Perception of AI-generated art: text analysis of online discussions

In this work we analyze comments on three subreddits related to AI-generated art to understand how people perceive the ability of AI to create art and the topics and moods of discussions in the context of widespread usage of pre-trained models. We used computational text analysis techniques such as LDA topic modeling and sentiment analysis with sentiment lexicons. As a result, we find that discussions on technical topics and descriptions of AI-generated art were mainly positive, while discussions on socio-cultural issues were mainly negative and took place in a subreddit focused on defending AI art. The findings suggest that Reddit users are interested in both the artistic and socio-cultural implications of AI-generated art, finding it risky and questionable.