Nature Does Not Clone

ai blog
July 12, 2026

Joel Kowalewski, PhD

The physicist's question

In February 1943, in a lecture hall at Trinity College Dublin, a physicist exiled from Austria stood up to answer a question that was not, strictly, one he was accustomed to asking. Erwin Schrödinger—one of the founders of quantum mechanics—had titled his public lectures “What Is Life?” His question was concrete: what must the hereditary substance physically be in order to carry the entire pattern of an organism faithfully across generations? His answer, published the next year as a small book, sent a generation of physicists into biology. Schrödinger’s once imaginative idea that the substrate of heredity was an aperiodic crystal, an orderly but non-repeating molecular structure bearing what he called a code-script, was a surprisingly close description to the actual structure Watson, Crick, and Franklin would discover in 1953. The gene, in other words, was not a message about matter. It was matter that was also a message. Schrödinger welded two things together that we have been trying to pry apart ever since; most recently, in AI research. The attractiveness of a reality that is at its core mathematical symbols floating in space that animate machines of various types would leave countless scientists as well as the general public reaching for machine metaphors inspired by industrialization.

An engineering blueprint is information separated from materials or substrates. It specifies relationships with soft constraints on materials. The relationship, for instance, between arbitrary parts A and B in a blueprint might require a semi-rigid solid rather than a fluid. Engineers are therefore often left with multiple options to encode the relations or information in the blueprint, and this suggests that the blueprint is substrate independent. If a mind is like a blueprint it could, in principle, be abstracted into pure information and reconstituted using different substrates—uploaded to silicon, copied, backed up, resumed after death. It is the quiet premise beneath the dream of building minds, beneath the fear of a machine that supersedes us, and beneath the sales pitch of digital immortality. It seems so intuitive that it is almost inevitable, and yet the distance between the dream and reality is vast in practice. Our brains simply lack a built-in “plausibility” or “reality” detector. It is bootstrapped, built up from individual experience, shared communally, and refined. This is why our core disagreements seem to go nowhere. We are not accustomed to debating reality itself. Because our experience is predominately shaped by industrial supply chains, people who lack diverse theoretical training rely heavily on machine metaphors to shape their definitions of what is real and plausible. As both a computational neuroscientist immersed in abstract math and theory and an experimentalist working with stem cells and genetic engineering throughout my career, I can state the biological perspective on this.

There is no true cloning in nature. No blueprint to read from and translate. Information transfer between physical systems is always lossy; even ordinary cell division does not hand down two identical cells. A biological blueprint is not separable from the material that composes it—matter and information coexist, interact, and change together—and the sentence “I will copy this living thing” does not, when scrutinized rigorously, describe a possible operation in the natural world to date. If that is a general biological rule, and it is hard to deny it credibility, whatever brains do (that is, their functional properties) cannot be separated from structure. Even if we replace hard substrate invariance with a softer version, in which the materials now matter, the task of finding materials that fully emulate the organization of brains likely points at something so brain-like in composition that it simply is a brain with neurons—which is consistent with the position that nature does not clone because cloning is a useless operation. Variability or noise, which is to say messy copying, is what keeps a lineage aligned with changing conditions. This runs contrary to the principles of a manufacturing supply chain, where the product must be compositionally cheap and highly reproducible. Whatever imperfections remain in any two industrial replicas persist, almost by design, to ensure failure rather than sustainability over time, in the way living systems endure. That is what necessitates version 2.0. Cloning, as we typically encounter it, is seen in the mass production of goods to satisfy economic objectives.

My aim is not to critique beliefs about AI, only to assert that mind is magic in the end, or that living matter hides a vital essence that physics cannot reach. The argument runs the other way. Biology, and the properties of biological systems such as “being alive,” “being intelligent,” or “being sentient,” are tied to the physical composition and organization of biological substrates. I leave some chance that there are, in theory, non-biological substrates that could carry these organizational properties—and this essay is an account of the historical and future difficulties in pursuit of our nature as thinking, sentient beings at the edge of technological advancements that control us and the biology that draws us towards chaos.

