A.I. In All The Wrong Places

Marc Böhlen,Ruolin Chen,Xiaoxu Dong,Srikar Gopaladinne,Hemanth Gorla,Divya Kandukuri,Sean Mansfield

2024-02-18

Abstract:This text describes experiences gained across a two-year test period during which two generations of Generative Artificial Intelligence (A.I.) systems were incorporated into an interdisciplinary, university level course on A.I. for art and design practices. The text uses the results from the courses to reflect on new opportunities for generative systems in art and design, while considering traps and limits.

Computers and Society

What problem does this paper attempt to address?

This paper describes the experience of integrating two generations of generative artificial intelligence (A.I.) systems into an interdisciplinary university art and design course during a two-year testing period. The author reflects on the new opportunities for generative systems in art and design through the course outcomes, while considering the pitfalls and limitations. The main focus of the paper is on the application of generative A.I. in education, particularly in the fields of art and design, discussing the potential, limitations, and how to better utilize these systems.

The article mentions several early generative design experiments, such as cellular automata and genetic algorithms, and notes that neural networks have gradually become the primary field of artistic experimentation. With the development of large-scale generative models such as the GPT series (GPT-2, GPT-3, and GPT-4) and the DALL-E image generator, these models have demonstrated powerful capabilities in text generation and image creation. However, they have also brought forth discussions regarding creative attribution, the ability to imitate human behavior, and the authenticity of generated content.

In the art and design course, students use these generative A.I. systems to create graphic novels and discover that while the models are capable of generating text, the coherence of the stories is lacking. With the introduction of new technologies such as GPT-3 and DALL-E2, the project shifted towards creating public websites to explore the application of generative A.I. in the public domain. The students eventually developed an A.I.-based recipe generator called "Neuro Recipe Box."

The paper mentions that although the new generation models perform better in certain aspects, they do not significantly surpass their predecessors in image generation, and the generated recipe instructions are sometimes not practical. The author emphasizes the importance of prompt engineering in the quality of generation and points out that although these systems demonstrate non-human innovation, their creativity and practicality still need improvement.

In conclusion, this paper aims to reflect on the impact of current generative A.I. in art and design education and propose potential improvements and scenarios to abandon for these systems.