Proceedings of - Association for Computational Creativity (PDF)

Proceedings of François Pachet, Amilcar Cardoso, Vincent Corruble, Fiammetta Ghedini (Editors) ICCC 2016 Paris | 27 June - 1 July  Sony CSL Paris, France http://computationalcreativity.net/iccc2016/ First published 2016 TITLE: PROCEEDINGS OF THE SEVENTH INTERNATIONAL CONFERENCE ON COM- PUTATIONAL CREATIVITY EDITORS: François Pachet, Amilcar Cardoso, Vincent Corruble, Fiammetta Ghedini ISBN: 9782746691551 Technical editor: Fiammetta Ghedini  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 Preface This volume contains the papers presented at ICCC 2016, the 7th International Conference on Computational Creativity held in Paris from June 26th to July 1st, 2016. The conference was hosted at Université Pierre & Marie Curie, in Paris. Computational creativity is the art, science, philosophy and engineering of computational systems which, by taking on particular responsibilities, exhibit behaviours that unbiased observers would deem to be creative. As a field of research, this area is thriving, with progress in formalising what it means for software to be creative, along with many exciting and valuable applications of creative software in the sciences, the arts, literature, gaming and elsewhere. The ICCC confe- rence series, organized by the Association for Computational Creativity since 2010, is the only scientific conference that focuses on computational creativity alone and also covers all its aspects. We received 82 paper submissions, a record number for ICCC, which confirms the growing interest for this field. Papers were submitted in five categories: 1) technical papers advancing the state of art in research, 2) system and resource des- cription papers, 3) study papers presenting enlightening novel perspectives, 4) cultural application papers presenting the usage of creative software, and 5) position papers arguing for an opinion. Each submission was reviewed by 4 program committee members and then discussed among the reviewers, if needed, to resolve controversial and borderline cases. Senior Program Committee Members led discussions and also prepared recommendations based on the reviews and discussions. In total, over 400 reviews and meta-reviews were carried out in the process. The committee accepted 51 full papers. Papers were presented either as oral presentations, posters or demos, depending on the nature of the contribution. The three-and-a-half days of the ICCC 2016 scientific program consisted in a series of exciting sessions for oral presen- tations of papers and a special session for posters and demos. The program included an invited talk by Todd Lubart, Professor of Psychology, entitled “Homo Creativus: A psycholo- gical perspective”. This conference included many events related to creativity and computers, all held on the Jussieu campus. Two workshops were held: the 4th International Workshop on Musical Meta Creation (MUME 2016), and the 4th Com- putational Creativity & Games workshop (CCGW16). Two tutorials were also organised, one on the Engagement-Reflection Model and another on Computational Creativity, organised by the PROSECCO project. A series of talks from ERC funded projects were given, with artists drawing in real-time during the talk, to experiment with novel ways of disseminating such projects. A concert of a Baroque comic opera “Casparo”, which tells the story of a humanoid robot, composed by Luc Steels (libretto by Oscar Villaroya) was performed. A special Flow-Machines session highlighting the main results of the pro- ject was held as well as short music performances with the interactive systems developed in this project. This year we inaugurated a video competition (11 submissions, 4 were retained, and 2 were given a prize offered by Sony CSL). The winner of the video competition was Alida Horsley for the video Hidden Pasts, Digital Futures.The best paper award has been awarded to Maximos Kaliakatsos-Papakostas, Roberto Confalonieri, Joseph Corneli, Asterios Zacharakis and Emilios Cambouropoulos for the paper An Argument-based Creative Assistant for Harmonic Blending. We thank our sponsors, from which we received very useful support: Lip6, UPMC, Sony CSL, The Journal of Artificial Intelligence Research, the PROSECCO network, AAAI. Special thanks to the ERC-funded Flow-Machines and ERCCo- mics projects. We thank the program committee, the senior program committee and other reviewers for their hard work in reviewing papers and the EasyChair platform that made our work easier. François Pachet Amilcar Cardoso Vincent Corruble Fiammetta Ghedini Program chair General chair Local chair Publicity chair July 2016 ii  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 Conference Chairs General Chair: F. Amílcar Cardoso, University of Coimbra Program Chair: François Pachet, SONY CSL Paris Publicity Chair: Fiammetta Ghedini, SONY CSL Paris Local Chair: Vincent Corruble, LIP6, UPMC (Paris 6) Senior Program Committee Oliver Bown; Design Lab, University of Sydney Simon Colton, Goldsmiths College, University of London Pablo Gervás, Universidad Complutense de Madrid Nada Lavrač, Jozef Stefan Institute Mary Lou Maher, University of North Carolina – Charlotte Nick Montfort, Massachusetts Institute of Technology Alison Pease, Imperial College London Rafael Perez Y Perez, Universidad Autónoma Metropolitana at Cuajimalpa Graeme Ritchie, University of Aberdeen Rob Saunders, University of Sydney Hannu Toivonen, University of Helsinki Tony Veale, University College Dublin Dan Ventura, Brigham Young University Geraint Wiggins, Queen Mary, University of London Program Committee Kat Agres, Queen Mary, University of London Wendy Aguilar, IIMAS - UNAM Josep Blat, Universitat Pompeu Fabra Giordano Cabral, UFRPE, Recife, Brazil Michael Cook, Goldsmiths College, University of London Joseph Corneli, Goldsmiths, University of London Vincent Corruble, LIP6, Universite