Unintended consequences in computational creativity can lead to better outcomes

@Joshua Woroniecki

Dominik Siemon

Breakages and unintended consequences are typical in human-AI collaboration. From the perspective of computational creativity, they can offer added value in the information work,  especially in creative processes, and therefore do not necessarily have to be addressed.

Computational creativity aims to develop computer programs and systems that can generate or recognise creative outputs, such as ideas, art, music, and stories. It involves using artificial intelligence, cognitive psychology, and other related disciplines to understand and replicate the natural creative process. 

Researchers in this field have developed a variety of techniques, including evolutionary algorithms, neural networks, and constraint-based approaches, to generate creative outputs or to evaluate the creativity of outputs generated by humans. A popular field of artificial intelligence is generative AI, which specifically focuses on creating new content, such as images, music, or text, by learning from existing data. 

However, one limitation that both humans and artificial intelligence have is their previous knowledge (i.e., the data that artificial intelligence is trained on) which limits their abilities to create something new. A common phenomenon with humans is that they often think with strong boundaries and limits and don't have an open solution space. The same applies to artificial intelligence and, more specifically, generative AI, as its limits are mainly based on the data it is trained on.

In one of my research papers, which I wrote during my PhD journey, I developed a so-called creativity support system, which allowed users to write down initial and unfinished ideas on certain topics. An intelligent algorithm then used natural language processing technologies to understand the idea written down by the user. Interfaces (APIs) of common social media platforms such as Facebook, Twitter, and Tumblr were queried by the system to find content that could further develop the idea. I called my algorithm "Alan" as it acted as an active partner in the creative process, commenting on the user's ideas based on the obtained social media data, similar to a generative AI system such as ChatGPT. In the process, mistakes were made from time to time, resulting in comments from Alan that were only remotely related to the idea. These breakages or functional errors produced comments that could still add value to the idea generation. In creative processes, this is called performing mental leaps or incorporating out-of-context content into the idea generation. This technique is also used in product design to generate completely new forms of design.

In summary, unintended consequences in computational creativity can lead to better or more creative outcomes because the computer can explore a wider range of possibilities than a human designer can on their own. Sometimes these unintended outputs have unexpected and creative characteristics.

Computational creativity and generative AI reshape our understanding of creativity by demonstrating that machines can generate content that rivals human-made works. As these technologies continue to advance, they will undoubtedly play an increasingly significant role in the creative industries, presenting both opportunities and challenges for artists, researchers, and society at large.