How to Mass-Manufacture Humanoid Robots

What kind of materials, minerals, and supply chain would we need to create ‘Westworld’-style hosts or ‘Blade Runner’ replicants en masse?

Illustration: Sandro Rybak

“I just do eyes!” may not be the most iconic words spoken in Ridley Scott’s Blade Runner, but they gesture toward a whole series of worldbuilding questions that have long fascinated fans (read: me, maybe only me). For a movie about victims of interplanetary exploitation seeking justice and standing up to corporate and state oppression, the supply chains of Blade Runner are surprisingly underexplored. If genetic designer Hannibal Chew does, in fact, “just do eyes” in his work for the Tyrell Corporation, who does the other parts? Is there a person who only does bones or skin? (According to a 1997 Blade Runner video game, there is.) And who exactly are Chew’s suppliers of vitreous humor? (Assuming that’s the material Chew’s putting into replicant eyes and not some alternative polymer.) Is he a subcontractor or a full employee? How much of Earth’s economy serves the manufacture of the off-world’s army of replicants?

Now, in the third season premiere of Westworld, the mystery that surely gripped audiences for the two years between seasons, in which the hosts fled their carefully curated environment for the real world, will again be broached: How the hell will the hosts manage maintenance on their fragile bioengineered bodies in a brand-new biome without the benefits of a full repair shop? How will their immune systems handle the new environments? What if Dolores gets an STD? What if Bernard needs cortical fluid again? Come to think of it, what’s in cortical fluid?

Of course, Westworld follows in a long sci-fi tradition of humanoid robots exhibiting total indifference to maintenance and supply chains except as short-lived McGuffins. In science fiction, a lot of attention goes to the software side of creating sentient A.I. — it’s a topic that lets otherwise-vacuous narratives pretend they’re really asking Smart Questions about “what it means to be human.” Meanwhile, the hardware (and, in some cases, wetware) side of these creations gets hand-waved away into a solved manufacturing problem — as much an externality as the extractive labor conditions that made the screen you’re using right now to read this article.

As numerous journalists on the electronics supply chain beat have documented, those extractive conditions can have devastating consequences. Don’t the materials, manufacturing, and waste management concerns that go into any large-scale humanoid robot future deserve similar scrutiny? While mass-manufacture of humanoid robots may not be a current or even near-future development, given the tendency of tech companies to treat dystopian literature as a VC pitch deck, it’s reasonable to assume someone will try to pull it off. And the persistent appeal of the humanoid robot as a fictional trope in part relies on ignoring its massive resource requirements. Once those costs are laid bare, the less philosophically ponderous rationales for making humanoid robots become a lot harder to ignore. Extrapolating on the supply chains required for science fiction’s visions of a world of sentient humanoid robots offers a more complex perspective on A.I. ethics and the sheer impracticality of engineering cybernetic organisms in the first place.

One would hope they’d stick to using medical-grade silicone blends, but in a scenario with multiple competing robot vendors, we may not be so lucky.

Of course, there isn’t just one kind of sci-fi robot design. Fictional humanoid robots tend to run a spectrum from fully mechanical (Ava in Ex Machina, the Stepford Wives, the synthetics of the Alien franchise) to fully bioengineered (replicants in Blade Runner and, depending on how you want to classify him, the Creature in Frankenstein). Cybernetic hybrids made of organic tissue and mechanical/electronic components (Terminators, Westworld’s hosts, the “skinjob” Cylons of Battlestar Galactica) that sit somewhere in between the mechanical and the biological are also popular. There’s also a spectrum of manufacturing scales—many of pop fiction’s humanoid machines are relatively bespoke high-end products or prototypes, not necessarily mass-manufactured.

Starting with just looking at the mechanical/biological distinction, let’s take a look at some of the extractive and materials requirements, maintenance, and end-of-life-cycle needs of humanoid robots.

Photo: Ingrid Burrington

On the fully mechanical end of the spectrum, the materials required for making a humanoid robot aren’t all that different from what’s used to make humanoid robots today: circuits, armature, actuators, an energy source, and, at minimum, a sleeve of material resembling human skin to contain all the electronic components. Sometimes you’ll also get a wig of hair. Hanson Robotics, the creators of Sophia the robot, is perhaps the most well-known company working on building humanlike robots. Sophia’s skin suit is made of a patented material Hanson calls Frubber, but unfortunately the company would not respond to a request for an interview, so we can’t know, exactly, what it’s made from. We can, however, read Hanson’s patent for a “human emulation robot system,” which suggests Frubber may not be all that different from existing elastomers (elastic polymers) on the market. The patent cites products from casting materials companies like BJB and Smooth-On, and the text notes that “any suitable elastomer may be used for the skin, such as Thermoplastic, Silicone, Polyvinyl, Polyethylene, Polyurethane [or] Polyolefin.” While not all polymers are directly derived from petroleum, petrochemicals and natural gas tend to play some role in the creation of these compounds.

