Society for the Study of Artificial Intelligence and Simulation of Behaviour (SSAISB or AISB), London, United Kingdom


Terrence Deacon :: In what sense could a machine be alive?

Published on Aug 29, 2014

Definitions of life are notoriously either vague and overgeneralized or fragmented and disjointed. Commonly cited properties include self-maintenance and reproduction, though being capable of evolving is often also considered relevant, and the presence or absence of one or another of these may not be critical. Slightly more descriptive accounts often include the reciprocal production of components, each from interactions among the others (as in the concepts of autopoiesis and hypercycle). Most definitions lack any account of the underlying mechanism that achieves these functions, however. ``Artificial life" simulations, which exhibit the tendency of algorithms or their graphical representations to reproduce and evolve, similarly tend to be agnostic about how these properties are physically realized, and ignore altogether the material and energetic basis of these processes. Probably the first and best-known effort to specify the properties required to define a living machine was produced by John von Neumann in his 1966 paper on self-reproducing machines. In this model there was a strict distinction between the instruction code for the design of the machine and the construction mechanism that implemented and ultimately copied these instructions. Though this logic was the precursor to the whole field of artificial life and cellular automata, its full physical realization remains beyond reach. I argue that this difficulty reflects the failure to define life mechanistically, i.e. by specifying the types of physical processes that reciprocally produce their component parts, explaining how these self-reconstituting processes maintain the integrated unity of the whole system, and demonstrating how the architecture of the whole can be represented in a form that doesn't become corrupted by turnover of components and factors that degrade system organization. A process called autogenesis is described in which multiple self-organizing processes are linked by virtue of each producing the critical boundary constraints that maintain the others. By virtue of this synergy of constraints they collectively generate a second-order constraint that maintains the integrity of the whole. Because this second-order constraint is neither a physical nor a chemical constraint, but rather a constraint on the relationships between lower-order constraint-generating processes, it persists unchanged despite turnover of substrates and partial degradation of system integrity. Though described in terms of molecular components, autogenesis is a generic process that should be realizable in diverse substrates and at diverse scales. I argue that a machine (in the widest sense of being a system of interacting parts) is alive if it is autogenic.?
 
Back
Top