Where the blueprint met the molecule

Schrödinger’s code-script needed an actual molecule, and for decades the smart money was on the wrong one. The chemist Phoebus Levene had argued that DNA was a monotonous four-unit polymer—his tetranucleotide hypothesis, which held the four bases in dull, equal proportion—far too simple, it seemed, to encode the baroque complexity of an organism. Proteins, with their twenty amino acids, looked like the only alphabet rich enough for the job. Then in 1944, the same year as Schrödinger’s lectures, Oswald Avery, Colin MacLeod, and Maclyn McCarty showed quietly and decisively that the “transforming principle” that carried heredity between pneumococcal bacteria was DNA, not protein. The blueprint had been located in the molecule.

The timing mattered, because the molecule arrived just as a new vocabulary was sweeping the sciences. Claude Shannon had given ‘information’ a rigorous, quantitative meaning in 1948, and molecular biologists seized it with both hands. Heredity became a coding problem: how could a sequence of four bases specify a sequence of twenty amino acids? The physicist George Gamow proposed an actual cipher—his overlapping “diamond code”—and in 1954 founded the RNA Tie Club, twenty members for the twenty amino acids, to break heredity as one would break an enemy’s cryptogram. Francis Crick’s ‘sequence hypothesis’ and his 1958 ‘central dogma’ recast molecular biology as the one-way flow of information from nucleic acid to protein, where information meant, precisely, the specification of a sequence. Henry Quastler was convening symposia on information theory in biology; the words that still organize the field—transcription, translation, messenger, reading frame—were borrowed wholesale from the engineering of communication. The historian Lily Kay would later describe this as the moment biology set out to write ‘the book of life.’ And a book is separable from its paper. The information-theoretic framing did not merely describe the gene; it installed the blueprint metaphor at the center of biology and made substrate independence feel like a finding rather than an assumption.

But the century then spent decades unlearning the very word blueprint alongside developments in far-from equilibrium thermodynamics, the thermodynamics of open systems like living systems and weather systems (or those systems that interact with the surrounding environment, exchanging matter and energy). While a blueprint is separable from the building it describes, structure and its observable properties (functions) are tightly bound in thermodynamically open systems. Thermodynamics is not responsible for the undoing of the blueprint metaphor. DNA did that on its own. Molecular biologists began to show that identical DNA in a liver cell and in a neuron yielded two radically different cells, because the “reading” is performed by the cell’s own machinery—the three-dimensional folding of the chromatin, the regulatory proteins that make some stretches of sequence legible and hide others. While early work had emphasized that DNA could be represented as two 1D text strings comprised of the letters A, T, C, and G for the 4 nucleotides, lending some support for the blueprint perspective and its value, at the very least, as an explanatory tool, reports on the significance of 3D geometry turned the molecule into the code, seriously questioning the merits of the simplification. Evelyn Fox Keller documented this paradigmatic shift in her historical overview, The Century of the Gene (2000), marking the gene-as-program metaphor obsolete by the start of the 21st century. The information does not seem to be in DNA the way text sits in a file; it is enacted by the whole system, and the sequence by itself underdetermines the organism, according to Keller. A reimagined metaphor for this new era in molecular biology would place DNA in a coordinated, biochemical stage production that ultimately conveys more than the screenplay ever could. But technological advancements in DNA sequencing and editing would drive renewed interest in computer metaphors over the next decades. Scientists began altering genes and entire genomes in virtuosic performances once restricted to nature alone. Images of scientist-as-creator and scientist-as-programmer rekindled interest in the attractive but reductive idea that life was a recipe, affirming that everything new was once old.

Structure Is Function

Substrate morphology (shape, size, etc.) critically influences local and global interactions in complex systems. This can be visualized with a sandpile simulation. Virtual particles “fall” and gradually form piles. The local perturbations create cascading avalanches. Once thought to be a largely substrate-invariant mathematical phenomenon the “sandpile” simulation illustrates how complex natural systems proceed toward self-organized criticality (SOC) over time, balancing order and chaos (sometimes described as being at “the edge of chaos”). That is the niche living systems occupy. Life must be dynamic enough to respond to perturbations, changing over time without falling into chaos. Researchers later showed that SOC depends highly on substrates when moving from virtual simulations and the substrate that originally inspired them (namely, sand) to physical simulations with different grains of rice. Grain shape significantly affected the organizational dynamics of the pile, further evidence that whatever mind is, should we take it to be organizational, and logically we should assume this (structure-activity patterns), biological substrates would be significant (proteins, nucleic acids, lipids).

