Co-Creative Production: Investigating the Impacts of Generative AI Models on Creative Satisfaction

• Owen He, Computer Science, University of Delaware

• Matthew Mauriello, Computer Science, University of Delaware

Visual asset production requires significant allocation of resources towards time and monetary compensation for artists, as a result this leads to a limited number of usable quality pieces. In recent years, the emergence of vast data driven generative artificial intelligence (AI) models such as Dall-E and Midjourney restructured the graphics design industry. These models allowed for the creation of assets in an inexpensive and efficient manner. With a descriptive text prompt, the models draw a correlation to generate a canvas from scratch. While in other cases, they can improve existing assets. For example, Adobe Firefly allows for the expansion of image borders through generative filling of the blank space. By removing major parts in the process of producing art or visual assets, the reduction in creative expression experienced by users remains unknown. This work investigates the psychological impacts of utilizing generative AI models, by measuring a user’s creativity satisfaction given various levels of task load and methods of producing creative works. Two surveying tools enabled the quantifiability of task load and creativity, the NASA Task Load Index (TLX) and Creativity Support Index (CSI) respectively. The project implements a text to character design generative AI model as well as a manual configuration interface through various sliders, wheels, and selection widgets to modify attributes. Participants recruited from Amazon Mechanical Turk (AMT), a crowdsourcing platform, will receive a creation task in which they have access to either the generative AI model or the manual customizer. Changes in user’s creative satisfaction between these constraints can be noted by comparisons between their CSI values. In replacing the process of producing visual assets with generative AI models, it may result in a reduction in the level of perceived creative freedom and satisfaction experienced by users, unless more engaging avenues of interacting with these models are introduced.