Pierre et Marie Curie (Paris 6) Alberto Diaz, Universidad Complutense de Madrid Mark d’Inverno, Goldsmiths, University of London Arne Eigenfeldt , Simon Fraser University Liane Gabora, University of British Columbia Ashok Goel, Georgia Institute of Technology Andrés Gómez de Silva Garza, Instituto Tecnológico Autónomo de México Hugo Gonçalo Oliveira, CISUC, University of Coimbra Jeremy Gow, Goldsmiths, University of London Kazjon Grace, University of North Carolina at Charlotte Raquel Hervás, Universidad Complutense de Madrid Amy K. Hoover, University of Central Florida Bipin Indurkhya, AGH University of Science and Technology Anna Jordanous, University of Kent Robert Keller, Harvey Mudd College Carlos León, Universidad Complutense de Madrid Antonios Liapis, University of Malta Maria Teresa Llano Rodriguez, Goldsmiths, University of London Ramon Lopez De Mantaras, IIIA – CSIC iii  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 Penousal Machado, CISUC, Department of Informatics Engineering, University of Coimbra Pedro Martins, University of Coimbra Brian Magerko, Georgia Institute of Technology Ruli Manurung, Faculty of Computer Science, Universitas Indonesia Jon McCormack, Monash University David Meredith, Aalborg University Diarmuid O’Donoghue, National University of Ireland, Maynooth Alexandre Miguel Pinto, University Of Coimbra Enric Plaza, IIIA-CSICS Senja Pollak, Jozef Stefan Institute and University of Ljubljana Matthew Purver, Queen Mary University of London Mark Riedl, Georgia Institute of Technology Pierre Roy, Sony CSL Marco Schorlemmer, Artificial Intelligence Research Institute, IIIA, CSIC Emily Short Adam M. Smith, University of California Santa Cruz Ricardo Sosa, SUTD, Singapore Oliviero Stock, FBK-irst Julian Togelius, New York University Tatsuo Unemi, Soka University Frank van der Velde, University of Twente Lav Varshney, University of Illinois at Urbana-Champaign Dekai Wu, HKUST Ping Xiao, University of Helsinki Georgios N. Yannakakis, Institute of Digital Games, University of Malta Martin Znidarsic, Jožef Stefan Institute iv  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 Table of Contents Preface................................................................................................................................................................5 Keynote Talk Homo Creativus: A psychological perspective...............................................................................................viii Todd Lubart Search Novelty-Seeking Multi-Agent System..................................................................................................................1 Simo Linkola, Tapio Takala and Hannu Toivonen Supportive and Antagonistic Behaviour in Distributed Computational Creativity via Coupled Empowerment Maximisation .....................................................................................................................................................9 Christian Guckelsberger, Christoph Salge, Rob Saunders and Simon Colton Mere Generation: Essential Barometer or Dated Concept? ...........................................................................17 Dan Ventura Searching for Surprise......................................................................................................................................25 Georgios N. Yannakakis and Antonios Liapis Role of Simplicity in Creative Behaviour: The Case of the Poietic Generator ...............................................33 Antoine Saillenfest, Jean-Louis Dessalles and Olivier Auber Evaluation Investigating Listener Bias Against Musical Metacreativity............................................................................42 Philippe Pasquier, Adam Burnett and James Maxwell Preference Models for Creative Artifacts and Systems.....................................................................................52 Debarun Bhattacharjya Evaluating digital poetry: Insights from the CAT............................................................................................60 Carolyn Lamb, Daniel Brown and Charles Clarke Dependent Creativity: A Domain Independent Metric for the Assessment of Creative Artifacts...............................68 Celso França, Luis Fabricio Wanderley Goes, Alvaro Amorim, Rodrigo Rocha and Alysson Ribeiro Da Silva - Regent Interaction Modes for Creative Human-Computer Collaboration: Alternating and Task-Divided Co-Creativity ............77 Anna Kantosalo and Hannu Toivonen Experience Driven Design of Creative Systems...............................................................................................85 Matthew Yee-King and Mark d’Inverno Applying Core Interaction Design Principles to Computational Creativity..........................................................93 Oliver Bown and Liam Bray Designing Improvisational Interfaces..............................................................................................................98 Jon McCormack and Mark d’Inverno Models of Creativity Visual Hallucination For Computational Creation.........................................................................................107 Leonid Berov and Kai-Uwe Kuhnberger Crossing the horizon: exploring the adjacent possible in a cultural system..................................................115 Pietro Gravino, Bernardo Monechi, Vito D. P. Servedio, Francesca Tria and Vittorio Loreto Computational Creativity Conceptualisation Grounded on ICCC Papers.....................................................123 Senja Pollak, Biljana Mileva Boshkoska, Dragana Miljkovic, Geraint Wiggins and Nada Lavrac v  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 An institutional approach to computational social creativity.........................................................................131 Joseph Corneli Understanding Musical Practices as Agency