Granular Criticality Simulator (Finite Run)

Grain Shape: Active Grains: 0

A complete map is not the organization

Two experiments would re-assert the limits of maps for a new generation. When Craig Venter’s team built the minimal synthetic cell, JCVI-syn3.0 (2016), they stripped a bacterium down to 473 genes, in an effort to discover the minimal set that could sustain free-living life. They “hollowed out” a biological cell to host their minimal genome. Despite the precise gene map, the task was repeatedly complicated by the non-linear interactions between genes. To determine “essential” genes they systematically deleted and restored them, measuring significance for survival. The first problem was degeneracy. Take gene pair, A and B, delete gene A but keep B and the organism survives. Delete gene B but keep gene A and the organism survives as well. Gene A and B are apparently non-essential. Delete genes A and B and the organism does not survive. Both genes A and B must be added back to the minimal genome. Yet their relationship is not disclosed in this process. The essential set could not be reasoned out in advance and numerous genes were marked as essential without any apparent function or critical role in sustaining life: 149 of the 473 genes, nearly a third, had no known function, and essentiality turned out to reside not in any single gene but in the standing organism, the complex, non-linear relationships of gene, epigenetic markers, mRNA transcripts, post-translational modifications (adjusting protein function), and the demands of the environment. The set had to be found by building and painstakingly testing the possibilities rather than derived from principles, which would be cost prohibitive and infeasible with more complicated organisms. At the end, what life is remained exactly as open as before. They recovered a working whole, not a reduced set, and no account of why it works. Effectively, the simplest task of reducing life to its constituent parts ended with an incomprehensible whole. Biology folding back upon itself. As the 21st century pushes toward new frontiers of AI research, in pursuit of far more complex organizations like human brains, may this represent at least one cautionary tale. We are nowhere near the dream of building ourselves. Much like Venter’s team we are tinkering in the dark without any theoretical grounding (first principles).

The other experiment removes the excuse that we merely lacked the map. We possess complete connectomes—every neuron and every chemical synapse of the nematode C. elegans, traced by hand and published by White, Southgate, Thomson, and Brenner in 1986, and now the entire adult brain of the fruit fly, roughly 140,000 neurons wired together, released by the FlyWire consortium in 2024. The full organizational map, in hand. Possessing it produced no designed brain, no mind, and not even a complete account of the animal’s behavior. The complete list of the parts and their connections turns out not to be the organization, because the organization was never a list. In the one case the enumeration could not be derived; in the other it was handed to us complete, and the living behavior still did not follow from it.

Separability is a matter of degree

The obvious objection is that biology recombines its own parts all the time, with no engineer present. Bacteria swap genes by horizontal transfer; whole organelles were once free-living bacteria, captured in the endosymbiosis that Lynn Margulis spent a career defending; genes duplicate and take on new roles. Separation and recombination are not only things a person does. The organism does them too.

The objection is real, and it does not reach the claim, because separability is not all-or-nothing but a matter of degree—and the degree is exactly what evolution sets. A transferred gene is a discrete piece of DNA, but its function is not discrete: transfer places that piece into a receiving cell that already supplies everything else the function needs. Nothing operates on its own; a component is moved into a compatible system that carries the rest. And how readily a gene moves tracks how connected it is to the rest of the organism. Loosely coupled genes—for metabolism, for antibiotic resistance—travel between species freely; the tightly integrated machinery of translation, the ribosome and its proteins, does not travel at all, because its identity is produced by the whole. This is separability measured, and it varies in a regular way. A claim that nothing is separable could not explain the variation; the graded claim predicts it. Organs cannot be exchanged between organisms without extensive matching and life support, and the same limit reappears at every scale. Reconstitution proposes to copy the least separable thing there is.

There is no fixed thing to copy

Suppose the objections all failed and the organization could, somehow, be specified. There would still be no single, stable organization for the specification to capture. Genetically identical organisms are not identical. Identical twins diverge in structure and temperament from the first cell divisions onward; CC, the first cloned cat (2001), wore a different coat from the donor she was copied from, because even the pattern of her fur was settled by the stochastic switching-off of one X chromosome in each cell and not by the genome at all. Development is not the execution of a program; it is a probabilistic process in which the environment is part of the outcome. There is no organization the twins share for a model to track.