Since I couldn’t get my hands on any Frubber, I decided to look at other industries trying to replicate skin-feel. This is how I ended up ordering a skin-tone sample set from Real Doll, proprietors of one of the higher-end sex dolls on the market. (My verdict: definitely “fleshy” in terms of density and elasticity, but more on the Gak side of the texture spectrum than actual skin.) Although Hanson Robotics doesn’t directly cite the sex toy industry in its approach to making artificial skin, the company is likely indirectly indebted to a long line of innovations in fuckable products.

As Hallie Lieberman, author of Buzz: A Stimulating History of the Sex Toy, observed in a phone interview, “Anytime we come up with new materials, we have a sex toy that comes along soon after.” Materials science innovations, from Bakelite to silicone, tended to be rapidly seized upon by sex toy manufacturers that were attempting to create products that felt not only good but also “lifelike,” or at least were advertised as such. According to its FAQ section, Real Doll’s silicone blend includes “a harmless oil suspended between the molecules of silicone, which gives the flesh a more realistic feel and greater elasticity” (this may have produced the Gak-like stickiness of my sample set) and, given proper care and maintenance, can continue to be used for at least a decade.

Leaping to use the latest new castable materials also led to major health and safety issues in sex toy manufacturing, since the inventors of, say, polyvinyl chloride weren’t necessarily thinking about whether people could safely insert that material into their body. As Lieberman notes, “The sex toy industry isn’t really regulated the way medical devices are,” which means consumers can be at risk of exposure to toxic chemicals like phthalates if they’re not buying from a reputable vendor. Additionally, sex doll owners sometimes tend to do maintenance and repairs to the dolls themselves using consumer silicone mold-making materials, and they may not know to look for medical-grade material. If we’re assuming a future in which a monopoly corporation is selling humanoid robots, one would hope they’d stick to using medical-grade silicone blends, but in a scenario with multiple competing robot vendors, we may not be so lucky.

E-waste could be a less dire problem for robot hardware than it is for smartphones and AirPods.

Underneath the synthetic skin, hardware supply chains for the fictional cyborg likely wouldn’t differ all that much from those underlying today’s consumer electronics — materials like lithium, cobalt, graphene, neodymium, silicon, tantalum, copper, and aluminum would need to be mined, processed, and fabricated into robot components. Considering the approximate amount of ore that has to be mined just to make an iPhone, scaling up to a full humanoid form would likely require a lot more extraction. Assuming demand for electric vehicles, weapons, and personal electronic devices increases alongside the emerging robot economy, this could create a serious commodities challenge — not to mention some massive ecological destruction (assuming we’re not simultaneously weighing the materials costs of creating an asteroid mining industry, which offers a whole new set of supply chain challenges).

Depending on the design of the robot and the business model of robot manufacturers, e-waste could be a less dire problem for robot hardware than it is for smartphones and AirPods. Part of the challenge of recycling consumer electronics for smelting comes down to size and design choices, like making lithium batteries harder to remove. Larger components, like an endoskeleton, might therefore be easier to smelt.

Recycling looks a little more complicated for sex toys and dolls. Most of the polymers used in making sex toys aren’t biodegradable, though there is at least one vendor currently using biodegradable PLA. There also isn’t much vertical integration for taking products back and reusing or recycling components, leading to yet more plastic in the waste stream and, apparently, oceans.

So, before we even get into the computational and engineering labor required to make a sentient A.I., it should be noted that building a world of widespread mechanical humanoid robots would require tremendous amounts of fossil fuel extraction. There are complex polymers to mimic human skin and mineral extraction for components, and these robots may or may not bear some serious health risks depending on whether vendors are considering if people are going to try to fuck the robots. (Of course people will try to fuck the robots.)

But it’s not clear that switching out synthetic polymers of our Stepford Wives for synthetic biology of our replicants would make for a less resource-intensive android supply chain.

Neonatal foreskin is one of the most commonly used sources for generating human skin in the lab, but no one in my reporting seemed able to point me to the vendor who “just does foreskin.”

While the key materials of tissue engineering, like scaffolding, cells, and nutrient sources to help the cells grow, technically don’t need to be mined from the earth, they do have to come from somewhere. For example, when it comes to generating human skin—a practice currently used in burn grafting and for more niche lab applications, like cosmetics testing—you need collagen and fibroblast cells. Bovine collagen is typically extracted as part of the end process of livestock slaughter via a hydrolysis process used to turn cow hides into leather. Rat tail tendons, presumably obtained from deceased lab rats, are also a common collagen source. But if you’re looking for human sources of collagen, the supply chain gets a bit murkier. Neonatal foreskin is one of the most commonly used sources for generating human skin in the lab. No one in my reporting seemed able to point me to the vendor who “just does foreskin,” but most likely it involves recovering materials that would otherwise end up in medical waste streams.

We won’t need to have an endless supply of foreskins at the ready to build a bioengineered robot army: Cells replicate, and you can get a lot of engineered skin from one foreskin. (A commonly cited and nightmare-inducing estimate from a 1998 journal article is five to six football fields’ worth.) Israeli firm CollPlant has genetically engineered tobacco plants that produce human collagen, so maybe we could skip the foreskins altogether. It’s impractical to keep football field–sized blankets of human skin lying around in storage, and technology already exists for creating spray-on skin for burn victims, so in all likelihood the addition of a skin layer could be an on-demand final touch for our humanoid robots.