Networks.................................................................................139 Andrew R. Brown A History of Creativity for Future AI Research..............................................................................................147 Mark d’Inverno and Arthur Still Visual Arts Deep Convolutional Networks as Models of Generalization and Blending within Visual Creativity............156 Graeme McCaig, Steve Dipaola and Liane Gabora X-Faces: The eXploit Is Out There...................................................................................................................164 Joao Correia, Tiago Martins, Pedro Martins and Penousal Machado Before A Computer Can Draw, It Must First Learn To See............................................................................172 Derrall Heath and Dan Ventura Creative Generation of 3D Objects with Deep Learning and Innovation Engines........................................180 Joel Lehman, Sebastian Risi and Jeff Clune Digits that are not: Generating new types through deep neural nets...........................................................188 Kazakci, Mehdi Cherti and Balazs Kegl Narratives Murder Mystery Generation from Open Data................................................................................................197 Gabriella Barros, Antonios Liapis and Julian Togelius Framing Tension for Game Generation.........................................................................................................205 Phil Lopes, Antonios Liapis and Georgios N. Yannakakis What If A Fish Got Drunk? Exploring the Plausibility of Machine-Generated Fictions........................213 Maria Teresa Llano Rodriguez, Christian Guckelsberger, Rose Hepworth, Jeremy Gow, Joseph Corneli and Simon Colton Language and Text Exploring the Role of Word Associations in the Construction of Rhetorical Figures....................................222 Paloma Galvan, Virginia Francisco, Raquel Hervas, Gonzalo Mandez and Pablo Gervas Meta4meaning: Automatic Metaphor Interpretation Using Corpus-Derived Word Associations.................230 Ping Xiao, Khalid Alnajjar, Mark Granroth-Wilding, Kathleen Agres and Hannu Toivonen One does not simply produce funny memes! - Explorations on the Automatic Generation of Internet humor.....238 Hugo Gonçalo Oliveira, Diogo Costa and Alexandre Miguel Pinto Poetry from Conceptual Maps - Yet Another Adaptation of PoeTryMe’s Flexible Architecture......................246 Hugo Gonçalo Oliveira and Ana Oliveira Alves Analysis of the correlations between the knowledge structures of an automatic storyteller and its literary production...254 Ivan Guerrero Roman and Rafael Perez Y Perez Structure Flexible Generation of Musical Form: Beyond Mere Generation.................................................................264 Arne Eigenfeldt, Oliver Bown, Andrew Brown and Toby Gifford Generative Choreography using Deep Learning............................................................................................272 Luka Crnkovic-Friis and Louise Crnkovic-Friis Investigating the Musical Affordances of Continuous Time Recurrent Neural Networks..............................278 Steffan Ianigro and Oliver Bown How Blue Can You Get? Learning Structural Relationships for Microtones via Continuous Stochastic Trans- duction Grammars..........................................................................................................................................286 Dekai Wu vi  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 A Music-generating System Based on Network Theory.................................................................................294 Shawn Bell, Liane Gabora Beyond the Fence Has computational creativity successfully made it «Beyond the Fence» in musical theatre?.......................303 Anna Jordanous The «Beyond the Fence» Musical and «Computer Says Show» Documentary...............................................311 Simon Colton, Maria Teresa Llano, Rose Hepworth, John Charnley,Catherine V. Gale, Archie Baron, François Pachet, Pierre Roy, Pablo Gervas, Nick Collins, Bob Sturm, Tillman Weyde, Daniel Wolff and James Robert Lloyd Blending Free Jazz in the Land of Algebraic Improvisation.........................................................................................322 Claudia Elena Chirita and J osé Luiz Fiadeiro An Argument-based Creative Assistant for Harmonic Blending.....................................................................330 Maximos Kaliakatsos-Papakostas, Roberto Confalonieri, Joseph Corneli, Asterios Zacharakis and Emilios Cambouropoulos A Process Model for Concept Invention.........................................................................................................338 Roberto Confalonieri, Enric Plaza and Marco Schorlemmer Optimality Principles in Computational Approaches to Conceptual Blending: Do We Need Them (at) All?...346 Pedro Martins, Senja Pollak, Tanja Urbancic and Amilcar Cardoso Learning to Blend Computer Game Levels.....................................................................................................354 Matthew Guzdial and Mark Riedl Software Platforms The FloWr Online Plat-form: Automated Programming and Computational Creativity as a Service..........363 John Charnley, Simon Colton, Maria Teresa Llano Rodriguez and Joseph Corneli Computational Creativity Infrastructure for Online Software Composition: A Conceptual Blending Use Case......371 Martin Znidarsic, Amilcar Cardoso, Pablo Gervas, Pedro Martins, Raquel Hervas, Ana Alves, Hugo Oliveira, Ping Xiao, Simo Linkola, Hannu Toivonen, Janez Kranjc and Nada Lavrac Dance CoChoreo: A Generative Feature in iDanceForms for Creating Novel Keyframe Animation for Choreography..380 Kristin Carlson, Philippe Pasquier, Herbert H. Tsang, Jordon Phillips, Thecla Schiphorst and Tom Calvert ROBODANZA: Live Performances of a Creative Dancing