The instability is not a defect to be filtered out; it is the thing itself. A lineage that held its organization perfectly fixed across generations would not be undergoing the process that makes it alive. And within a single individual the organization is remade without pause—through development, learning, ageing, and the turnover of very nearly every molecule in the body. The organization is a moving target at every scale, between individuals and within one across time. A snapshot of it is a single frame lifted from a process, and the process, not the frame, was the living thing.

Information is physical

The intuition that information floats free of matter has a real and respectable pedigree—the same one that had intoxicated the biologists. Shannon’s abstraction cut information loose from both meaning and medium: a bit is a bit whether it travels on a copper wire, a radio wave, or a wax cylinder, and that move was a stroke of genius that made the entire digital world possible. It is also true—for the narrow, precise thing Shannon defined. The overreach is the slide from “a bit can be re-encoded across media” to “everything that matters about a system is bits.”

The physicist Rolf Landauer spent a career on the correction, and compressed it to three words: information is physical (Physics Today, 1991). There is no information without a physical basis; the abstraction always has a physical and actual cost associated with it. Landauer’s earlier work, which he published in 1961, grounded the formerly metaphysical “information” with an elegant math equation: erasing a single bit of information must dissipate at least a minimum quantity of heat, kT ln 2; the logical operation of computers was unavoidably a physical event with a thermodynamic cost, a message that seems especially relevant today as AI model complexity unstably outpaces the physical cost of the mathematical operations needed to generate outputs.

In many ways, Landauer already rejected many of the speculative predictions about AI potential and digital immortality. Information is not a ghost residing in the matter. It is a configuration of the matter—and configurations of matter, unlike the idealized bits of a proof, do not lift off cleanly from what they are made of. Living systems such as ourselves shoulder the thermodynamic costs as well and are very much physical computers. Simply put, there are no ghosts residing in our brains either, just matter that dynamically reconfigures to manage the costs, unlike the classical computers running today’s AI models.

The gene was not a message about matter. It was matter that was also a message.

While this challenges a pure virtual, substrate-free existence, speculative theorists and proponents of machine consciousness and mind uploading have turned their attention to the fine grained structure of the brain as it undergoes these organizational changes. The logical conclusion drawn from a brain-as-a-physical computer is that its computation (e.g. configuration, biological state or physical state) would determine its observable outputs. By replicating the momentary fine-grained organization, we could reproduce the momentary state, possibly without biological brains. Yet “fine-grained” is not well defined in biological systems. The physical organization is fractal-like or scale invariant. We run into the same dilemmas as the ghostly or metaphysical minds of old. Something must still be “lifted,” an abstract slice of the brain’s organization deemed sufficient to encode brain states.

At the finest grain physics can reach, the argument becomes a theorem: the no-cloning theorem, proved by William Wootters and Wojciech Zurek and independently by Dennis Dieks in 1982, shows that an arbitrary unknown quantum state cannot be duplicated at all. Classical bits, mathematical constructs introduced by Shannon and formalized as Information theory, copy freely. Physically instantiated information, as Landauer’s work suggested as well, is governed by laws, preventing certain actions. When a system state is physical, that state is not copyable.

Although this shift to the microscopic is abstract, we encountered this concept previously. Genes embedded in deep, hierarchical networks could not be transferred because they become much closer to organismal states than the classic imagery of a gene as a discrete and exchangeable unit. The deeper into the substrate, and the web of relations bound to the substrate, the more copying, separability, and transferability are seemingly prohibited by laws of nature. This suggests subjective experience, while computational and physical, is also computationally irreducible. That means there is no algorithm or set of mathematical operations we can invent capable of dividing the experience up into sensible parts. To understand, is to observe over time.

Speculative theories aside, intelligence is encountered alongside life. Our everyday experience dictates that intelligence is indeed inseparable from life. That implicitly transforms artificial intelligence into artificial life; the measurement, then, may not be “is this intelligent?” or even “is this conscious?” but rather “is this alive?”, and the state of “being alive” is uncontroversially deeply relational and substrate-bound. Living organisms cannot utilize virtual amino acids. As it pertains to life, we already draw a line between human engineered machine simulations and physical reality. I would, however, be doing myself and the field of AI a disservice if that was the whole story.

True, there is a law-like rejection of copying or transferring deeply relational properties or states in nature, and by necessity, it seems, that would include life and intelligence. But evolution builds structure and function from suboptimal starting points, bringing redundancy along. Subsequently, it is feasible, in principle, that the observable organization (e.g., an example of life that we observe) is just overly complex, inefficient, and that there is a simpler, more abstract projection (call it “artificial”) that nature simply could not discover. Not because it does not exist. So I want to caution against relying entirely on biological examples to make claims about the theoretical limits of AI.

