Meanwhile, within the corpse the fluids heat, the soft bones tepefy,
And creatures fashioned wonderfully appear;
First void of limbs, but soon awhir with wings,
They swarm…

— Virgil, Georgics

The obscure tenth-century Arabic text Liber vaccae (The book of the cow) contains one of the first detailed accounts of the magical procedure for creating a synthetic human. [3] According to the book, the artificial generation of a “rational animal” requires the deposition of human sperm into the womb of a cow, which must be sealed inside a windowless building and fed on a diet of animal blood until it gives birth to an entity entirely devoid of skin. The newborn creature must then rest in the dark, submerged in a powder of sunstone, sulfur, magnetite, and zinc oxide to allow its human skin to grow. Enclosed in a glass vessel and locked away from sunlight, the being begins moving and feeding on its mother’s milk and blood. The book claims that after a year has passed, the entity, now entirely human-like in appearance and intelligence, will begin uttering secrets to its maker, revealing “all things that are absent.” The creation of such artificial beings, which came to be known as homunculi in the European Middle Ages, was one of the most ambitious goals of Western alchemy, alongside the transmutation of metals into gold. The most widely cited recipe for creating a homunculus is recorded in a sixteenth-century occult text titled De Natura Rerum (On the Nature of Things), commonly attributed to the German occultist Paracelsus, although this is still debated. According to this version, human sperm must be sealed in a glass vessel and incubated in horse dung until it acquires a human-like appearance. Fed on milk and blood, this being grows into an all-knowing creature capable of revealing hidden truths.

More than a systematic doctrine, Western alchemy was an eclectic set of theories and practices revolving around the manipulation of inorganic and organic matter. Despite being closer to magic than to modern science, the philosophical debates surrounding alchemical practices began to raise profound questions on the relationship between the natural and the artificial, questions which remain increasingly relevant in the face of today’s emerging technologies. In his book Promethean Ambitions, historian William R. Newman (2005) frames Western alchemical practices as attempts to materially replicate, not merely imitate, the intimate workings of nature. Alchemy’s ambition of transmuting metals into gold rested on the assumption that geological processes occurring naturally within the womb of the earth over enormous spans of time could be reproduced, accelerated, and perfected in the alchemical laboratory. Beginning in late antiquity, alchemists’ accounts point to the search for a radical continuity between natural and technical realities and a deep questioning of the metaphysical prominence of nature over artificial processes. Alchemy, Newman argues, proposed a paradigm of technics that diverged profoundly from that of the “mechanical tradition,” which included the fabrication of early automata. Whereas mechanical technology aimed “to conquer nature by imitating it with contrary materials,” alchemy “occupied a privileged rank among its believers in its claim to alter the deep structure of matter in a way that was purely natural” (Newman 2005, 23). The ontological continuity between natural and artificial to which alchemists aspired, in other words, rested on the possibility of understanding and harnessing the same material processes unfolding spontaneously in the natural world.

Its antiquity notwithstanding, the homunculus problem was already a problem of artificial intelligence. Although the spontaneous generation of animals from decaying organic matter was, since Aristotle, widely accepted as uncontroversial, replicating the same process with human life was more complicated. This was due to the necessity of generating a “rational soul” alongside an artificial human body—or, in modern terms, its consciousness. As such, homunculi were not just naive proto-scientific experiments but speculative devices testing the limits of human technology. Those who claimed to have successfully produced them argued that even the most mysterious natural qualities, such as intelligence, could be recreated, accelerated, and perfected artificially. This goal, however, could not be achieved through mere mechanical mimicry but required working alongside the material complexity of natural processes.

By considering alchemists’ fantastical accounts without quickly disregarding them as bizarre products of unenlightened times, we can begin to see that many of the questions springing from alchemical lore are still profoundly relevant to today’s technological landscape. Being both ensouled and enfleshed, natural and artificial, the homunculus asks: Can cognitive processes be replicated entirely artificially, or is technology bound to reiterate a faulty simulation of nature? And what role does matter play in the artificial generation of intelligent beings? These ancient questions, which still haunt today’s debates on AI and synthetic biology, lead us to the heart of the intersection between materiality, intelligence, and technology, where machines come to terms with the question of the substrate.

