The classic 1978 Kraftwerk song We Are the Robots was oddly popular in India when I was growing up. As a kid, I didn’t get whatever subversive message there was to the song, and I couldn’t quite make out any of the lyrics besides the titular refrain. I just unironically resonated with the headline sentiment and the stylized mechanistic musical arrangement, with processed vocals.
The opening Russian-English lines (the group was German, so I don’t know why the song is in Russian+English) go:
Я твой слуга [I am your servant] Я твой работник [I'm your employee]
We're charging our battery And now we're full of energy We are the robots We are the robots...
The song has an associated music video, featuring the band members playing their instruments with stereotypically robotic movements and flat affects:
The funny thing about human impressions of robots these days is that actual robots don’t move in stereotypically machinic ways anymore, or make stereotypically machinic sounds.
Advanced humanoid robots, such as those from Boston Dynamics (not the Tesla clanker1), move in much more fluid and organic ways, and can certainly dance human dances better than I can. Their sounds have a more complex musicality to them. And all machines, whether or not they are consciously designed in biomorphic ways, seem to be getting more idiomatically organic in how they are built and how they operate. They are even starting to exhibit emergent patterns of affect that feel organic. You can sense something like internal emotional states that rhyme with human ones like uncertainty, panic, and exhilaration for example. In modern robots, these are typically epiphenomena of functional behaviors rather than being designed in through some effort to engineer cosmetic “affective computing” features. A robot programmed with a sufficiently sophisticated exploration algorithm will look like it is exploring. It doesn’t need to be equipped with a designed “exploration face.”
What should we make of the actual robot dance? Futurama points the way.
There’s a really funny scene in the episode Fear of a Bot Planet (S1E6) where Bender the robot is befuddled by the humans doing the robot dance to help him celebrate “Robanukkah,” and goes, “hey, you guys are good! How the hell do you do that?” In the show, Bender himself aspires to be a country musician, and practices a rather human style of dancing that includes a move he calls “the Bender.”
The robot dance is just the newest in a long tradition of dances that draw inspiration from mechanisms. In pre-modern dance forms, for instance, we find performance styles inspired by puppets, such as the wayang wong of Southeast Asia. Many traditional Indian dances are essentially stylized interpolation sequences between sculptural poses, which often have associated religious symbolism. The temple sculptures you see depicting dancers aren’t pure representations; they document, in a somewhat prescriptive way, the stylized poses that form the scaffolding of the dance forms themselves. I suspect the dance and sculptural traditions co-evolved. In a documentary I watched long ago, a classical Indian dancer described the differences among styles as follows: Bharatanatyam uses straight-line movements between poses, Kathak uses circles, and Kuchupudi uses arcs. This is how roboticists talk. Or used to, in the pre-deep-learning era.
The robot, which Bender struggled to grok, is a dance from the 80s, and today is incorporated into the popping genre, which itself is derived from the the older (60s?) boogaloo genre, from which modern hip-hop dancing is also descended. Boogaloo, like classical Indian dances, is based on interpolation between poses, held statically for a beat or two. The distinct element appears to be the bit of exaggerated jitter or dangling exhibited when entering or exiting poses, which clearly reveals that observations of machinery starting and stopping abruptly, and underactuated parts swinging, provided the inspiration.
It is important to note, however, that whatever the particular technological era that inspires a dance style — sculpture, puppetry, monumental architecture (the modern dance pioneered by Isadora Duncan) or 20th century machinery — it says almost nothing about technology per se. But it does say a lot about the human attraction to the machinic. The underlying sentiment of the Kraftwerk video and song is the exception. Generally, our cultural performances seek to mimic the contemporary machinic because the aesthetics of machine states (poses) and movements appeals to us, not because they represent oppression and control.
Affinity, not alienation, defines the default evolving relationship between the organic and machinic. The machinic as contiguous with the organic, rather than representing a hostile Other separated from it by an unbridgeable gap. A mode available to the artist as much as to the engineer, for incorporation into experimentally expanded senses of being.
