Exactly so.
Mission critical systems are built to be tolerant of small errors.
It is the small error that turns into a big one (the cascade) that
one dreads. For example, when the missile fails to take off and
you only find out after the other one is in the air. Or the humanIntheLoop
when a sub-sub-sub contractor decides to save a few shekels in
manufacturing, doesn't stiffen the board, it shakes loose in flight,
and it has the component that tells the warhead to separate.
AKA, oooopsie...
The truly bad one is the unk-unk: unknown-unknown.
len
From: Michael Kay [mailto:]
> I guess the serious question is whether the cost of a few laughable
> failures of the heuristic algorithms outweighs the cost of building
> ontologies maintaining a lot of semantic metadata in the instances?
> My guess is "no", by a couple orders of magnitude.
>
> Of course the cost-benefit calculation would be very different for a
> military system or a mission-critical enterprise system.
I wouldn't be so sure. You can spend billions on cataloguing all your
information and still make laughable mistakes.
