) have been used to induce states of relaxation, sleep, alertness or increased concentration.^
(2) Westworld (1973) was a dystopian science fiction feature film set in a futuristic Wild West themed amusement park with malfunctioning android cowboys. It was one of the first films to explore aspects of sentience and artificial intelligence in human-like droids. “We aren’t dealing with ordinary machines here. These are highly complicated pieces of equipment, almost as complicated as living organisms. In some cases, they’ve been designed by other computers. We don’t know exactly how they work”, explained one of the scientists in the film.^
(3)In the short story by Borges, “The God’s Script”, the main protagonist Tzinacán reads a divine ‘Script of God’ in the patterns of a jaguar’s fur. Worked on shortly before his death, Alan Turing’s early computational morphology theories were based on modelling the patterns of animal skins using mathematical systems. Turing meticulously mapped his reaction-diffusion algorithms on graph paper by hand (without computers), to simulate the morphological emergence of those trademark patterns. E.g.: https://www.brainpickings.org/2016/03/01/alan-turing-morphogenesis-diagrams/