Creativity and AI: A Conceptual Blending approach

260 Pages·2006·2.11 MB·English

Creativity and AI: A Conceptual Blending
approach
Francisco Cˆamara Pereira

To Ana
Abstract
The main motivating factor behind this book is the appealing, yet often controversial,
goal of computational modelling of creativity. A cross-disciplinary study of creativity
is a mandatory first step and this will help us construct a synthesis from an AI pers-
pective. From this, we focus on a Model of Concept Invention and, more practically,
on a computational system, Divago.
The Model of Concept Invention is built over the expected principles for a creative
system. It is formalized, although in a computationally idealistic manner, i.e. its
implied complexity prevents it from a feasible implementation. Divago is the partial
instantiation of this abstract model and comprises the main technical substance of
the book. This system is centered on an implementation of the cognitive linguistics
framework of Conceptual Blending, as well as on a mapping algorithm based on
Metaphor work.
Divago was subject to experimentation in a range of applications and analyzed
according to methodologies that have been proposed with the area of Creativity and
AI.Othervalidationproceduresarefollowed,namelyinthecomparisontootherworks
and to Conceptual Blending literature.
ii
Acknowledgements
In parallel to the work behind this book, a whole novel could have been written, with
many more actors than appears in the final document. The truth is that, without
these actors, it would certainly lose much of its value.
My supervisor, Prof. Dr. Am´ılcar Cardoso, has been a key actor from the first to
the last chapter, sometimes as a patient advisor and listener, other times as a friend.
Some colleagues (and friends) had the leading role on occasion, even if they were
unaware of it. In fact, a substantial part of this book was a result of talks, collabo-
rations and constructive arguments with Tony Veale, Pablo Gerv´as, Jo˜ao Leite and
Geraint Wiggins.
Also, without the rest of the AILab crew (before and after ECOS), a lot of science
and friendship chapters would certainly be irreparably lost. These were Penousal,
Paulo, Grilo, Xico, Leonor, Prof. Carlos Bento, Prof. Ernesto Costa, Jorge Tavares,
Andr´e, Miguel Ferrand and Ana Alves.
Special thanks must be made to Paulo Ribeiro, Jo˜ao Pedro Martins, Bruno Leit˜ao
and Bruno Marques who have contributed with work that is also strongly linked to
the thesis that originated the book.
A large part of this novel was done within the environs of the Departament de
Informatics Engineering of the University of Coimbra, and I must not forget to thank
my colleagues there and particularly the staff, Isabel, Paula, Manuela and M´arcia,
for their support.
The other side of the story, with different scenery and people, starts at home.
Ana has given the strongest support and love as only she can give and has frequently
drawn from a limitless reservoir of patience. Kali and Sansao have also been patient
and supportive, in the way that only cats and dogs can be. Kali has left us during
the last phase, and we feel a deep sorrow, for this and future novels would never be
the same without her.
Finally, my parents and sister deserve a special mention for the many times that
I missed them, especially when they needed me. And for the support they have given
iii
from the beginning.
I must also kindly thank the reviewers of this (and previous) publications for the
severalconstructiveandincisivecommentsthatdefinitelyhelpedimproveitspotential
as a contribution to science.
Table of Contents
Table of Contents v
1 Introduction 2
1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Creativity 8
2.1 Creativity theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Divergent Production . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.2 Bisociation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.3 The Systems Model . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.4 Boden’s taxonomies . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1.5 Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.6 Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Computational Creativity . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.1 Two accounts for characterizing creativity . . . . . . . . . . . 30
2.2.2 Creative systems . . . . . . . . . . . . . . . . . . . . . . . . . 36
3 Working with Concepts 45
3.1 What is a Concept? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Building concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.1 Concept Formation . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2.2 Concept Invention . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Mixing concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.1 Conceptual Combination . . . . . . . . . . . . . . . . . . . . . 49
3.3.2 Conceptual Blending . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Metaphor and Analogy . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.4.1 Structure Mapping Engine . . . . . . . . . . . . . . . . . . . . 66
3.4.2 Conceptual Scaffolding and Sapper . . . . . . . . . . . . . . . 68
3.4.3 Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4 A Model of Concept Invention 80
v
CONTENTS
A Top-Down approach . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1 A Creative General Problem Solver . . . . . . . . . . . . . . . . . . . 80
4.2 Description of the Model . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5 Divago 95
5.1 Overview of the Architecture . . . . . . . . . . . . . . . . . . . . . . . 95
5.2 Knowledge Base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.3 Mapper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.4 Blender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.5 Factory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.6 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.6.1 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.6.2 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.6.3 Pattern Completion . . . . . . . . . . . . . . . . . . . . . . . . 124
5.6.4 Maximization/Intensification of Vital Relations . . . . . . . . 126
5.6.5 Unpacking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.6.6 Web . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.6.7 Relevance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.7 Elaboration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.8 Divago as a creative system . . . . . . . . . . . . . . . . . . . . . . . 134
6 Experiments 136
6.1 The Boat-House . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.2 The Horse-Bird . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.2.1 Evaluating the Optimality Pressures . . . . . . . . . . . . . . 147
6.2.2 Finding the pegasus . . . . . . . . . . . . . . . . . . . . . . . 150
6.3 Noun-noun combinations . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.3.1 Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.3.2 Free generation . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.3.3 Comparison to C3 . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.4 The creature generation experiment . . . . . . . . . . . . . . . . . . . 172
6.4.1 Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
6.4.2 Free generation . . . . . . . . . . . . . . . . . . . . . . . . . . 176
6.5 The established Blending examples . . . . . . . . . . . . . . . . . . . 183
6.5.1 Experiments with isolated principles . . . . . . . . . . . . . . 187
6.5.2 Experiments with combination of principles . . . . . . . . . . 193
6.5.3 Some considerations . . . . . . . . . . . . . . . . . . . . . . . 197
6.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7 Conclusions and Future Directions 204
A The effect of input knowledge 211
vi
CONTENTS
B Established examples of Conceptual Blending 214
C The Generalized Upper Model Hierarchy 223
D Programming the frames 226
D.1 Syntax and Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
D.2 Programming of the frame . . . . . . . . . . . . . . . . . . . . . . . . 227
D.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
E Instances in the house-boat experiment 235
F Experiments, Databases and other documents 238
Bibliography 240
vii
f
1