A Culinary Computational Creativity System - Computational Creativity Research: Towards Creative Machines - page 321

Fig. 16.1 Block diagram of computational creativity system that produces a work product and a
work plan
The designer is the lead component of the system, though its task depends heavily
on the other modules. Once system users select target categories of work products
and constituents, it generates combinations of constituents that satisfy these inputs.
As the number of possible combinations rapidly exceeds what current computers
can handle, it becomes necessary to introduce additional filtering criteria. The gen-
eration of ingredient combinations is explained in Sect. 16.3 . Once the combina-
tions of constituents are selected, their proportions must be determined. Section 16.4
describes the generation of ingredient proportions through a distributional confor-
mance method based on the learned properties of the dish to be prepared.
The assessor examines the combinations produced by the designer, and evaluates
themalong several dimensions. Certain dimensions are specific to the creative domain
of interest, in particular metrics measuring the quality of the proposed artifacts—the
flavor pleasantness of a recipe in the culinary arts, or the longevity of a fragrance in
perfumery. Others are generic enough to be used in various contexts, such as novelty
[ 15 ] or cost. Although we briefly review the assessor in Sect. 16.3 , it is described in
much more detail elsewhere [ 16 ]. The designer and the planner are the central focus
of this chapter.
Next, the work planner determines how a specific work product, selected by the
system user based on the outputs of the designer and assessor, can be executed. The
nature of the work plan varies with the creative field. For material work products, it
typically entails constituent proportions andmanufacturing instructions. Section 16.5
discusses the generation of recipe instructions. There is also a need to estimate the
time durations or efforts of the various atomic recipe steps. Section 16.6 explains
how this can be done by solving an inverse problem to deal with the fact that original
recipe data typically only lists timings for larger “molecular” recipe steps. Finally,
we conclude the chapter in Sect. 16.7 .