But the kind of tissue generated for grafts and research is a really superficial layer of what constitutes “skin.” “Full-thickness skin is a fairly complex organ that needs to be layered in a very specific way,” says Oron Catts, director of SymbioticA, a lab at the University of Western Australia dedicated to artistic explorations of life sciences. With collaborator Ionat Zurr, Catts has been using tissue engineering as an artistic medium for more than 20 years, which has given him more than enough time to become skeptical of some of the visions of fully bioengineered living forms as a more environmentally friendly or efficient approach.

“When engineering tissue, you’re essentially outsourcing the body of the organism to technology,” Catts says. “You need to predigest all of the nutrients. You need to protect it from infection. You need to do all of the things that a body does by itself.” All these activities have their own energy and resource costs, and the hardware and equipment used in that process is as much a product of extractive industries as our mechanical humanoid robots.

On top of the lab costs, there’s the work that goes into making it possible to take your cybernetic organism out of the lab. “If you bring a completely manufactured living system into this world, it’s going to deal with all the living systems we take for granted, like bacteria and parasites,” says Jeremiah Johnston, programs director at cellular agriculture nonprofit New Harvest. Westworld briefly acknowledged this reality in its first season with a reference to host Maeve getting an MRSA infection, but host immunology hasn’t really come up since in the series. While Johnston speculates that there are ways of constructing safeguards for your bioengineered robots — some careful engineering of major histocompatibility complex proteins, borrowing from fungus or soil bacteria biology and making cells that basically “pump out antibiotics,” perhaps — there’s so much about biomes that we don’t understand and can’t control for once we step outside of a lab. In Johnston’s words, “Biology works because you create more disorder than the order you’re trying to create.” The bioengineered robot vendor, in other words, is going to have a hell of a time with quality assurance and maintenance.

While the waste streams of bioengineered robots might not hold toxic minerals, there are still toxic components to be reckoned with.

“Waste is essentially unavoidable” in living systems, as Johnston points out. Whether that waste looks like the watered-down urea produced in human waste or is engineered into a slightly less toxic form by borrowing from other species (do Westworld hosts actually produce birdlike white shit? We’ve never seen them pooping!), there’s always going to be some waste stream to reckon with. And as far as recycling components goes, there are possibilities of using some bioengineered components as scaffolding for new tissue, but a circular economy approach to cybernetic organisms is unlikely. “The only form of recycling is compost,” Catts jokes.

If you trace the supply chain requirements and hypothetical life cycles of humanoid robots, it begins to make sense why a lot of the examples from popular science fiction tend to be highly niche, high-end models: Actually doing it at scale would be incredibly resource intensive and expensive.

They’re also impractical for a lot of industrial applications. A warehouse robot that lifts and moves pallets doesn’t need convincing skin so much as it needs to be good at perceiving space and have at least one very strong lifting arm. And as the wide world of sex toy design demonstrates, anthropomorphism in devices isn’t a fundamental utility.

Janet Lieberman Lu, CTO and CPO of distinctly not-anthropomorphic sex toy company Dame Products, wrote via email when asked about utility and anthropomorphism in sex toys, “Penises are great, but they’re overconstrained in their design.” Rather than produce a simulacra for sexual pleasure, Dame’s products try to, in Lieberman Lu’s words, “enhance what the human body can already do.” And while Lieberman Lu discouraged stigmatizing sex doll users, she also expressed skepticism of sci-fi narratives around sex robots. “No one should be worried that 20 or 50 years from now, they’ll have replaced all human sexual interaction,” she says. “If that were true, they’d already be more popular.”

Hallie Lieberman, the sex toy historian, noted that within both sci-fi narratives and the niche world of sex dolls, the goal of “can we create the perfect person” tends to actually translate to “can we create an artificial thing that (usually) looks like a woman,” and perfection typically has more to do with control than creativity. Despite all the rhetorical blather about consciousness and selfhood, ultimately most narratives around building a humanoid robot aren’t actually about making a machine — a thing — that’s exactly like a person; they’re about responding to a frustration that people can’t be — and can’t be treated — more like things. And while the A.I. supply chains of real life may lack in synthetic skin and immune system design challenges, the goal of making human beings into disposable and easily controlled components arguably persists.

When it comes to forecasting the future of A.I., most experts today go to great eye-rolling pains to emphasize it won’t look like the Terminator or any of the humanoid robots of cinema. Based on the resource needs of those futures, they’re probably correct — filling the world with robot-humans would have a massive environmental impact and limited returns on investment. Then again, impracticality and environmental damage across the supply chain hasn’t stopped a lot of technological development. There are plenty of people who “just buy cobalt,” “just train the model,” or “just manage the data centers” that consume vast amounts of energy, reproduce structural biases, and further alienate and displace human labor. It wouldn’t be a stretch for the market to offer a niche for someone who just does eyes.

Update: A previous version of this article misstated Janet Lieberman Lu’s first name. It has been corrected.

Precarious author (Networks of New York), educator, and artist (currently, Pioneer Works).

Get the Medium app