While the computational scientist is advised to never confuse the model for reality, this may be the one exception, given “model” is assigned a physical existence other than what a classical computer simulation grants; a physical substrate of transistors that is conceptually similar to modeling brains with pulleys, gears, and levers. In short, the standard computer simulation gets the abstraction part right but then makes the substrate wholly irrelevant. The complex, non-linear and physical organization of a living organism cannot be forced onto the linear, mechanical operations of computers. That form of abstraction is at best a 2D rendering of life, better known as a drawing.

We may not be able to reconstitute life and mind using silicon chips, but that does not rule out the possibility of physical abstractions of life, mind, and intelligence that are foreign to us in the 21st century. Whatever hope remains must begin and end with math. That amounts to the instantiation of math in physical substrates that can do the math. Not a ghostly projection of math onto the physical world.

Nature does math, humans represent math

Clearly, mathematical descriptions are indeed substrate-invariant. A triangle’s angles sum to 180 degrees whether the triangle is scratched in sand or cut from steel; a sorting procedure is the same algorithm running in silicon or built, absurdly but validly, out of LEGO. If a mind and intelligence were nothing more than a mathematical object—a program—then substrate independence would simply follow. This remains the mainstream philosophy of mind. Formally, the Church–Turing tradition and Hilary Putnam’s multiple realizability (1967) hold that mental states are functional roles, and a functional role can be realized in any medium that implements the function. But I have already provided examples showing structure and function are not two separate things in biology; the imagery of a “blueprint” of life and intelligence is metaphorical.

We have already encountered a soft substrate-dependence view of mind, where mind leans on its substrate only lightly—any material that carries the organization will serve. Biology, the one domain where the question can actually be tested, shows the opposite as the rule. Full gestation requires a womb; no built device raises a healthy child from conception. The liver and the pancreas cannot be manufactured; what medicine offers are bridges and single-variable control loops, not organs. Wherever the biological property itself is what is wanted, the arrow points back to biology. The exceptions are precise, and they mark the boundary rather than erasing it: we replace hearts and limbs, and little else, because a heart is a pump and a limb is a lever and a strut. Their function is mechanical, and mechanism was the part that was already substrate-neutral. Replaceability tracks how nearly mechanical a function is—substitutable where the property is structural, not substitutable where it is the living organization itself.

A printer like math is also apparently substrate-invariant: knowing what a printer does, I can build a crude one from LEGO and then extend the meaning of “ink” to melted plastic until the set of all printers quietly includes 3D printing. A brain does not sit still for it. Neurons do not simulate computation the way a chip simulates a printer; they are the computation, physically reconfiguring themselves—rewiring, re-weighting, growing—in order to perform it, so that the “program” and the “hardware” are one and the same event. A universal theory of computation is, in this respect, like the physics of weather. We can understand the physics of weather completely and still be unable to invent a hurricane, because a hurricane is not a description waiting to be instantiated on demand—it is dynamically constructed by its substrate, moment by moment, and to have one you must assemble actual atoms and retain the pertinent relationships. Also, the Navier-Stokes equations describe tornadoes, making them computable natural phenomena. But the computer simulation is lacking. The abstract, physical instantiation (model) is a vortex in a fluid. Vortices in these lab setups do not, however, generate wind and debris. Admittedly, this is irrelevant and actually desirable. Could life and intelligence be conceptually similar to this tornado-to-vortex mapping? That is the major, unresolved question in AI. Does the abstraction preserve reality, reconstituting it? Or is it a drawing that seeks to explain something that it is not and cannot be? Vortices do explain tornadoes, Navier-Stokes does determine the behavior, but the lab reproduction does not create an actual tornado and relies on external materials (scale models of physical buildings). Something, physical and irreducible is regularly added back, highlighting that the mathematics can be complete while still being physically impoverished.

Image Panel: Minds and Tornadoes. From the abstract mathematics to the still abstract but nevertheless physically realized “vortex,” tornadoes are examples of computable phenomena. Vortices are not tornadoes, and yet this may not be a critical distinction. Human self or mind could be like the tornado.

Particle Image Velocimetry visualizes vortex flow. Source: Sylverarts / Getty Images

CFD simulation of a tornado's pressure and velocity. Source: Kitware Inc.