Humanoid..........................................................388 Ignazio Infantino, Agnese Augello, Adriano Manfré, Giovanni Pilato and Filippo Vella Interactive Augmented Reality for Dance......................................................................;................................396 Taylor Brockhoeft, Jennifer Petuch, James Bach, Emil Djerekarov, Margareta Ackerman and Gary Tyson vii Proceedings of the Seventh International Conference on Computational Creativity, June 2016  Proceedings of the Seventh International Conference on Computational Creativity, June 2016 Keynote Talk 2016 Homo Creativus: A psychological perspective Todd Lubart Biography Todd Lubart is Professor of Psychology at the Université Paris Des- cartes, and Member of the Institut Universitaire de France. He received his Ph.D. from Yale University and was an invited professor at the Pa- ris School of Management (ESCP). His research focuses on creativity, its identification and development in children and adults, the role of emotions, the creative process and intercultural issues. Todd Lubart is author or co-author of numerous books, research papers, and scientific reports on creativity, including the books Defying the crowd: Culti- vating creativity in a culture of conformity (NY: Free Press, 1995), Psychologie de la créativité (The psychology of creativity, Paris: Co- lin, 2003), and Enfants Exceptionnel (Exceptional Children, Bréal, 2006). He is the co-founder of the International Centre for Innovation in Education (ICIE), and the associate editor of Gifted and Talented International. Abstract What is creativity and what are its psychological underpinnings? More than a century of research in psy- chology provides an initial understanding of the definition of creativity, sources of individual differences, and ways to measure them. The “ingredients” of creativity including both cognitive and personality facts will be highlighted. Then the way these ingredients come into play during the creative process of producing ideas will be explored based on work in diverse domains, such as the fine arts, literary composition, design and engineering. The role of a favorable environment, including social and technological facets will be dis- cussed. Finally, work on the appreciation and uptake of creative productions in the field will be presented. viii  SEARCH  Novelty-Seeking Multi-Agent Systems Simo Linkola Department of Computer Science and HIIT University of Helsinki [email protected] Tapio Takala Department of Computer Science Aalto University School of Science [email protected] Hannu Toivonen Department of Computer Science and HIIT University of Helsinki [email protected] Abstract This paper considers novelty-seeking multi-agent sys- tems as a step towards more efficient generation of cre- ative artifacts. We describe a simple multi-agent archi- tecture where agents have limited resources and exer- cise self-criticism, veto power and voting to collectively regulate which artifacts are selected to the domain i.e., the cultural storage of the system. To overcome their in- dividual resource limitations, agents have a limited ac- cess to the artifacts already in the domain which they can use to guide their search for novel artifacts. Creating geometric images called spirographs as a case study, we show that novelty-seeking multi-agent sys- tems can be more productive in generating novel arti- facts than a single-agent or monolithic system. In par- ticular, veto power is in our case an effective collabora- tive decision-making strategy for enhancing novelty of domain artifacts, and self-criticism of agents can signif- icantly reduce the collaborative effort in decision mak- ing. Introduction Novelty is often considered a central component of creativ- ity (e.g. Boden (1992)). Obviously, an artifact that is not novel can hardly be considered creative. This paper studies the capability of cooperative multi-agent systems to seek and produce novel artifacts, and the effects of social decision- making strategies on this capability. Our focus is on seek- ing novelty; other aspects of creativity, such as surprise and value, are left for future work. According to the systems view of Csikszentmihalyi (1988), creative systems consist of three intertwined parts: individual agents, society and domain. A set of interacting agents forms a society. The domain is a cultural component constructed by the society by selecting artifacts worth pre- serving. Each part in the system is in constant interaction with other parts, e.g. individuals try to learn from the do- main and bring about transformations, while it is the society that collectively decides which transformations are valued and stored in the domain. In this work, we view the agent society as a whole, and consider the artifacts introduced to the domain as the end result of the agent population’s cultural knowledge of the artifact type. From this point of view, it is important that the agent society is capable of distributed self-regulation in controlling which artifacts are accepted to the domain. We examine how the number of agents, the amount of their collective resources and their access to the domain amalgamate with decision-making strategies of the society. Specifically, we are interested in how self-criticism, voting and veto power (the ability of individual agents to reject arti- facts) enhance the overall novelty of artifacts accepted to the domain. Further on, we study how much work the system has to do to produce a certain amount of domain artifacts. In our case study, we use simple agents that create spirographs. Our main contribution is the study of overall novelty of domain artifacts produced using different social decision- making strategies, especially self-regulation and veto power. This paper is structured as follows. After reviewing re- lated work in the next section, we describe the