Machinic Stereotypes
Much of what we might call the machinic aesthetic in modern culture reflects the workings of a particular narrow class of machines: Those whose behaviors are governed by discrete state transitions between stable static states. The machinic aesthetic is a set of stereotypes because actual machines, increasingly, do not conform to them.
Sculpture is of course the simplest and mathematically most degenerate example. A traditional static sculpture or piece of statuary is a technological artifact whose governing state machine features a single state and no transitions. Puppetry, depending on the style, involves many more static states and a few dynamic ones (depictions of horseback fighting in Rajasthani puppetry, for example, involve wild pendular swinging). It’s less obvious in music, but the music of Kraftwerk clearly draws inspirations from the sounds of machinery. Many modern musical genres have a generally syncopated state-transition quality to them, while genres that have a nostalgic vibe tend to be smoother and more flowing. This is a distinct machinic evolutionary trend, not to be confused with the more familiar trend towards increasing dissonance.
The machinic aesthetic in the arts is inspired by machines, obviously, but it’s worth taking a closer look at the specifics of why machines behave like machines. Where do machinic stereotypes come from?
Industrial-age machines feature a lot of orthonormal movement axes (due to the limitations of rotary and piston parts, and machined contact surfaces), connecting states that are stabilized by mechanical stops, spring-loaded bumpers, and so on. Simpler control techniques and less capable actuators (like pneumatic) produce abrupt, jarring motion primitives, and force a separation of translational and rotational movements. More sophisticated control techniques and hardware allow for primitives that feature smooth accelerations and decelerations, and simultaneous translations and rotations.
Artistic aspiration can gallop ahead of available technological inspiration. In an interview about Robocop, the classic 80s movie, Peter Weller, who played the titular cyborg cop, revealed that he and his choreographer actually came up with a style of slinky, curvy motion primitives that the rigid robocop suit did not allow him to use. So sadly the character moves like an old-fashioned jarring-orthonormal-primitives machine because of technical limitations, not lack of artistic imagination. I’d really like to see a reboot with the original movement style they’d come up with. I bet we can do actual physical suits now (not just CGI) that would allow for the original artistic vision.
From the way Weller described the stillborn robocop motion grammar that never made it to the screen, the inspiration was likely the precise multi-input/multi-output servo-mechanisms controlled by digital controllers that began entering the industrial workplace in the late 70s. Machines using such mechanisms tend to make complex smooth humming sounds when they move, rather than clanking, whirring and thumping sounds, and can produce fairly arbitrary motion primitives (since they can move and turn in multiple dimensions at once). They also have less of a state-to-state pose-interpolation “locking” operating style, and look more like they’re flowing smoothly, dynamically, and not very repetitively, only pausing where they actually need to.
The takeaway from this particular example is that artists, like engineers, often move with, or even ahead of, machinic evolution. The people who are bound by obsolete machinic stereotypes are typically ones who neither work with machines, nor observe them closely for artistic inspiration. They have a primitivist urge to look away instead.
If you are paying attention, you’ll know that machines are rapidly evolving past the pose-to-pose transition/interpolation manner of operation that historically provided the inspiration for machine-inspired art. The flowing, organic quality we associate with “nature” and “naturalistic” arts is now better embodied by machines.
Machines that Flow
In modern mechanical engineering, any motion curve and velocity profile you can define mathematically in a computer program, you can now produce with a suitably expressive machine. The movement can also integrate feedback from sensors, and accommodate large variations in how work inputs present themselves. This creates an increasingly flowing, organic quality to how modern machines operate.
The challenge isn’t actuation of particular limbs and joints, but macro-stability of the whole, given everything you could “say” with a given set of motion primitives. The robotics lab I worked in as a postdoc at Cornell pioneered the use of precisely such complex motion primitives (computed using calculus of variations techniques that require modern computers) for robotic soccer contests. They dominated the contest for years because the other teams continued to use stereotypically robotic grid-based motion primitives, with decoupled translations and rotations. I suspect state-of-the-art wheeled soccer robots could already defeat human soccer players. Humanoid soccer robots will take a while longer. Within human-like constraints of speed and power, the human body is not actually the best machine for playing soccer, anymore than it is the best machine for playing chess, or cranking widgets on an industrial-era assembly line.