A vortex ring in water. Source: Allexxandar / Getty Images

A forced funnel vortex. Source: genkur / Getty Image

Church-Turing gives us vortices, not actual tornadoes. And this gap could be (and likely is) considerable for minds. We simply cannot compress complex weather systems into a glass box in a lab. What can get “compressed” is tacitly about physical substrates (materials), even for complex, non-living systems. So it is presently unclear if quasi-biological substrates (those satisfying the complexity requirements of biological systems in physical models or simulations) even exist. Try as we might, the arrow seems to point back to biology itself.

While I have made this point about biological complexity before, a few more examples might be instructive. For instance, complete gestation requires a womb; no built or artificial device to date raises a healthy child from conception, despite persistent efforts. The liver and the pancreas cannot be manufactured; what medicine offers are bridges and single-variable control loops, not organs. Not complexity. Exceptions are precise and mechanical: hearts and limbs are replaceable, and little else, because a heart is a pump and a limb is a lever and a strut. Their function is mechanical; mechanism is substrate-neutral.

Replaceability in biology tracks how nearly mechanical a function is. When the structural and functional organization is relationally isolated—and not substitutable where it is the living organization itself (FIG. 03). This seems to further dampen enthusiasm around techno-utopianism and AI superintelligence parade, should the reference be biology. If intelligence is later rebranded “control” because biological intelligence turns out to mean the processes whereby organisms regulate or control their internal states amid changing inputs, then AI superintelligence would simply imply something foreign to biology at this time; being “super-natural” or “above-nature” in the sense of providing superior control or regulation than what nature could ever evolve. The vortex-in-a-box example perfectly illustrates this. Who cares if it is not a tornado? We might be similarly inclined to one day say, who cares if it is not like biology?

The strong pull of biological sensationalism in AI gets its character not from science but entertainment. Subsequently, AI emerges from this context in the 21st century, driven not by the scientific and philosophical issues that I have raised throughout, but classical engineering under commercial constraint. AI is the product optimized toward a written specification—the very method whose specification, as the previous sections have shown, runs against biology. Biology required none of it: no specification, no engineer, no profit, only optimization over deep time toward no stated aim. Even our biologically inspired creations are built for profit, which is to build for obsolescence. This is not a limit of today’s technology waiting to be lifted. It implies a hard substrate-dependence in industrial manufacturing—that anything falling outside of this scope is increasingly an impossibility. The hope necessarily rests on a fully non-physical, substrate-free blueprint to “squeeze” through the narrow opening. I have targeted this perspective and exposed its limits using biological examples. Biology is a poor reference point for current AI developments. Biology is the sales pitch: “Look, we are creating ourselves.” Sensationalism is a commitment to a belief that fiction is real and lived. While maybe some dreams are or can be real and lived, sensationalism gives us the dream that persists as the world drifts further away.

Economic Obstacles

Biological substrates are incompatible with today’s economic requirements. The more important the precise properties of biological substrates become the wider the gap between reality and imagination. This establishes another barrier to AI progress. Not only do we need major theoretical advances in neuroscience and computer science but materials science as well. Then we must contend with the reality that these developments may not be economically realizable for reasons such as material costs, availability, scalability, staffing, and the low probability of getting a reasonable return on what would be a sizable investment. Historically, it is much easier to turn the science problem into a marketing and advertising problem, sidestepping these obstacles altogether (e.g. new and innovative, “bio-inspired” etc.). The plot (right) shows values from biological and non-biological materials for two key metrics in industrial manufacturing. As substrate dependence deepens, the loftier goals of AI research seem less achievable than we might imagine.

Hopcroft, M. A., Nix, W. D., & Kenny, T. W. (2010). “What is the Young’s Modulus of Silicon?” Journal of Microelectromechanical Systems, 19(2), 229-238.

Rivnay, J., Inal, S., Salleo, A., Malliaras, G. G., Vandewal, K., & Facchetti, A. (2016). “Organic electrochemical transistors.” Nature Reviews Materials, 3(2), 1-14.

Budday, S., Nay, R., de Rooij, R., Steinmann, P., Wyrobek, T., Ovaert, T. C., & Kuhl, E. (2014). “Mechanical properties of macroscopic brain tissue.” Acta Biomaterialia, 15, 315-324.
Sze & Ng, Physics of Semiconductor Devices).
Inal et al., Nature Communications, 2017).