novelty- seeking agent architecture. We then illustrate and evaluate the architecture using spirographs as the artifacts. Related Work Multi-agent systems are a large research area (for an overview, see, e.g., Shoham and Leyton-Brown (2009)). Within the field, our work can be characterized as a system with multiple autonomous agents, where the agents diverge in information they possess (they each have a location and some memory) but not in their interests (they all aim to gen- erate novel artifacts). Further on, the agents are cooperative rather than competitive. The focus of this work is on cre- ativity of agent systems and more specifically on novelty- seeking agents. Next, we briefly review related work on creative agents; a more comprehensive overview can be ob- tained from the review of computational social creativity by Saunders and Bown (2015). We build our research upon existing work on creative and curious agents, especially work done by Saunders and Gero. Saunders and Gero (2001a) present a curious agent searching for novelty in the space of geometric images pro- duced by a spirograph. The agent learns a categorization of the produced images by showing them as input to a self- organized map, or SOM (Kohonen 1995). The novelty of a new image is computed as the pixel-wise deviation from the best matching cell’s image in the SOM. The agent’s curios- ity is modeled as a tendency to make smaller mutations in the generating parameters when more novelty is found. This 6 1 Proceedings of the Seventh International Conference on Computational Creativity, June 2016  helped the agent to concentrate on areas in the parameter space where more variability was found. In another experiment they let a society of agents seek novelty in images produced by genetic programming (Saun- ders and Gero 2001b). The agents have variable degrees of curiosity, modeled as a hedonic function that gets its maxi- mum at a certain level of novelty. The agents communicate through their creations, giving positive feedback to those ar- tifacts that match their hedonic function. Societal forma- tions, such as cliques, were found to emerge. We have adopted a similar approach, simulating a so- ciety of communicating agents that try to produce novel spirographs. However, we do not utilize the hedonic func- tion but seek only to maximize novelty. Moreover, the agents in our experiments do not learn a model, such as a SOM, of previously seen artifacts. Instead, they memorize a limited number of the encountered artifacts as they are. This is a simpler solution and also less sensitive to parameters of the model (e.g. those of SOM). Sosa and Gero (2005) have studied design as a social phe- nomenon with change agents (designers) and adopter agents (consumers). They conclude that emergent social phenom- ena — such as gatekeepers and opinion leaders — can stem from simple social mechanisms, and that the effect of an individual on a society depends both on the individual at- tributes and on the social structures. Gabora and Tseng (2014) have studied a society of agents capable of inventing and imitating ideas, and of realizing the ideas as actions. In their work, each agent has a set of limbs and the agents make actions by moving the limbs. Gabora and Tseng (2014) observe that societies where agents can chain simple actions to more complex ones obtain higher average fitness and that self-regulation increases the mean diversity of the actions. Finally, Lehman and Stanley (2008) introduce a novelty search where the main interest is not, per se, in satisfying certain objective goal. Instead, the aim is to find a diverse set of behaviors, i.e. behaviors that are novel enough with respect to other behaviors in the set. The search for an ex- panding set of novel behaviors often leads to a point where a fixed objective goal is also satisfied. Our work has a similar interest, a set of novel behaviors or artifacts, but we consider multi-agent systems without central control. Agent Architecture We now describe our architecture of a novelty-seeking agent system. The designs of individual agents and the society of agents have been kept as simple as possible. We make no claims of the novelty of the architecture; rather, our contri- bution is in the aim to maximize the diversity of artifacts created and the experimental results concerning factors be- hind the resulting diversity. We outline the big picture of the architecture first and then give the details. We have a society (population) S of homogeneous agents. Each agent Si ∈ S has a fixed amount of resources at its dis- posal, in particular a constant amount of individual memory; in other respects, the agents are identical. We model the behavior of the population via iterations: at each iteration, each agent creates a candidate artifact based on its current position and memory. Agents then proceed to collectively decide which of the candidate artifacts to add to the domain. In our model, the agents can be self-critical and choose not to present their own artifact as a potential candidate. They can also exercise veto power to reject other agents’ candidates. The agents are cooperative so self-criticism and especially the veto power are intended to be used for the benefit of the society, not of any individual agent. We will next more closely explain how individual agents function, and then how the multi-agent system operates as a whole. Individual Agents We consider agents that have a generative function produc- ing artifacts from one or more parameters. In our model (following Saunders and Gero (2001a)), the agents live in the generative function’s parameter space and can only ex- plore different artifacts by moving in the parameter space. Agents appreciate artifacts based on their novelty: the more novel the artifact is to the agent, the more it is ap- preciated. To this end, each agent has a limited memory of artifacts, and a