In industry, CNC machines (which are simpler than robots) had already far exceeded the limits of human dexterity by the 80s. Human machinists can typically only control one movement dimension with any precision. Operating a lathe, for instance, only requires you to move the bit precisely in the radial direction, and somewhat coarsely in the axial direction. Humans can only produce simpler geometries with a 2d milling machine. More complex geometries require computer control. Milling operations under computer control machining complex parts often have a flowing quality to them. They can be mesmerizing to watch.
In robotics, an emerging example of the rise of flowing, organic operating qualities can be found in drones (which are simpler than ground-based robots in many ways). Early drones, about 20 years ago, were being programmed using what are known as maneuver automata. This was done by having expert human pilots fly the drones around using remote controllers, and recording the maneuvers (such as barrel rolls) that transitioned between trim states (states like steady forward flight). These then supplied the grammar and vocabulary of motion that autopilots could use, modeled using sophisticated mathematical frameworks (automata theory, Lie algebras…). The limit again was math and computation, not mechanics. Drones programmed this way had a clear “pose-to-pose” quality to how they flew.
But more recently, drones are being programmed to fly in ways no humans could fly. And this isn’t just a matter of higher g-forces or tighter turning radii achievable with unpiloted vehicles. Not only can machines now produce motions that no humans can produce (directly with their bodies, or indirectly through pilot controls), no humans can even conceive of them. It takes machine learning to discover and grammatize higher-order motion primitives that exploit the full mobility envelope of a given machine. And the result is motion languages that are increasingly beyond the human brain to learn. Motion languages that flow in ways where humans can’t easily perceive the underlying boundaries between primitives, and governing grammars.
We are reaching the limits of the ability of the organic to conceive of, grok, or imitate the machinic, even symbolically or allegorically, through art. Yet, paradoxically, machines are looking more organic than ever.
Hyperorganicity
The evolutionary tendency of machines, given improving material and computational capabilities, is not towards perfection of static human notions of the machinic (artistically mimicked or conceptually modeled), but towards the hyperorganic.
By hyperorganic, I mean an evolutionary mode that drives increasing complexity along both machinic and organic dimensions. This gives us machines from qualitatively distinct evolutionary design regimes. Machines that exhibit organic idioms but defy comparisons with specific biological organisms, and also exhibit alien aspects that don’t fit organic idioms.
This is not as complicated a point as it might seem. The ball-and-socket joint is idiomatically organic (your hip joint is an example). The (macro-scale) wheel-and-axle is idiomatically machinic. A machine that features a ball-and-socket joint more complex than any natural example, and wheels, would count as hyperorganic. Certain Boston Dynamic robots are already there.
Karl Schroeder observed that any sufficiently advanced technology is indistinguishable from nature (a snowclone of Arthur C. Clarke’s line about magic). This is likely too weak a proposition now. Technology gets to parity with the organic, in terms of informational complexity, then begins to go past it to hyperorganic regimes. And art sometimes leads the evolution and points the way there. The anticipation of flowing motions by Robocop is one example. Another is in AI. I read some speculation that the depiction of alien language in Arrival was part of the inspiration of the transformers architecture underlying much of modern AI, which is ironic given author Ted Chiang’s turn against AI. I don’t quite buy the theory (technological origins are never that simple) but it’s hilarious and fun-to-believe origin story.
Today’s robots are much more flowing and organic than stereotypically machinic robots that whose movements resemble the robot dance. They will soon be hyperorganic as advances in soft materials, weird actuators, and AI-discovered motion languages continue. The hyperorganic future is arriving not just linguistically, but materially.
We tend to miss this dynamic because we identify most easily with technology that visibly has roughly the complexity of human self-models. We harbor contempt for technology that is simpler than we perceive ourselves to be (unless you’re a particular sort of mystic seeking to become one with a simple tool like a sword or hammer), and fear of technology that is more complex.