Biological computers and homunculi share a condition of double ambiguity: they confound the boundary between natural and artificial, and they complicate the demarcation between substrate and function. As their cognitive capacities arise directly from the material conditions of their embodiment, they both frame substrates not as passive carriers of function but as decisive agents in their own right, even capable of overcoming the constraints of their own nature. The ambiguous condition of these entities points to a deeper question: how is our understanding of the relation between organisms and machines shaped by our assumptions about substrates? In the domain of AI, this question becomes especially significant. To frame machine intelligence as either a flawed imitation of its natural model or the expression of a universal process abstracted from embodiment leaves little room for the power of substrates to emerge. Set against silicon chips, the hybrid materiality of biological computers throws the substrate/function binary into a profound crisis. To grasp the scope of this crisis, we first need to excavate the dominant paradigms through which the relation between organisms and machines has been conceptualized historically, paying particular attention to the place substrates have been assigned within them.

One productive way to approach the entanglement of machines and organisms is by investigating the use of metaphors across engineering and biology. Machine metaphors circulate widely in contemporary technoscience and are commonly understood as purely descriptive statements, for example, “the brain is a computer,” “DNA is software,” or “the organism is a machine.” As simple as they may appear, these statements implicitly express, through their use of “being,” a wide spectrum of possible relations, ranging from loose functional analogy (“organisms and machines share similar functions”) to strict ontological equivalence (“organisms and machines are unequivocally the same thing”). The ambiguity of such metaphors has been the subject of intense debate in the philosophy of science, where their divergence from transparent representations of biological and technical realities is often emphasized. [10] What relation is truly expressed by statements like “the brain is a computer,” and what role must substrates play for such statements to hold? Although machine metaphors seem to suggest proximity between natural and artificial beings, they often entail an implicit hierarchy in which either the machine or the organism serves as the archetype for the other. As we will see, this hierarchy affects not only the ontological status of artifacts but also the relationship between technologies and their substrates.

In 1747, Julien Offray de La Mettrie published L’homme machine (Man—Machine), a provocative essay in which he proposes that all physical and mental faculties of human beings can be explained as the product of mechanistic processes. [11] In his provocation, he followed René Descartes’s conception of the bête-machine, which argues that animal bodies can be described through the same mechanistic principles governing inanimate matter. While Descartes limited the scope of his mechanistic conception to explaining physiological processes, thus preserving the spiritual nature of human rationality, La Mettrie saw no distinction between mental and biological phenomena, believing both to be ruled by the same mechanical principles. [12] Both Descartes’s and later La Mettrie’s mechanistic understandings of biology were undoubtedly influenced by the mechanical technologies of their time: L’homme machine abounds with references to springs, clocks, and perpetual motion, while Descartes was known to cultivate a profound personal fascination for automata. [13] Reading La Mettrie’s work today is instructive on the historical situatedness of technological metaphors in biology, as it provides a clear example of how specific technocultural paradigms often shape broader philosophical understandings of life and intelligence. Although strictly bound to its historical context, La Mettrie’s vision of the man–machine also serves as a paradigmatic example of a broader, still widespread, epistemic stance that I define as technomorphism.

Technomorphism approaches the machine/organism binary from a peculiar angle. Instead of understanding technology as shaped based on natural processes (as per the traditional Aristotelian dictum, “art imitates nature”), technical reality is raised to the status of a universal archetype capable of encompassing both natural and artificial beings. From a technomorphic standpoint, the statement “the organism is a machine” implies that the category of “machine” is broad and powerful enough to serve as an archetype for the organism. Importantly, another implication immediately follows: that the category of “machine,” insofar as it encompasses both nuts and bolts and blood and flesh, points to a reality conceived as independent of embodiment. This “substrate indifference,” which erases any role for substrates in shaping natural or artificial realities, is a common trait of technomorphic statements. Consider, as an example, the following passages from La Mettrie’s work:

"Let’s look in more detail at these springs of the human machine. All the body’s movements—vital, animal, natural, and automatic are carried out by them. Aren’t all these mechanical? . . . So the soul is only a principle of motion, a tangible material part of the brain that we can safely consider as a mainspring of the whole machine, which visibly influences all the other springs and seems indeed to have been made first; in which case all the others are a mere by-product of it. . . . I am not mistaken; the human body is a clock, a huge and complex and finely designed clock. (La Mettrie [1748] 2017, 23–29)"

From the 1970s onward, computational metaphors in molecular biology began circulating widely, becoming so generally accepted that they made their way, largely unquestioned, into biology textbooks. Over time, several scholars have, knowingly or unknowingly, echoed La Mettrie’s confident technomorphic statements (“I am not mistaken; the human body is a clock, a huge and complex and finely designed clock”), revisiting them in computational terms. In doing so, they sustained a vision in which material substrates, whether artificial or natural, are framed as incidental to the technical functions they support. What do we really mean when we state that cells are computers, or that bodies are clocks? At their most literal, technomorphic statements enact a metaphysical flattening, where machines and organisms are related by strict ontological equivalence and substrates appear as marginal attributes of a higher, transcendent reality.

Among the most intriguing examples of technomorphic thinking in the information age are the writings of Efim Liberman, who, beginning in the mid-1970s, developed a theory of the cell as a “molecular computer” (Liberman 1979). Liberman, writing from the standpoint of Soviet cybernetics, argued that the cell (and, by extension, the brain) should be understood as a stochastic, parallel-successive molecular computer operating with “molecule-words” (DNA, RNA, proteins) and “molecular devices” (ligases, polymerases, ribosomes), whose interactions are governed by chemical programs inscribed in genetic sequences. In this stochastic computer, Brownian motion acts as a sort of “search engine,” binding complementary molecular addresses in a massively parallel computing substrate. For Liberman, the product of such massive computational effort within the cell is, essentially, the prediction of the future. “The molecular computer operates with word-molecules according to the programme, recorded in DNA,” he writes, “with the aim of predicting an outer situation in the next time-moment and selecting of a correct answer by synthesis of suitable proteins and other substances and also by macroscopic motion” (Liberman 1979, 111). While noting at multiple points the peculiarity of living cells vis-à-vis ordinary computational machines, Liberman offers a rigorous demonstration that “molecular cell computers” (which include both microscopic cells and macroscopic brains) are “universal computers,” entirely equivalent to any other universal computational machine irrespective of its embodiment or architecture. [14]

The notion of a “universal computer” serves as a powerful lever for several technomorphic positions whose content, while differing slightly in each specific elaboration, has remained substantially unchanged since the 1970s. Most recently, Blaise Agüera y Arcas (2024) has revived Liberman’s visionary ideas in his essay What Is Intelligence? He similarly proposes a reading of microbiology as fundamentally computational, evoking Alan Turing’s groundbreaking vision of a read–write machine that, given enough time and enough tape, could theoretically simulate any other computational machine. Demonstrating the possibility of such a computer, known today as a universal Turing machine, leads to the powerful conclusion “that computability is a universal concept, regardless of how it’s done” (Agüera y Arcas 2024). Agüera y Arcas leverages this universality to propose that living organisms are but one instance of a broader archetype, namely computation, which may occur indifferently across substrates and architectures. Much like Liberman, he also recognizes profound differences between cells (which he too defines as massively parallel and stochastic machines) and human-made computers, specifically focusing his comparison on Von Neumann architectures. [15] He also concludes that these differences, while making certain kinds of computational processes more efficient than others, are ontologically inconsequential, since “any computer can emulate any other one” (Agüera y Arcas 2024). Here, the relationship between machine and organism is not merely functional or discursive and far exceeds the boundaries of metaphor, as the author notes when he states that “it’s not a metaphor to call DNA a ‘program’—that is literally the case” (Agüera y Arcas 2024). In this radical technomorphism, natural processes become intelligible only insofar as they can be subsumed under a machine archetype whose validity transcends substrates, space, and time.

