Information Technology Reference
In-Depth Information
the level of parts as well as wholes. Second, and again like Walter's, Ashby's
science was a science of a heterogeneous universe: on the one hand, the brain,
which Ashby sought to understand; on the other, an unknown and cognitively
unknowable (to the homeostat) world. Performative interaction with the un-
knowable was thus a necessary constituent of Ashby's science, and in this
sense the homeostat returns us to an ontology of unknowability. And, third, a
discovery of complexity also appears within Ashby's cybernetics, though this
again requires more discussion.
In chapter 3 we saw that despite its simplicity the tortoise remained, to
a degree, a Black Box, capable of surprising Walter with its behavior. The
modern impulse somehow undid itself here, in an instance where an atomic
understanding of parts failed to translate into a predictive overview of the
performance of the whole. What about the homeostat? In one sense, the ho-
meostat did not display similarly emergent properties. In his published works
and his private journals, Ashby always discussed the homeostat as a demon-
stration device that displayed the adaptive properties he had already imagined
in the early 1940s and first discussed in print in his 1945 publication on the
bead-and-elastic machine.
Nevertheless,
combinations
of homeostats quickly presented analytically
insoluble problems. Ashby was interested, for example, in estimating the
probability that a set of randomly interconnected homeostats with fixed
internal settings would turn out to be stable. In a 1950 essay, he explored
this topic from all sorts of interesting and insightful angles before remarking
that, even with simplifying assumptions, “the problem is one of great [math-
ematical] difficulty and, so far as I can discover, has not yet been solved. My
own investigations have only convinced me of its difficulty. That being so
we must collect evidence as best we can” (Ashby 1950a, 478). Mathematics
having failed him, Ashby turned instead to his machines, fixing their param-
eters and interconnections at random in combinations of two, three, or four
units and simply recording whether the needles settled down in the middle
of their ranges or were driven to their limits. His conclusion was that the
probability of finding a stable combination probably fell off as (1/2)
n
, where
n
was the number of units to be interconnected, but, rather than that specific
result, what I want to stress is that here we have another discovery of com-
plexity, now in the analytic opacity of multihomeostat setups. Ashby's atomic
knowledge of the individual components of his machines and their intercon-
nections again failed to translate into an ability to predict how aggregated
assemblages of them would perform. Ashby just had to put the units together
and see what they did.
