Can AI Be Creative? A Quest for Refining the Question

MALENA LEÓN

Lisboa

Workshop; III. A.I. Aesthetics workshop; 2023

Centro de Filosofia das Ciências da Universidade de Lisboa; Fundação para a Ciência e a Tecnologia (Portugal)

The question of whether a computational AI algorithm can be creative depends on how creativity is conceived. In my view, there are approaches to defining creativity that make the question of AI's capacity for creativity uninteresting. In recent decades, there has been a broad consensus in the literature regarding a bipartite definition of creativity, according to which creative products are, on the one hand, novel or original, and on the other hand, valuable: they excel in their field, are relevant, and functional (Kaufman and Glăveanu, 2019; Sawyer, 2012). Some scholars argue that a third component should be added to this definition: creative products should result from processes carried out by agents who act with understanding and purpose (Gaut, 2003, 2010; Carruthers, 2006; Davies, 2009; Kieran, 2014; Paul and Stokes, 2018). From my perspective, adopting this agent-focused position makes the question of AI's creativity uninteresting, as this perspective leads to a resounding "no'' as an answer, regardless of the actual capabilities of the programs in each case.From an opposing perspective, it could be argued that creativity is a property of the products (Briskman, 2009), and all products that meet the bipartite definition are considered creative. According to this view, if a program produces a result that passes the creativity "Turing test," we should answer affirmatively to the question of whether AI algorithms can be creative (as Dennett, 2001, seems to suggest). However, I believe these approaches are uninteresting as they do not help us better understand the phenomenon of AI programs producing seemingly creative results.In my opinion, a more intriguing perspective emerges from Margaret Boden's conception of creativity, which considers creative products to be original, valuable, and surprising (Boden, 2004). The latter element is crucial because it implies that it is not merely a result that has not occurred before but rather one that could not have occurred before. Boden examines the involved process and the rules that were being followed to generate the product. This approach allows us to evaluate whether a system is capable of transforming its own rules and assessing the appropriateness of these transformations. I intend to argue that this type of approach provides a more explanatory and interesting way to advance our understanding of the phenomenon of computational creativity.