function which can measure a distance be- tween any two artifacts. An agent can memorize artifacts it sees during the process to its memory. If the memory is full, memorizing a new artifact will erase the oldest one. An agent calculates the novelty of a new artifact as the minimum distance between the new artifact and any ar- tifact currently in the agent’s memory. More precisely, an agent Si with artifact memory Mi of size m, Mi = (A1, A2, . . . , Am), calculates the novelty Ni(A) of artifact A to be Ni(A) = min A′∈Mi d(A, A′), (1) where d(·) is the distance function. Pseudocode for the behavior of a single agent is given in Algorithm 1; details are given in the text below. Algorithm 1 Agent behavior during a single iteration 1: invent a new artifact close to the agent’s current location and move to the new location 2: if the new artifact passes self-criticism then 3: memorize the new artifact 4: publish the new artifact as a candidate for the domain 5: end if 6: participate in social decision making to select which artifact, among candidates published by all agents, is added to the domain 7: select and memorize artifacts from domain To invent a new artifact and to move to a new location (line 1), the agent considers a fixed number of possible new locations using random walk in the parameter space (called a search beam). For each possible location, it then considers the artifact produced by the respective parameter values and chooses the one with maximum novelty with respect to the agent’s own memory. It then moves to the corresponding position in the parameter space. 7 2 Proceedings of the Seventh International Conference on Computational Creativity, June 2016  In order to model self-criticism, agent Si has a novelty threshold si which it uses to determine if the created artifact is novel enough for its liking (line 2). If the created artifact passes the threshold, i.e. if Ni(A) ≥ si, the agent memo- rizes the artifact and also publishes it as a potential domain artifact candidate (lines 3–4). In a single agent setting, these published artifacts will create the domain on their own. Multi-Agent Architecture To keep our model simple, the multi-agent system runs with minimal agent-to-agent interaction. The interactions are done solely via generated artifacts and are twofold: (1) agents use collective decision making to select artifacts to the domain D, and (2) agents can examine and memorize current domain artifacts in D to guide their own search. In each iteration, domain artifact candidates are published by individual agents. The selection to the domain takes place in two phases (line 6). First, agents exercise veto power: any agent Si rejects any other agent’s artifact A whose calculated novelty is below a threshold vi, in a manner similar to self-criticism. Formally, given a set C of candidate artifacts, the set C∗ = {A ∈ C | ∀Si : Ni(A) ≥ vi} (2) of candidates survives to the next step. Second, agents vote on which remaining artifact in C∗ to add to the domain. (If C∗ is empty, none is added.) The vot- ing procedure considers the calculated novelties of artifacts in C∗, and the winner is the artifact A∗ which is considered on average most novel: A∗ = arg max A∈C∗ � 1 |S| � Si∈S Ni(A) � . (3) The artifact A∗ is then added to the domain D. Agents have access to the domain artifacts which they can examine and memorize (line 7). Memorizing an artifact will add it to the agent’s memory (and erase the oldest artifact from the memory if its full). In our model, agents have two means to explore domain artifacts: draw k artifacts at ran- dom or select the closest k artifacts in the parameter space. We will denote these domain artifact memorizing strategies as randomk and closestk. In both strategies the agent mem- orizes the artifacts blindly in the sense that a single artifact can appear multiple times in the agent’s memory. The domain is a set of artifacts, but for notational pur- poses we consider it as a temporally ordered sequence of artifacts D = (A1, A2, . . . , A∗). This allows us later to de- note all the artifacts in the domain up to the jth artifact by Dj = (A1, A2, . . . , Aj). Case study: Spirographs We illustrate the novelty-seeking agent architecture by gen- erating spirographs, a type of geometric images, like Saun- ders and Gero (2001a) did. While generation of a spirograph is a mechanistic process given the necessary parameters, finding parameter values that produce creative spirographs — in our case more specifically novel ones — is a non-trivial problem. Spirograph Spirograph is a toy used to draw epicyclic curved patterns with two interlocking gears of different sizes. A rotating gear (g) of radius r is positioned next to a fixed gear (G) of radius R such that the gear’s teeth interlock. A pen fixed to some point in g at distance ρ from the center draws a pattern when the gear is rotated. Points on the curve are given by equations x = (R ± r) cos(θ) + ρ cos(θ + t) (4) y = (R ± r) sin(θ) + ρ sin(θ + t) (5) where the sign of r determines whether g is exterior or inte- rior to G. θ is the rotation of g’s center around G, and t is the rotation of g self, given by t = θ(R − r)/r. (6) The pen’s movement is cyclic, returning to the starting point when both gears have made an integer number of rotations, i.e. when θ = 2πN/R, where N is the least common mul- tiple of r and R. Small N gives distinguishable calligraphic patterns, whereas shaded circular bands result when r/R tends towards irrational (N → ∞). A real physical spirograph is constrained by R > 0 and ρ < r, and r < R if g is inside G. In our experiment, we use an abstract computational toy, allowing any (real) values in the formula. Without loss of generality, R can be fixed and r, ρ defined relative to that. Values of ρ > r (meaning that the pen is outside of g) and ρ < 0 are also possible, though the latter only produces mirrored equivalents of positive