While this is most obvious with technologies that reproduce, automate, or surpass human functions and behaviors in legible ways (most obviously humanoid robots), this is actually true of all technology, including technologies that do things humans never did, like fly.
Consider airplanes. Most airplane enthusiasts resonate most strongly with airplanes from the era where the gross morphology of aircraft was human-like in complexity (World War II). Airplanes through World War I were too simple and contraptiony. They did not have the cohesion and humanish contraption factor (ratio of complexity to design integrity) to allow us to identify strongly with them. But at the other extreme, modern airplanes all look strangely similar, falling into far fewer geometric varieties than WW 2, and too cohesive and monolithic. For example, the “flying wing” design is hard to project a 4-limbed primate bodily identity onto, compared to say a WW2 fighter. WW2 aircraft represent an anthropomorphic identification sweet-spot.
One sign: It is far harder to individually identify modern aircraft, compared to WW2 aircraft, unless you really the subtle details to look for. Aerodynamically, they have converged to a few deceptively simple forms, beneath which a lot of invisible and non-human complexity and machinic design integrity lurks (I suspect this convergence is about to be reversed, by the way, by an explosion of new hyperorganic flyable forms).
The case of stealth airplanes is illuminating and worth a detour. The original stealth fighter, the F-117, has a weird, clunky appearance, full of strangely intersecting flat planes. Nobody I know finds it to be attractive. It looks like a contraption because it is. But not for the reasons you might think. The plane looks the way it does because computers in the 70s and 80s were not powerful enough to do complex ray-tracing to minimize radar profiles. So they solved for “low-poly” profiles. Later generations of stealth aircraft could rely both on better computers and better materials to engineer stealth features without sacrificing aerodynamics. So the airplanes have returned to the trajectory of increasingly smooth, aerodynamic profiles they were on. Once again, the limiting factor was computing, not mechanical imagination. The ugly Tesla cybertruck supposedly looks the way it does because of the difficulties in constructing automotive bodies out of stainless steel. But I suspect in that case, there is also a deliberately reactionary machinic aesthetic at play. On the plus side, it has hyperorganic maneuverability, thanks to the first automotive use of fly-by-wire type controls.
There is a sublimated sort of anthropomorphism at work in stories like this. A machine does not need to look obviously anthropomorphic for us to project our identities onto it. It does not even need obviously humanistic pareidolia-fodder elements, such as headlights that can be mapped to eyes. The necessary (and likely sufficient) condition is that it presents with a complexity comparable to human self-models in visual (or more generally, visual-auditory) modes. We trust and identify with machines that can be unbundled similarly to how we unbundle ourselves. This typically means about 7±2 visual lumps to the morphology for example (the so called “magic number), and operating principles summarizable at the level of (say), “id, ego, superego” theories of the psyche. Steam engines are easier to identify with than electric motors, which in turn are easier to identify with than gas turbines.
We might say that humans like machines that are informationally impedance matched to human self-models.
Hyperorganic Asymptotes
A thing — whether organic or machinic — can of course be vastly more complex than human self-models, both structurally and behaviorally. Things that are more structurally complex present as statically inscrutable even when they aren’t doing anything. A good example is the exposed plumbing and wiring in the cross-section of a rocket booster, such as that of the Saturn V that you can see at the Johnson Space Center in Houston. Another good example is the innards of a modern CPU. Or an aerial view of a modern metropolis.
Things that are more behaviorally complex present as chaotic even if they’re not. The rhythms of early industrial era factories were pleasing, even if high-tempo. You can vaguely sense the exciting orderliness of activities in progress. They were legible enough to sustain humor, such as the famous scene of Lucy working the accelerating chocolate production assembly line. But the rhythms of modern technologies, such as the view of evolving software presented by the evolving state of code repositories (there are several excellent visualizations) look chaotic. Even experienced programmers struggle to get attuned to them. Making widely accessible jokes about software projects is harder than making jokes about industrial factories.
