Criticism, Culture, and the Automatic Generation of

Researchers wishing to create computational systems that themselves generate artworks face two interacting challenges. The first is that the standards by which artistic output is judged are notoriously difficult to quantify. The larger AI community is currently involved in a rich internal dialogue on methodological issues, standards, and rigor, and hence murkiness with regard to the assessment of output must be faced squarely. The second challenge is that any artwork exists within an extraordinarily rich cultural and historical context, and it is rare that an artist who is ignorant of this context will produce acceptable works. In this paper we assert that these considerations argue for case-based AI/Art systems that take critical criteria as parameters. We describe an example system that produces new bebop jazz melodies from a case-base of melodies, using genetic programming techniques and a fitness function based on user-provided critical criteria. We discuss the role that such techniques may play in future work on AI and the arts. Introduction: Constructing Artists Applications of computers to the arts date from the earliest days of computing. The use of AI technologies in the arts has a long history as well, particularly in music (Balaban et al. 1992). The majority of these uses fall into two categories: systems that perform “art understanding” tasks of some sort (e.g., music analysis systems), and systems that function as “intelligent” tools for use by human artists (e.g., (Rowe 1993)). Recently, however, a new category of systems has begun to emerge; a category of systems that are designed to be artists. By this we mean that such systems, which we will call “constructed artists,” are supposed to be capable of creating aesthetically meritorious artworks on their own, with minimal human intervention. Harold Cohen’s Aaron system is an early example of this category, and one of its few clear successes to date (McCorduck 1991). Aaron is a system that creates original drawings, each unique and potentially surprising to Cohen, that have iThe authors acknowledge the support of the Dorothy and Jerome Lemelson National Program in Invention, Innovation, and Creativity. *An experiment combining public assessment with genetic techniques similar to those described later in this paper is currently in progress via mosaic on the internet. The address is: http:Nporsche.boltz.cs.cmu.edu:SOOl/htbin/mjwgenfo~. been exhibited in galleries internationally. Aaron was constructed through a laborious process of “tutoring” by Cohen, himself an accomplished painter, that spanned over a decade. More recently, work has proceeded on constructed artists that function as poets (Kurzweil 1990) music composers (Ames & Domino 1992), and aesthetic agents in virtual worlds (Bates 1992). A literature has also emerged on the computational underpinnings of artistic creativity more generally (e.g., (Boden 1991)). Aesthetic Judgements The philosophy of art, which in the Western tradition dates at least from Plato, has never been an area of widespread agreement (see, e.g., (Dickie & Sclafani 1977)). The range of theories regarding the bases of aesthetic value, judgement and criticism is extraordinary, and the debates show no signs of near-term resolution. This presents a problem for AI scientists wishing to produce computational artists: How do we know when we’ve got one? How do we know if version A is better than version B, or vice versa? Without the ability to answer such questions the science of artist construction cannot proceed, and these questions seem to be inseparably linked to the murky issues of aesthetic judgement. The larger AI community is currently involved in a dialogue on methodological issues, standards, and rigor; many are calling for the adoption of experimental methods from more traditional sciences, for the use of standard examples and criteria of assessment, etc. If those of us working on constructed artists cannot judge our systems without first resolving all of the open questions regarding the judgement of artworks, then we will be on shaky methodological ground indeed. Fortunately, it is possible to separate the two kinds of judgement; we describe one approach to doing so below. The artworks of Cohen’s Aaron have been judged by the artworld and by the museum-going public. According to some theories of art this is the best, or even the only, form of assessment by which to judge the quality of a work (Danto 1978). But this sort of judgement has a high price both in terms of human resources and in terms of time. The science of artist construction will proceed quite slowly if each iteration of each system can be assessed only by organizing a public show and by waiting for critical reviews.2 Of course, Cohen himself also served as a critic of Aaron’s performance, and he was presumably able to apply the results of his judgements to the improvement of the program Art 3 From: AAAI-94 Proceedings. Copyright © 1994, AAAI (www.aaai.org). All rights reserved.