[from Mary Powers 9904.22]
Some comments on the control theory thread beginning 9904.20 or so:
About control engineers knowing what they are doing - they are designing
devices that _do_ things in the "real world", from thermostats on up. That
the real world (including what their control systems are doing) is a
perceptual phenomenon in their minds is irrelevant (they think). That the
devices themselves also deal with a perceptual phenomenon that is a function
of the transducers they are given is perhaps an unusual point of view but
one which can certainly be helpful. Events like Three Mile Island would
never have happened if the engineers had considered what the control system
they designed could perceive - which was not the condition of the valve, or
the flow of water which depended on the position of the valve, but simply
the information that a signal to the valve had indeed been sent. This is
all the system could sense, and it was blind to the fact that the signal had
failed to activate the valve. So it proceeded on the basis that the valve
was activated, with the unfortunate results we all know. Obviously the
engineers involved did not imagine correctly what needed to be sensed.
In WTP's career of designing control systems, he found that taking the
system's "point of view" was a strategy that solved many problems so easily
it felt like cheating. But this maneuver, of imagining what the system he
was designing could perceive with the sensors he was giving it, occurred to
him mainly because he was learning engineering control theory and developing
PCT essentially simultaneously - a fortuitous circumstance unlikely to
happen again. This is what makes his view of control systems unusual, but
not wrong. It may, indeed, be better than what most engineers emerge from
school with.
It is rather insulting to claim that all PCT is based on chapter 1 of
control system texts. A great deal of what comes in later chapters has do
do with achieving the results one wants from the system one is building -
and involves stability issues, the inevitable crankiness and variations in
the supposedly standardized parts one is using, various unexpected
consequences of joining one system to another, etc., etc. Something people
tend to forget, I think, that underlying Bill's discussion of control lies a
technical knowledge (from later chapters in those books) that rarely
surfaces because it isn't called for in this forum. And one major reason it
isn't required is the fact that on the net we are dealing with _living_
control systems, which have evolved over millenia and eons to _be_ stable
and to work. The work Richard Kennaway is doing with Bill off the net on a
_simulated_ bug definitely requires those later chapters.
As for why choose PCT over traditional engineering: the thing that is being
sneered at here, PCT's "simplicity", is its virtue. Can you imagine a
bacterium packing the equipment necessary for model-based control? To be
able to describe the performance of bacteria, and increasing complex ones
such as humans, by the proliferation of the same simple element into huge
hierarchies, is parsimonious and elegant, and takes evolution into account
as well as things-as-they-are.
Mary P.