The mechanist fallacy and the prospect of artificial life

The philosophy of mechanism treats all physical reality as though it were a machine. Is an organism a machine? Under what circumstances could a machine become an organism? Clear answers to such questions are important to evaluate the feasibility and desirability of artificial life.

The answer to the first question is negative: an organism is not a machine, because it is not an artifact. The answer to the second question follows from an understanding of how the philosophy of mechanism leads falsely to the conclusion that natural reality can be formally exhausted in thought and recreated as artifact. A machine can become an organism only by designing itself, from the bottom up, as organisms in effect have done. An artificial organism cannot be both autonomous and fully subject to human control, any more than natural organisms are. This trade-off presents a watershed choice: to create artifacts as tools of human intent or to foster autonomous systems that may elude human control and pose a threat to us and all life.

Much of the optimism of genetic engineering rests on treating organisms as mechanisms, whose genetic program is their blueprint. But no natural thing is literally a machine, because (as far as we know) natural reality is found, not made. The quest to engineer the artificial organism from the top down rests on the theoretical possibility to analyze the natural one exhaustively, just as simulation relies on formal coding of the thing to be simulated. But, unlike machines and other artifacts, no natural thing can be exhaustively analyzed. Only things that were first encoded can be decoded.

As a way of looking, the philosophy of mechanism produces artifacts at a glance.  While this has been very fruitful for technology, imitating organisms is not an effective strategy for producing them artificially, because it can only produce other artifacts. The implicit idealist faith behind theoretical modelling and the notion of perfect simulation is that each and every property of a thing can be completely represented. A ‘property’, however, is itself an artifact, an assertion that disregards a potential infinity of other assertions. The collection of properties of a natural thing does not constitute it, although it does constitute an artifact.

A machine might be inspired by observing natural systems, but someone designed and built it. It has a finitely delimited structure, a precise set of well-defined parts. It can be dismantled into this same set of parts by reversing the process of construction. The mechanistic view of the cosmos assumes that the universe itself is a machine that can be deconstructed into its “true” parts in the same way that an engine can be assembled and disassembled. However, we are always only guessing at the parts of any natural system and how they relate to each other. The basic problem for those who want to engineer life is that they did not make the original.

We cannot truly understand the functioning of even the simplest creature and its genetic blueprint without grasping its complex interactions with environments that are the source and reference of its intentionality. Just as a computer program draws not only upon logic and the mechanics of the computer but also upon the semantically rich environment of the programmer (which ultimately includes the whole of the real world), so the developing embryo, for instance, does not simply unfold according to a program spelled out in genes, but through complex chemical interactions with the uterine environment and beyond. The genetic “program”, in other words, is not a purely syntactic system, but is rich in references that extend indefinitely beyond itself. The organism is both causally and intentionally connected to the rest of the world. Simply identifying genetic units of information cannot be taken as exhaustive understanding of the genetic “code”, any more than identifying units of a foreign language as words implies understanding their meaning.

Simulation involves the general idea that natural processes and objects can be reverse-engineered. They are taken apart in thought, then reconstructed as an artifact from the inferred design. The essence of the Universal Machine (the digital computer) is that it can simulate any other machine exhaustively. But whether any machine, program, artifact, model, or design can exhaustively simulate an organism—or, for that matter, any aspect of natural reality—is quite another question.

The characteristic of thought and language, whereby a rose is a rose is a rose, makes perfect simulation seem feasible. But there are many varieties of rose and every individual flower is unique. The baseball player and the pitching machine may both be called pitchers, but the device only crudely imitates the man, no matter how accurately it hurls the ball. Only in thought are they the “same” action. When a chunk of behavior (whether performed by a machine or a natural creature) seems to resemble a human action, it is implicitly being compared not to the human action itself but to an abstraction (“pitching”) that is understood as the essence of that behavior. Similarly, the essence or structure of an object (the “pitcher”) is only falsely imagined to be captured in a program or blueprint for its construction. Common sense recognizes the differences between the intricate human action of throwing and the mechanical hurling of the ball. Yet, the concept of simulation rests on obscuring such distinctions by conflating all that can pass under a given rubric. The algorithm, program, formalism, or definition is the semantic bottleneck through which the whole being of the object or behavior must be squeezed.

One thing simulates another when they both embody a common formalism. This can work perfectly well for two machines or artifacts that are alternative realizations of a common design. It is circular reasoning, however, to think that the being of a natural thing is exhausted in a formalism that has been abstracted from it, which is then believed to be its blueprint or essence. The structure, program, or blueprint is imposed after the fact, inferred from an analysis that can never be guaranteed complete. The mechanist fallacy implies that it is possible to replicate a natural object by first formalizing its structure and behavior and then constructing an artifact from that design. The artifact will instantiate the design, but it will not duplicate the natural object, any more than an airplane duplicates a bird.

If an organism is not a machine, can a machine be an organism? Perhaps—but only if, paradoxically, it is not an artifact! What begins as an artifact must bootstrap itself into the autonomy that characterizes organism. An organism is self-defining, self-assembling, self-maintaining, self-reproducing—in a word, autopoietic. In order to become an organism, a machine must acquire its own purposes. That property of organisms has come about through natural selection over many generations—a process that depends on birth and death. While a machine exhibits only the intentionality of its designers, the organism derives its own intentionality from participation in an evolutionary contest, through a long history of interactions that matter to it, in an environment of co-participants.

Technological development as we know it expresses human purposes; natural evolution does not. The key concepts that distinguish organism from machine are the organism’s own intentionality and its embodiment in an evolutionary contest. While a machine may be physical, it is not embodied, because embodiment means the network of relationships developed in an evolutionary context. No machine yet, however complex, is embodied in that sense or has its own purposes. Indeed, this has never been the goal of human engineers.

Quite apart from feasibility, we must ask what would be the point of facilitating the evolution of true artificial life, aside from the sheer claim to have done it? The autonomy of organisms limits how they can be controlled. We would have no more control over artificial organisms than we presently have over wild or domesticated ones. We could make use of an artificial ecology only in the ways that we already use the natural one. While it is conceivable that artificial entities could self-create under the right circumstances—after all, life did it—these would not remain within the sort of human control, or even understanding, exerted over conventional machines. We must distinguish clearly between machines that are tools, expressing their designers’ motivations, and machines that are autonomous creatures with their own motivations and survival instincts. The latter, if successful in competing in the biosphere, could displace natural creatures and even all life.

If we wish to retain human hegemony on the planet, there will be necessary limits to the autonomy of our technology. That, in turn, imposes limits on its capabilities and intelligence, especially the sort of general and self-interested intelligence we expect from living beings. We must choose—while we still can—between controllable technology to serve humans and the dubious accomplishment of siring new forms of being that could drive us to extinction. This is a political as well as a design choice. Only clarity of intention can avoid disaster resulting from the naive and confused belief that we can both retain control and create truly autonomous artifacts.