International Encyclopedia of Systems and Cybernetics

A device created by W.R. ASHBY to demonstrate some properties of the ultrastable systems.

In St. BEER words: "The Homeostat is precisely a machine for settling down again after it has been disturbed" (1967, p.117). As such a property is very general in complex systems which must "cope with an unpredictable environment,… (conveying) random information" (p.118), the homeostat can be considered a behavioral modelling device of considerable interest. In effect, "we are returned to the old issue of machine design, and that of adaptive behavior to situations not envisaged by the designer" (Ibid).

ASHBY gives a complete description of this "machine", whose main points are that: "The homeostat consists of four units, each of which carries on top a pivoted magnet. The angular deviations of the four magnets from the central positions provide the four main variables" (1960, p.100)…

Each unit may take 25 different positions and its outputs are transmitted in a randomized way to the other units and influence their activity, thus providing 25, i.e. 390.625 combinations of parameter-values.

ASHBY explains that: "The field of the four main variables has only one state of equilibrium (at the centre), which may be stable or unstable. Thus the system is either stable and self-correcting for small impulsive displacements to the needles, or unstable and self-aggravating, running away to the limits of the throughs. Which it will be depends on the quantitative details of the primary feedbacks, which are dependent on the values on the stepmechanisms" (p.1 04).

As stated by E.von GLASERSFELD, ASHBY's device would gradually learn to "… make appropriate connections by learning on the basis of its own experience which activities help to reduce which error-signals" (1988, p.117).

However, as stated by St. BEER: "… if a homeostat in an equilibrial state is bombarded without respite by wave after wave of input data, it will go into an oscillation from which it can never recover… if the periodicity of the input is very much faster than the cycle time of the homeostat itself, then clearly it has no opportunity to settle" (1968, p.384).

Thus, either the periodicity of the input must be extended or, if it can be proved repetitive, an adaptive algorithm must be evolved.

Numerous behaviors of many kinds of systems can be mimicked on this not too complicated device, as for example the tendency to stability or instability, the ways of training, some failures of adaptation, etc…

The homeostat, which is now somewhat forgotten, could still probably be very useful.

On the semantic convenience of the name "Homeostat", ASHBY wrote: "It was given the name of "Homeostat" for convenience of reference, and the noun seems to be acceptable. The derivates "homeostatic" and "homeostatically", however, are unfortunate, for they suggest reference to the machine, whereas priority demands that they be used only as derivatives of CANNON's "Homeostasis" (1960, p.100).