We typically react to both static inscrutability and chaotic dynamics with anxiety, and in extreme cases, even with terror or horror. And the more hyperorganic technology gets, the worse it gets.
In a post a few months ago, I tagged one flavor of the horror — fear of oozification. Oozification — evolution towards gray goo that simultaneously seems like a regression to primordial ooze and progress towards alien intelligence — is one asymptote of evolution of the machinic towards the hyperorganic.
Another asymptote is towards monolithicity, which I wrote about in another post. Where ooze evokes a kind of primordial horror based on what is hyper-present, the monolithic can, under some conditions, inspire terror based on what is visibly absent (ht
for this distinction). Terrifying conditions obtain, I think, at the intersection of monolithicity and monumentality, which, as I argued a couple of weeks ago, is about proportions rather than scale. But when monumental proportions coincide with monolithicity and particular scales, you can evoke terror. In movies, this is often depicted through platonic-primitive imagery. The mysterious alien orb. The blocky robot in Interstellar. The smooth pyramid. The impossibly large cube.What can make this terrifying is that there are clear signs of great complexity somewhere, but none of it is visible. The scales being either epic, or worryingly small, establish a worrying distance from human scales, and signal either terrifying magnitude or density of latent powers.
A third asymptote of hyperorganicity is what we might call hypermodularity. When a technology presents as Lego-like composable blocks, but the visible grammar and vocabulary are clearly too simple to account for what is going on. If you were swarmed by a bunch of different levitating parts that had different shapes and coupled in complicated but comprehensible ways (think Iron Man’s suit), you might try to figure out the underlying assembly logic and hack it. But if you were swarmed by (say) a bunch of featureless levitating cubes that seem to self-assemble into weird alien geometries, hinting at advanced group theory beyond human mathematical capabilities, you’d possibly be terrified.
Humans are attracted to the machinic while it is evolving through the particular range of complexity that is our own range of self-models, with much of the visible complexity in idiomatically organic rather than machinic dimensions. We like mirror-like machines that offer us simplified, but still recognizable reflections of ourselves. With perhaps a bit of an imitable machinic aspect that we can incorporate into our own sense of ourselves.
Machinic Mirrors
Our simplistic self-models, of course, don’t exhaust our own actual complexity. As emergent meatbags comprising billions of cells, entire microbiomes within our bodies, and a range of engineering complexity that exceeds the span of semiconductors to space stations, of course, we aren’t accurately modeled or constrained by (say) id-ego-superego grade psyche models or Dasein-grade metaphysics, any more than our bodies are accurately modeled by stick-figure drawings.
Any such radically limited model of Being/mind/body can only be a limiting self-perception. Living in ways that feel true to such limiting self-perceptions is what we declare to be authentic. Technology that allows us to live in more complex ways without falling apart profanes not our literal biological natures — anything our bodies can do without irreversible damage or death is arguably legal in an engineering sense — but these limiting self-perceptions.
Over one lifetime, we don’t update or evolve our self models much. Over multiple generations, however, we do. So machines that provoke anxiety in older adults do not provoke such anxieties in younger adults who imprint on more complex self-models at formative stages of young adulthood. Douglas Adams’ 3 rules of technology capture this effect:
Anything that is in the world when you’re born is normal and ordinary and is just a natural part of the way the world works.
Anything that's invented between when you’re fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it.
Anything invented after you're thirty-five is against the natural order of things.
If you apply this idea alongside the tendency towards hyperorganicity in machines, you get a co-evolutionary model, where each successive generation gazes upon its reflection in a richer machinic mirror, and locks on to a more expressive organic identity.
Machinic mirrors drive our evolving human sense of being human.
Your self-models tend to form and ossify between 15-35, and because the machinic tends to evolve towards and past the organic, older people look more stereotypically machinic, and less human, than you perceive yourself to be. And younger people appear more dissolute, presenting both inscrutable structure and chaotic behavior. Co-evolving with technology means all generations are inclined to view older and younger generations as dehumanized. Every generation views itself as the peak of humanness. Only the youngest generation living generation is likely to be right.
