The 'Bisociation Engine' (bEngine) is a collaborative, interdisciplinary project that attempts to computationally model specific aspects of human creativity. Rather than employing top-down processes such as propositional logic, the bEngine takes a generative approach that begins with the recognition of micro-level semantic, linguistic & structural associations between lexical items, then recursively assembles these into larger units of meaning. Arthur Koestler first coined the term 'bisociation' to distinguish between 'routine thinking' which occurs on a single plane, and 'the creative act' which, he states, 'always operates on more than one plane.' This paper presents a range of attempts by the authors to reverse-engineer 'bisociative' creative thought in software. More specifically it addresses difficulties and potential solutions for translating these complex parallel processes into code.

A particular focus of the 'Bisociation Engine' project thus far has been the human capacity for association, specifically between disparate areas of experience. An initial output of this research is the generative installation entitled 'the Architecture of Association' (AoA). Implemented as a sculptural grouping of 100+ suspended LCD screens, the AoA draws associative links between elements of a large multimedia database containing text, images, and video from the history of computation. bEngine algorithms are employed to 'intelligently' recognize semantic, linguistic and structural relationships between these database elements in real-time. These relationships (and their relative strengths) are used to situate media items in physical/architectural space, creating an evolving recombinant collage rich in associative potential.

dhowe@mrl.nyu.edu
Daniel Howe
Media Research Lab

bseaman@risd.edu
Bill Seaman
RISD Digital+Media

keywords: creativity, bisociation, association, artificial intelligence

While aesthetic practices in photography, film and music have undergone massive transformation due to affordances provided by computational tools, the practice of creative and critical writing has remained largely unaffected (word processors aside). This paper explores the potential impact of algorithmic tools on the writing process and consequently on the way in which we must read and analyze computationally inflected texts.

As 'source' material for our investigations we use a series of 'digital texts' created by Brown University undergrad and graduate students enrolled in the Spring 2007 'Electronic-Writing' Workshop. Exploring the results of three exercises in generative writing (recombination, context-free grammars and Markov-chains) we focus both on our own observations of the output and students' descriptions of the use and impact of computational techniques on their own writing, reading and critical practices.

Since the works in question are open in multiple senses (publicly available as web applets, 'template' files, and source-code), our analysis proceeds at multiple levels: from the surface text, to the intermediate 'grammars' employed, to the program code, to the multiple layers of software and hardware that constitute our experience of the text.

Our paper examines generative literary practice from three perspectives: as affordances and tools for practicing writers; as a pedagogical strategy for teaching procedural practices to humanities students; and as an emergent 'text' for critical interpretation of contemporary (digitally-encoded) literary practice.

anton_soderman@brown.edu
Braxton Soderman
Brown University

dhowe@mrl.nyu.edu
Daniel C. Howe
Media Research Lab

keywords: generative literature, e-writing, code, pedagogy, aesthetics, algorithm