An in-depth review of the history of machine metaphors in biology lies beyond the scope of this paper. Targeting technomorphic statements as “reductionist” is also not my intention: such a critique would demand much deeper engagement with an extensive and ongoing debate in the philosophy of science. [16] Ultimately, technomorphic positions are legitimate attempts to break down the rigid separation between the natural and the artificial, which has historically cast machines as imperfect imitations of natural models. Faced with increasingly lifelike technologies, the importance of this challenge is unquestionable, and technomorphism has convincingly addressed it. The problem emerges in how technomorphic stances pursue this continuity, achieved at the cost of a structural erasure of the agency of substrates and disregarding their capacity to shape not only how functions are performed but also what kinds of functions are possible in the first place. If biological flesh and hard silicon are nothing but indifferent media supporting the same universal technical function, then constructing devices that hybridize living and nonliving substrates would ultimately make no difference to the ontology of our machines.

At stake in this debate, then, is the failure of our dominant paradigms to account for the increasing hybridization of machines and organisms on the one hand and the agency of substrates on the other. While technomorphism pursues continuity through substrate indifference, those who uphold the opposite hierarchy preserve discontinuity by essentializing the natural. Evoking once again the Aristotelian notion that “art imitates nature,” we may define their stance as biomimesis: the assumption that nature may never be fully replicated artificially and that technical reality is but a faulty imitation of natural processes whose true essence lies beyond technology’s grasp. Biomimesis remains a profoundly influential stance, especially in debates surrounding the possibility of artificial consciousness, where it often goes hand in hand with the notion of intelligence as “substrate-bound.” This is the idea that mental functions are a specific product of the brain and the brain alone, implying that no other substrate can produce comparable effects. Perhaps the most exemplary of such positions has been articulated by John Searle (1990, 32), who emphasizes the distinction between true intelligence and its artificial mimicry:

"Brains are specific biological organs, and their specific biochemical properties enable them to cause consciousness and other sorts of mental phenomena. Computer simulations of brain processes provide models of the formal aspects of these processes. But the simulation should not be confused with duplication. The computational model of mental processes is no more real than the computational model of any other natural phenomenon. One can imagine a computer simulation of the action of peptides in the hypothalamus that is accurate down to the last synapse. But equally one can imagine a computer simulation of the oxidation of hydrocarbons in a car engine or the action of digestive processes in a stomach when it is digesting pizza. And the simulation is no more the real thing in the case of the brain than it is in the case of the car or the stomach. Barring miracles, you could not run your car by doing a computer simulation of the oxidation of gasoline, and you could not digest pizza by running the program that simulates such digestion. It seems obvious that a simulation of cognition will similarly not produce the effects of the neurobiology of cognition."

Such positions remain anchored to a fixed idea of nature, life, and intelligence, preventing us from imagining any possible contamination between technical and biological realities. They are incapable of accounting for the ontology of an increasing number of hybrid beings, such as biological computers, in which organisms and machines exist in growing proximity. The technomorphic solution to this problem, however, affirms a new hierarchy as it dissolves the old. It is now no longer an essentialized concept of nature but one of technology—be it in the form of a seventeenth-century automaton, a nineteenth-century telegraph, or a twentieth-century computer—to which embodied realities of both organisms and artifacts are made to conform. Each in its own way, both biomimesis and technomorphism erase substrate agency: the first by declaring the natural substrate fixed and unchanging, the second by rendering all substrates interchangeable. Neither of these inherited paradigms can address the messy entanglements of machines and organisms, substrates and functions that arise from contemporary technology.

Alchemical sources insist that before the homunculus could talk, revealing its intelligence to its creator, the alchemist had to wait for its human skin to grow. The process of fabricating homunculi, as previously discussed, was akin to an accelerated and technically enhanced birth, but this acceleration could not entirely circumvent the time and meticulous care required for the cultivation of a biological substrate. Homunculi, as we have seen, could be produced only through specific combinations of mineral, animal, and human materials, following precise proportions, conditions, and durations dictated not by the alchemist but by materiality itself. While alchemy expressed the possibility of bridging the gap between natural and artificial realities, this possibility was always rooted in the intrinsic intelligence of substrates and their hypernatural capacity for contamination, transformation, and emergence. The question of the substrate is fundamental to any philosophical articulation of the boundary between natural and artificial, especially in the context of AI. While those who affirm a metaphysical supremacy of nature over technology often appeal to the idea of functions as being strictly substrate-bound (for example, “mind is a product of the brain, and the brain alone”), technomorphism insists on the indifference of substrates and the universality of functions (for example, “any computational machine is equivalent to any other”). What sets the world of alchemy apart from both these paradigms, making it especially relevant for outlining a new machine ontology rooted in an agential understanding of matter, is that its substrates are neither immutable nor indifferent: they are open to artificial manipulation while remaining powerful in their own right.