val- ues (the pen is in a reversed position w.r.t. g’s center). Compared to Saunders and Gero (2001a) the main differ- ence is that we also let the pen radius ρ vary, giving us two parameters to mutate while traversing the search space. A Spirograph-Generating Agent We will now describe in detail how a spirograph-generating agent in our experiments behaves. As described above, we run our agents in a simulation where each agent is triggered to act on every iteration. Agents follow the procedure illus- trated in Algorithm 1 every time they act. Agents live in the 2-dimensional parameter space of spirographs, where the location of an agent is determined by its values for r and ρ. Each point (r, ρ) in the param- eter space corresponds to a single spirograph defined by r, ρ, and R = 200. Agents are initialized to start at ran- dom locations in the continuous parameter space by draw- ing the initial location (r, ρ) from the uniform distribution r, ρ ∼ U(−199, 199). Spirographs are first drawn as 500×500 greyscale images where gear G is located in the center. Because r can be negative (gear g is exterior to G), some areas of the param- eter space actually produce plain white images as the whole spirograph is drawn outside the image. To reduce the spirograph generation time, each spirograph is drawn with only 20 full rotations of gear g around gear G’s center. This has the effect that some spirographs are only drawn partially, but as neither the completeness of the spirographs nor the generating function is in the focus here, 8 3 Proceedings of the Seventh International Conference on Computational Creativity, June 2016  (a) Agent’s movement (b) Generated spirographs Figure 1: A single agent’s behavior, its movement in the 2- dimensional parameter space (1a) and generated spirographs (ordered left-to-right, top-to-bottom) (1b). it does not affect the experiments. Finally, to reduce evalu- ation time, spirographs are rescaled to 32×32 greyscale im- ages. For inventing a new spirograph, an agent located in a point (r, ρ) in the parameter space considers a fixed amount of new points around it. Each new point (r′, ρ′) is sam- pled from a two-dimensional normal distribution with r′ ∼ N(x, 8) and ρ′ ∼ N(ρ, 8), then both r and ρ are clamped to −199 ≤ r, ρ ≤ 199, and a spirograph corresponding to the point is created as described above. For each new spirograph, its novelty is calculated as in Equation 1, and the spirograph considered the most novel is selected. The difference d(·) between two images, used in the equation, is defined as the Euclidian distance between the 1024–element vectors formed from grey-scale values of each 32×32 image’s pixels. Although this does not fully correspond to perceptual distance between images, it tech- nically serves our purpose. Figure 1 illustrates a sample of 25 iterations of a single agent’s behavior, its movement in the parameter space and the spirographs it has created. Evaluation We next report on empirical evaluation of the proposed agent architecture using spirographs as the creative artifacts. The questions we aim to answer empirically are the fol- lowing. (1) How does the number of agents affect the nov- elty of artifacts produced to the domain? (2) What is the effect of the beam size on the performance? (3) How does self-criticism of agents affect the novelty, and what is the ef- fect of the veto power? (4) How does agents’ access to the domain affect novelty? We also study how these factors af- fect the rate at which artifacts are introduced to the domain. Experimental Setup Novelty can be difficult to define in many domains, and it obviously depends a lot on the back- ground. In the experiments of this paper, the novelty of each artifact added to the domain is measured in relation to the artifacts that the agent society has already added to the do- main. Such a measure allows comparison across different Simulation parameter Default value Target domain size, |D| 200 Number of agents, |S| 16 Self-criticism threshold, si 3.2 Veto power threshold, vi 3.2 Total search beam width 256 Total agent memory 512 Memorization strategy closest3 Table 1: Default parameter values for the experiments. systems that aim to produce novel artifacts of the same type, whether they are single-agent or multi-agent systems. Let Aj denote the artifact added to the domain D as its jth artifact. The novelty of Aj is measured as its distance to the nearest artifact already in the domain: N j(Aj) = min A′∈Dj−1 d(Aj, A′), (7) where Dj−1 is the set of artifacts in the domain before Aj is added to it. Further on, we define N 1(A1) = 0. Based on the novelty of individual artifacts in the domain, we define an aggregate measure as the average over all arti- facts’ novelties: N ∗(D) = 1 |D| − 1 � 2≤j≤|D| N j(Aj), (8) and use N ∗(D) to compare performance of different system configurations. In the experiments, we simulate the agent system until a fixed number (200) of artifacts has been accepted to the domain and compute their mean novelty N ∗ as the measure how novel the artifacts in the domain are on average. The effort needed to produce a given number of artifacts varies across different settings since the exercise of self- criticism and veto power can result in iterations with no can- didate artifacts at all. We therefore also study the number of iterations of the agent system needed to produce the arti- facts. Each agent has some resources, in particular a fixed amount of memory and a search beam (the number of lo- cations it considers per iteration). To make comparisons fair across different numbers of agents, the total amount of these resources in the society are kept constant when the number of agents varies. (There are other aspects that affect the computational complexity but they are ignored here. For instance, with the above division of a constant amount of memory across agents, a society consisting of a smaller number of agents makes a larger total number of comparisons between arti- facts in the search beams and the memory. On the other hand, a larger society spends more efforts on mutual evalua- tion, vetoing, and voting on candidate artifacts produced by the society.) The default parameter values of our experiments are listed in Table 1. The total search beam width and agent memory are divided equally to agents. 