Abstract
Artificial life studies life-like organisation across possible substrates, and classifies its bottom-up systems by where the artificial organism appears: in a physico-chemical, physico-mechanical, or computational medium. We argue that this taxonomy leaves a second axis underspecified, the causal role of the computing substrate itself, and that the omission carries a cost. The stakes of this missing axis are visible in the field’s recent stocktaking: virtual artificial life has not yet exhibited a system that satisfies the full requirements of life (autopoiesis, agency, and open-ended adaptation), with thermodynamic abstraction and the brittleness of symbolic substrates recurring as limitations (Stepney, 2025). We propose substrate-coupled artificial life (SCAL) as a research programme in which the physical dynamics of the substrate participate causally in the production, persistence, and adaptation of life-like organisation, rather than serving as an inert execution layer. The framework is grounded in three observations about conventional silicon: 1/f noise from charge-trap dynamics at MOSFET interfaces, a cache-hierarchy latency gradient that satisfies Lewontin-style criteria for differential persistence among self-reinforcing memory-topology structures, and DRAM operating far from equilibrium as a Prigogine-class dissipative structure. We specify the experimental requirements such a programme implies in conventional silicon and set out specific falsification criteria. The work is theoretical and pre-experimental: it defines the framework, identifies the experimental commitments required to test it, and specifies the conditions under which it would fail.