Douglas Adams’ three laws require a caveat though — you only imprint on contemporary technology if the world appears to be broadly improving and technology appears to be contributing to that impression based on casual impressions. In other words, you have to like what you see in the machinic mirror when you’re ready to lock-in an identity.
Note that only casual impressions matter here. Arguments, sound or not, about whether or not technology is actually improving the world are moot. When contemporary technology appears to be contributing broadly to an apparent improvement of the world, more young people imprint on it and develop corresponding self-models. When the world appears to be in decline, and technology appears to be contributing to that decline, more young people are tempted into a reactionary turn. Often they turn more reactionary than living older generations. There are at least some signs that this is true of significant segments of Gen Z.
This can lead to weird retrograde periods in cultural evolution. Two British movies, both featuring Om Puri, portray this phenomenon. In East is East, he plays a conservative Muslim father dealing with a progressive son. In My Son the Fanatic, he plays a liberal Muslim father dealing with son who has turned to fundamentalism.
The evolution of human self-models is not necessarily monotonic in the medium term, even if in the long term it drives us towards an increasingly complex sense of our own humanity.
Nowhere is this clearer than in the most complex machines we build — organizations. Aging GenX and Boomer managers demand a return to 9-5 work, struggle to address low motivation, have no good ideas for dealing with “quiet quitting,” and declare meaning crises. It is effectively a demand that we put on primitive suits and return to an older ideal of stereotypically robotic work. And many in Gen Z seem eager to transform themselves into stereotypical robots and comply.
Meanwhile younger founders adopt more fluid org forms, cheerfully organize in seemingly inchoate structures and behaviors, and build little subcultures that seem remarkably free of the “meaning crisis” their reactionary peers and fearful parents clutch pearls about. They too are modeling themselves on robots, except that these are the emerging hyperorganic, swarming ones that don’t do the robot dance.
The very underwhelming Tesla humanoid “robot” (apparently still human controlled, so really more of an autonomous cyborg suit) was labeled a clanker derisively by critics, and the term clanker has since found notoriety as the name of an AI bot that allows you to launch memecoins on blockchains. Ironically, this clanker is actually a lot more organic in its behavior, though not in a humanoid way.
Subscribe to Contraptions
Janky, wobbly models of everything from philosophy to technology and everything in-between.
I kept thinking about motorcycles while reading this. If you've ever ridden one, when they're moving slowly you're very very aware of the 'old school' machinic nature of clumsy state to state transitions. Release the clutch too fast it stalls, kaput. Lose footing it falls over, kaput. Even backing up, adjusting the handlebars, moving forward, is like slow motion Boogaloo with the always present _weight_ of the thing that to this day makes me cautious.
The parking lot and low speed moves on a bike are where I feel the most aware of 'the machine' and it's seemingly inorganic peculiarities. I assume—for my own financial and physical safety—it is limited, mechanical, and only capable of very specific movements or it breaks or doesn't do what I want. Turning a bike around in a driveway is like doing the robot on a 2D plane.
"Speed stabilizes the vehicle."
Once you're moving though, talk about machines that flow. All of the things that felt jarring, careful, or clumsy now just flow. Once velocity removes the issue of balance you feel nimble and fused to the bike.
It being an "all 4 limb" experience helps I suppose. It is still humanistic but not like driving a car. It's less like it is made for you to get in and more to augment you—especially certain types. A bike can handle more of those motion profiles or velocity curves. You look at a 30mph flagged windy road and you want to hit it at 50, 60.
Now I am not saying they're anywhere near hyperorganic. The main design has barely changed in a century. It is just the experience of getting on one, getting through the low speed stuff—in my case to get out of stoplight territory—and finally flowing feels like a reminder of the direction of progress.
do you have any examples of motion grammars (that a human can understand) of "It takes machine learning to discover and grammatize higher-order motion primitives that exploit the full mobility envelope of a given machine."