9 4 Proceedings of the Seventh International Conference on Computational Creativity, June 2016  (a) Novelty N ∗ (b) Iterations Figure 2: Effect of the number of agents on the novelty N ∗ (2a) and on the effort required to produce 200 novel artifacts to the domain (2b). Points at the right ends of the panels are for the baseline method Mono. Results We now report our experimental results with the above- described architecture of novelty-seeking agents. Population size The effect of population size on the over- all behavior of an agent system is of key interest. Ideally, a multi-agent system should have emergent properties that a single-agent system does not have while not introducing excessive overhead due to agent communication and coordi- nation. Figure 2 shows how the behavior of our multi-agent sys- tem is affected by the number of agents in the society. Dif- ferent lines show different search beam widths; for now, consider the shapes of the curves, we will return to a com- parison between them below. Panel 2a shows that the overall novelty N ∗ of artifacts added to the domain increases with the number of agents. This is a desired effect for an agent architecture and indicates that agent collaboration, in particular the selection of arti- facts to the domain works effectively. The effect is clearer with smaller beam widths (lower lines in the figure). Panel 2b complements the picture by showing the corre- sponding effort, expressed in terms of the number of itera- tions required to produce 200 novel artifacts to the domain. Here, we observe a less trivial behavior when the number of agents increases. First, the required effort drops until about 4 agents. This is explained by the fact that a larger number of agents can search a more diverse set of options. The required effort starts to increase, however, when the number of agents grows further. When the number of agents grows, the soci- ety also becomes collectively more critical about the novelty of candidate artifacts. In our case, some 16–32 agents seem to be the critical amount, but the exact amount is of course dependent on the application. The two panels of Figure 2 illustrate an inherent trade-off in systems like this: the more critical the society, the higher the novelty of its output is but smaller in size. Based on the figure, in our setting some 4–16 agents seem to give a good compromise between quality and efficiency. We next briefly compare the results of the multi-agent sys- tem to three different simple alternatives. First, a comparison to a single-agent system with other- wise similar functionality and identical resources (Figure 2, leftmost points of the lines) shows that as a rule, a multi- agent system produces more novelty and often in less time than a single agent. Second, an efficient and simple method to obtain 200 spirographs is to sample 200 random points uniformly from the parameter space. Artefacts produced this way have an average novelty of N ∗ = 1.14, markedly lower than the novelties obtained by agent systems with at least two partic- ipants (3.5–3.9). Third, consider a monolithic hybrid between the two base- lines above called “Mono”. Mono has no location in the pa- rameter space and so it does not use random walk. It in- stead samples points uniformly from the parameter space at each iteration and, like our agents, chooses the best of them at each iteration. The Mono system also exercises self-criticism/veto with the same threshold as the agents. In contrast to our agents, Mono has a complete memory of the domain artifacts and is maximally informed in that sense. A comparison to the novelty obtained by the Mono base- line (panel 2a, separate points at the right end of the panel) shows that from approximately four agents up, agent soci- eties are competitive with and even outperform the mono- lithic system with complete memory. At the same time, the agent system is more effective in producing the 200 artifacts, up to some 16 agents (panel 2b). Search beam width Let us now consider the different search beam widths in Figure 2. First, a comparison of the relative performances of differ- ent search beam widths gives the expected results: a wider search finds more novel results (2a) and does it more effec- tively (2b). Among the different beam widths, the narrower ones tend to be more interesting because a common assump- tion in multi-agent systems is that the agents are relatively simple and operate under severe resource constraints. In contrast, when the beam width grows without limit, agents start to have complete information about the search space. As already suggested above, different search beam widths behave differently when the number of agents is changed. As a rule, the number of agents has a larger effect when the search beam is narrow. This is natural, since with nar- row beams the individual agents are more constrained. A larger number of agents helps overcome the limitation and find more novel results (2a). On the other hand, when the number of agents becomes large, self-criticism and espe- cially the veto power hit the constrained agents harder and they need a longer time to find novel results (2b). Selection of candidates to the domain We now move on to consider how different methods to select candidates to the domain affect the behavior of the society. This is the central social aspect of our model: we model social interaction by 10 5 Proceedings of the Seventh International Conference on Computational Creativity, June 2016