Hille, Ion Channels of Excitable Membranes)

Select Property to Compare:

What cannot be translated must be lived

Turing and Church assure us that anything expressible as machine instructions can be simulated on any universal computer—but not everything is expressible as machine instructions. Ask why life evolved nervous systems in the first place, and the answer is energetic: a body must continually find and spend energy simply to keep existing, and a nervous system is what lets it do so on more demanding terms. You cannot translate the energetics problem into a program. It has to be physically encountered, paid in real thermodynamic currency—lived. A simulation of hunger is not hungry; a simulation of a storm leaves the ground dry. Under a framework of pure mathematical optimization the body is an embarrassment—leaky, costly, mortal, inefficient—and so the recurring techno utopian dream is to shed the body and retain the pattern. But for a living system the body is not the packaging around the intelligence. It is the problem the intelligence exists to solve. Francisco Varela, Evan Thompson, and Eleanor Rosch gave this its name, the embodied mind (1991): cognition is not a computation performed upon a body but an activity of a body engaged with a world.

Can science build a mind?

Explanation and construction are not the same (FIG. 04); the construction of minds demands what science is not historically aimed at: construction. Explanation can be partial—a grasp of principles, a working theory that leaves detail unspecified. Construction cannot: to build a living thing, or a mind, a person must specify its organization completely enough to carry the specification out, and that specification is an enumeration of relations that are not captured by scientific methods and instruments at this time. It was illustrated for the simplest, functional engineered; it did not follow from complete neural maps (connectomes); there is simultaneously no fixed organization for science to capture, because the organism is a process that never holds still long enough. Science is assembled from snapshots. Adding detail yields a more accurate representation, not the thing represented. Accordingly, construction is less about what science does or has been successful doing. It extends chiefly from industrial manufacturing and economics where construction is far from an unbounded space of possibilities. The question is not about the theoretical possibility of building minds at all. And that is what I have mainly targeted in this article. That remains very much in doubt from my perspective, but I am not prepared to label it impossible in principle because I have relied entirely on a “common sense” argument. I connect life to mind and assign evidence of fairly strong substrate dependencies in biology for complex non-linear relationships. Whatever mind and life are, they are not exchangeable, discrete and separable units like the antibiotic resistance genes bundled in plasmids and transferred between bacteria in FIG. 01. They are deeply relational and substrate bound processes, greatly complicating their reconstruction with anything other than biological substrates (FIG. 03).

So a future in which we explain what life and mind are is entirely conceivable in my view. That is what science does. A procedure that builds either one by enumerating and assembling its relations is not—and not because life or mind is more than physical, but because the specification a person would need is one no person can complete. I did, however, offer one additional option that would involve abandoning biological references altogether:

If intelligence is later rebranded "control" because biological intelligence turns out to mean the processes whereby organisms regulate or control their internal states amid changing inputs, then AI superintelligence would imply something foreign to biology; being "super-natural" or "above-nature" in the sense of providing superior control or regulation than what nature could ever evolve

That would pull AI research in a very different direction. Strong economic constraints continue to drive the research toward profitable and easily realizable ends, with an emphasis on sensationalism. My claim above is still seemingly a substrate-dependent claim that importantly hinges on very real, significant scientific advances in our capacity to explain mind, life, and intelligence. We would need to accept that our subjective accounts are flawed —intelligence, as we encounter it, may be the messy, incomplete version of something more generic that evolution has never quite reached and may never reach free from humans pursuing it.

Which returns us to Schrödinger in wartime Dublin, and his search for what at that time was the unknown molecule of our existence. That molecule, its 3D structure, and the complete genomes of many living organisms would be reported over the next 80 years, seemingly encoding the line between life and non-life. Yet the one simple question he started with still remains unanswered in 2026: What is life? To his credit, he asked the question and made it precise enough to take seriously. Three-quarters of a century later we are still working on it, relationship-by-relationship. There is a strange freedom in accepting that a person is not a file to be copied and a mind is not a program to be resumed elsewhere: it returns us to the only place a life is ever actually lived, here, in a transient body, once. The task of AI was never to escape the substrate. It is to understand the astonishing thing that happens when matter, arranged just so, begins to understand itself—a question that is ours alone to pursue, and we have barely begun to ask with proficiency.