In the context of technology, its models, and its practices, any design approach presupposes, implicitly or explicitly, a substrate ontology: a cultural and operational vision of how natural and artificial are divided and of the role that materiality plays in negotiating that boundary. To decode this relationship, we may—borrowing and to some degree repurposing a term first proposed by Yuk Hui (2024)—adopt a perspective of technodiversity: rather than prescribing a single framework, different substrate ontologies emerge from specific material configurations and processes, speaking to the broader cosmology in which our technical artifacts are immersed. [21] The homunculus, which embodied a view of substrates as agential participants in both natural and technical processes, is far from the only myth centered around the possibility of artificially generating life. To depict a different substrate ontology, it is worth briefly turning to another, widely influential technocultural figure, in which matter is instead framed as a passive carrier of function: the golem.

The golem is one of the most enduring myths of artificial life in European culture, emerging from the Jewish mystical and philosophical canon. According to this tradition, humans can imitate divine creation not by biological or chemical processes, as alchemists would, but through symbolic and linguistic operations. As William R. Newman explains in Promethean Ambitions, “the major goal of the golem was to demonstrate the power of Hebrew verbal magic, the same power that allowed the world to be created ex nihilo” (Newman 2005, 235). The golem myth is embedded in early Jewish mysticism, especially the Sefer Yetsirah (Book of Formation), which represents the act of divine creation as a process of permutating the letters of the Hebrew alphabet. Accounts of the golem’s creation depict it as a human-like creature formed from clay or mud, imbued with life solely by the inscription of a magical word—emeth, the Hebrew word for truth—on its forehead. Such creatures, infamous for turning all too often against their makers, would revert into formless dirt as soon as the first letter of the magical word was erased from their bodies, leaving behind the inscription meth, the Hebrew word for death. Newman articulates the differences between the homunculus and the golem, and their respective technocultural legacies, in the following terms:

"The golem . . . inhabits a different thought world from that of our other homunculi. If it were not rash to draw a modern comparison, perhaps one could say that the golem belongs to the realm of “hard” artificial life, the world of robotics, cybernetics, and artificial intelligence, where ordinary biological processes are obviated or simulated by nonbiological means. The homunculus proper is a child of the “wet” world of in vitro fertilization, cloning, and genetic engineering, where biology is not circumvented but altered. (Newman 2005, 187)"

The lore of the golem points to a specific substrate ontology, largely shared by today’s computational technologies, where the embodiment of the artificial being is entirely dependent on, and ruled by, symbolic logic. The formless mud of which the golem’s body is composed stands in stark contrast to the lively, festering, protean material out of which homunculi were thought to emerge. Just as we framed the homunculus myth as a model for an agential substrate ontology, we may understand the golem as a forerunner of the notion of substrate indifference, a paradigm in which matter is understood as a passive support for the creation and transmission of a disembodied technical “truth.” The significance of the golem myth for cybernetics was not lost on the latter’s founder, Norbert Wiener, who named his book God and Golem, Inc. after this same cultural figure (Wiener 1990). In his book, Wiener engages with the golem figure to reflect on the relationship between technology and life, speculating on the capacity of cybernetic machines to learn and replicate autonomously. In a thought experiment highlighting the subordinate role of substrates in early cybernetic understandings of life, Wiener (1990, 36) went so far as to suggest “that it is conceptually possible for a human being to be sent over a telegraph line.” Although Wiener’s proposal may sound outlandish today (primarily because of the outdated communication infrastructure it refers to), the idea of such radical substrate indifference has persisted and evolved. One of the most interesting recent descendants of this philosophy is the OpenWorm project, which, since 2011, has been pursuing the goal of creating a perfect digital twin of the nematode Caenorhabditis elegans, modeling its body with single-cell precision. The project aims to reconstruct C. elegans’s entire physiology to produce a functional model capable of exactly replicating and predicting its behavior in a digital environment. [22]

