machine built in 1948 by W. Ross Ashby, a psychiatrist working at Barnwood House hospital in Gloucester, from war-surplus parts. It did nothing useful. Its point was that it would find its own way back to stability after being disturbed, by changing itself.

what it is

Four identical units, each with a pivoted magnet carrying a vane in a trough of water; the position of each vane set a current fed to the other three. Each unit had a uniselector, a telephone-exchange stepping switch, wired so that when the unit's needle strayed outside set limits the switch stepped, giving the connections new, effectively random, values. The machine kept stepping until it reached a configuration in which every needle stayed within bounds (Ashby, 1952) [paraphrase, not a quotation].

what it was for

Ashby called this ultrastability: a system that, when its first-order feedback fails, uses a second loop to change its own parameters until the first works again. He had argued since 1940 that adaptation in organisms is the same thing as stability, and the homeostat was the argument built in metal. If someone reversed a connection, or tied a unit's controls, the machine would search and settle again. See Ashby.

what's contested

Was it a model of the brain or of evolution by trial and error? Random search is hopelessly slow for anything large, which Ashby knew; his answer was that real adaptive systems must be broken into many nearly independent parts that can settle separately. Pickering (2010) reads the homeostat as a different picture of the mind altogether, performative rather than representational: a device that copes without modelling its world [paraphrase, not a quotation]. Others saw only a demonstration of a principle.

methodological critiques

The homeostat shows that ultrastability is possible, not that brains use it. Like Grey Walter's light-seeking robot tortoises (Walter, 1953) (Holland, 2003), it is an existence proof built to settle an argument about whether simple mechanisms can produce lifelike behaviour.

later reception

Ashby developed the ideas into An Introduction to Cybernetics, with the law of requisite variety, and later the good regulator theorem (Conant & Ashby, 1970). Echoes appear in self-organising systems, in the "search until stable" logic of some learning algorithms, and in debates about systems that change their own objectives; see specification gaming.

in the tour: the room called "Governors and feedback" has a placard called "The homeostat" ···

References

1. Ashby (1952). Design for a Brain. London: Chapman & Hall. The 1954 Wiley printing is scanned at the Internet Archive. archive.org/details/designforbrain00ashb
2. Ashby (1940). Adaptiveness and Equilibrium. Journal of Mental Science. doi:10.1192/bjp.86.362.478
3. Pickering (2010). The Cybernetic Brain: Sketches of Another Future. Chicago: University of Chicago Press.
4. Walter (1953). The Living Brain. London: Duckworth.
5. Holland (2003). Exploration and high adventure: the legacy of Grey Walter. Philosophical Transactions of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences. doi:10.1098/rsta.2003.1260
6. Ashby (1956). An Introduction to Cybernetics. London: Chapman & Hall; New York: Wiley. Scanned at the Internet Archive. archive.org/details/introductiontocy00ashb
7. Conant & Ashby (1970). Every good regulator of a system must be a model of that system. International Journal of Systems Science. doi:10.1080/00207727008920220

see also