A Culinary Computational Creativity System - Computational Creativity Research: Towards Creative Machines - page 320

16.1 Introduction
Traditional work in computational creativity has typically focused on domains such
as visual arts, writing, and music [ 1 - 5 ]. In characterizing human creativity, how-
ever, it has been found that there are at least three distinct groupings of creative
domains: expressive creativity (visual arts, writing, humor); performance creativity
(dance, drama, music); and scientific creativity (invention, science, culinary) such
that abilities within one member of a category transfer over to other members of the
category [ 6 ]. Given the potential business impacts of creativity in science and inven-
tion [ 7 , 8 ], we focus our efforts here on scientific creativity, rather than expressive
or performance creativity. In particular, we take culinary creativity as our specific
domain; the basic techniques developed here could also be useful for other kinds of
scientific creativity, especially in those settings where artifacts are to be constructed
from components.
We adopt a definition of creativity used in human creativity research [ 9 ]: Creativ-
ity is the generation of a product that is judged to be novel and also to be appropriate,
useful, or valuable by a suitably knowledgeable social group . By defining princi-
pled descriptions of artifacts as compositions of parts, we compute novelty using
information-theoretic measures that have been validated with psychology experi-
ments across domains [ 10 ]. For appropriateness and value, we draw on fundamental
chemical properties of flavor and neurobiological properties of flavor perception to
define measures [ 11 , 12 ]. In both the novelty and quality dimensions, the scientific
nature of the domain lends itself to data-driven assessment. The ability to assess
enables the selective step of creativity, which is arguably just as important as the
generative one [ 9 , 13 ].
Notwithstanding, the central focus of this chapter is on the generative algorithms
we use for creating novel and flavorful culinary recipes. Note that unlike Morris
et al. [ 14 ], who also looked at culinary creativity, generating ingredient proportions
and recipe steps (with timings) rather than simply the list of ingredients is critical for
our work. Our system produces a complete work plan for physically creating a dish.
A block diagram for our computational creativity system is presented in Fig. 16.1 .
Data from a domain knowledge database is categorized and fed to the work product
designer, which interacts with the work product assessor to suggest one or more
work products, and with the work planner to generate the corresponding work plan.
Applied to the culinary domain, the work product designer outputs combinations
of ingredients and their proportions—the recipes—and the work planner produces
preparation instructions.
The domain knowledge database consists of a collection of existing work products
and their constituents, including information pertaining to their properties, quality
assessments, styles, and cultural associations. The knowledge categorizer is respon-
sible for classifying these products and constituents, defining an ontology that helps
users specify the profile of the novel products they wish to create. This is described
in Sect. 16.2 .