We have already seen how technomorphic positions rest on two core assumptions: that functions are universal and that their material instantiations are largely interchangeable. In this view, technical models (for example, “computer” or “machine”) run the risk of becoming metaphysical archetypes, while the specific materiality of substrates is treated as secondary to the functions they support. Technomorphism assumes that substrates, while making certain operations easier or faster than others, are functionally inconsequential, since any computational machine is equivalent to any other. Biological computers, however, evoke a vastly different substrate ontology than their silicon counterparts, just like the homunculus and the golem provide a vastly different answer to the same philosophical question. To articulate why that is, let us consider two aspects of biological computers that complicate the relationship between computation and its substrates: (1) nano-intentionality and (2) onto-/phylogenetic diversity.

The term nano-intentionality is borrowed from neuroscientist W. Tecumseh Fitch, who proposed a materialistic account of cognition based on the embodied plasticity of biological matter (Fitch 2008). According to Fitch, the problem of understanding how brains produce subjective experiential states (a power often referred to with the term “intrinsic intentionality”) can be attributed to the capacity of all eukaryotic cells, from amoebas to the neurons in our brains, to perceive their surroundings and respond adaptively by modifying their morphology. Human-scale cognitive powers, therefore, are the result of an extensive stratification of nano-intentionality, which originated as a primordial evolutionary strategy in single-celled organisms. “An amoeba changes its structure, moving about ‘seeking’ nutrients, engulfing food particles, following chemical gradients in a purposive fashion,” Fitch (2008, 166) explains. “Such behaviour provides the prototypical example of what I’m calling nano-intentionality.” Fitch insists on this embodied plasticity as an attempt to identify the “missing link” between human cognition and silicon-based AI. While cells can adaptively modify their morphology in response to external circumstances, transistors are engineered as hard substrates and designed to remain morphologically unchanged. In questioning the cognitive equivalence of silicon and biological substrates, Fitch also advances a critique of technomorphic positions, arguing that while there is nothing mysteriously “special” about living organisms, there are profoundly consequential differences between machines as we know them and biological bodies.

"[T]he “cell = machine” truism obscures an equally important fact: that a eukaryotic cell is unlike any machine ever produced by humans. A cell has specific, causal powers, possessed by no currently available machine, and it is these powers I wish to bring into focus with the term “nano-intentionality.” Eukaryotic cells respond adaptively and independently to their environment, rearranging their molecules to suit their local conditions, based on past (individual and species) history. A transistor or a thermostat does not—nor do the most complex machines currently available. This is a practical difference between cells and machines (I can see no reason in principle that machines must lack such causal powers), but it is nonetheless a profound one. (Fitch 2008, 159)"

Biological computers such as DishBrain introduce an additional layer to this discussion, showing that nano-intentionality is not a prerogative of natural bodies, but that it is possible, in principle and in practice, to design “nano-intentional” devices. While silicon-based AI relies on its substrate remaining unchanged (and interfering as little as possible with its operations), SBI relies on the power of substrates to self-organize and form dynamic connections autonomously. In the case of DishBrain, researchers observed the emergence of functional connectivity among neurons even when they were spatially distant from one another, demonstrating that biological computers respond to environmental stimulation and training with nanoscale, substrate-level plasticity. An additional implication of nano-intentionality is that the plastic behavior of substrates, as Fitch notes, is “based on past (individual and species) history.” In the context of biological computers, this diversity is reflected in functional and behavioral differences based not only on training protocols but also on the different types of neural cultures employed. DishBrain was shown to perform differently depending on the cell culture’s origin, even when the electrical inputs and training procedure remained unchanged (Kagan et al. 2022). While mouse neurons were shown to be better at playing the game initially, human neurons appeared capable of faster learning, seemingly engaging in more “explorative” in-game movements. The system’s intelligent behavior, in other words, preserves an inherent contingency, since it depends on the specific physiology of the biological neural network, which in turn is determined by ontogenetic and phylogenetic factors unique to each cell culture.