"Minimal designer intervention" is not just nice, it is paramount. It can
also be stated as minimal interference with the designer. The optimum
interface would simply let the designers do what they do already
write and draw anything they want during the course of the conceptual
design process. Any additional SECONDS of participation required by the
interface on the part of the designer during the working session must be
jealously metered out. If the designer feels the interface gets in the way
of doing design, it won't be used in actual practice.
The interface should not only be easy to use, but require NO designer
participation at all to function at the lowest level. And any
participation the designer does choose to donate must be handsomely
rewarded; ie great payoff to the designer personally for small increments
of increased participation. In particular we strove to avoid mandating the
use of syntax
driven menus in order to specify designs. Filling out forms
is a valuable data entry technique in general, but often anoying during the
conceptual design of mechanical artifacts.
There must also be minimal knowlege engineer participation
a goal of a
few hours per month per designer.
Another criteria in the project was to accept the full richness of raw
design notebooks as generated during design
that is, to let the
designers create whatever text graphic information structures help them do
design. And then during the map making, the system should deal with those
aspects of that information that it can, but SAVE all the information to
have it available for re processing again some time in the future (when we
understand more about text graphics and design activity). This could be
called a "text graphic based reasoning" approach, parallel to text based
approaches, but including graphics.
A final criteria was upward compatibility with more traditional AI
techniques. The global, raw text
graphics first philosophy proposed here
has certain advantages, but we also wanted at the same time to support
in depth analysis of particular kinds of information in the notebook, as
well as formal model based approaches to capturing design rationale. The
ideal would be an integration of these various techniques the use of
in depth analysis and formal models would produce nicely mappable features,
and at the same time design information maps would themselves be rich areas
for application of analytic and formal methods.