Information Technology Reference
In-Depth Information
to a cost function. This alternative is the “most robust” to the uncertainty on the
states of nature. In this model, dealing with uncertainty does not consist in
designing (the set D of alternatives δ ι is closed); it consists in acquiring knowl-
edge to reduce the uncertainty on θ. Risk management is an uncertainty reduc-
tion process, not a design process. Still a research project can precisely be
financed to reduce some technical uncertainty or, in marketing, a market
research project can gain knowledge to reduce uncertainty on consumer
behavior.
Hence the decision making perspective is very general and does not depend on
invariance structures; but it enables only a limited form of action. Our goal in
this paper is to cast this decision making approach into a design paradigm.
We will see how this operation helps (1) to identify a new set of concept, namely
“generic concept”; (2) some specific features of the design of generic concepts;
(3) dynamic models with repeated interactions between “invariants” and related
design space.
13.2.2 Casting the Decision Model into a Design
Framework: The Logic of Generic Concepts
To make a first step to cast decision making into a design perspective, let's rework
the equation of the Waldian model. For sake of simplicity we consider that the
sampling is reduced to 0—ie there is no opportunity to gain more knowledge on
the states of nature. Suppose that we can design an alternative δ n+1 that would be
better than all the other alternatives. It is easy to prove that the only property
required by δ n+1 is:
ð
ð
8i ¼ 1 ... n,
C θ, δ nþ1
ð
ÞμðÞdθ <
C θ, δ i
ð
ÞμðÞdθ
ð13:2Þ
Θ
Θ
Or, without simplification:
ð
μðÞd x
8i ¼ 1 ... n,
C θ, δ nþ1
ð
Þλ x δ nþ1
ð
ÞL x , θ
R n Θ
ð
μðÞd x
<
C θ, δ i
ð
Þλ x
ðÞL x , θ
δ i
R n Θ
In a design perspective, this equation actually is the brief (in C-K theory:
the concept) of a “robust” design. This concept still depends on a priori probability
μ. Actually such a δ n+1 following Eq. ( 13.2 ) would be the best for all μ in the
domain M:
 
Search WWH ::




Custom Search