In April we wrapped up the second edition of the Creativity, AI and Labor reading group. The second cycle of 4 appointments covered topics around artificial aesthetics, mass image generation, creative testing paradigms, and the evolving relationship between human creativity and machine capabilities (to learn more about the first cycle, check out this blog post). Through close engagement with Lev Manovich and Emanuele Arielli’s Artificial Aesthetics alongside complementary works, our discussions interrogated fundamental assumptions about what constitutes art, creativity, and aesthetic judgment in an era of generative AI.
The reading group emerged from a recognition that current discourse around AI and creativity often operates within restrictive frameworks that privilege certain forms of creative expression while rendering others invisible or “non-machine-readable.” Rather than accepting binary distinctions between human and artificial creativity, our sessions sought to unpack the complex negotiations, adaptations, and co-evolutions occurring between users, models, and creative outputs.
Throughout these conversations, we consistently encountered what we termed “discontinuities” in generative workflows: the gaps between creator and creation, imagination and computation, linguistic input and algorithmic pattern matching. Rather than viewing these discontinuities as problems to be solved, we began to see them as productive sites for understanding the hybrid, distributed, and fundamentally collaborative nature of contemporary creative practice.
The summaries below capture the texture of our discussions and the evolving questions that guided our collective inquiry into the aesthetic, technical, and cultural dimensions of AI-mediated creativity.
Theme 1/January: Aesthetics and recognition
Readings:
- Manovich, L., &Arielli, E. (2024).
- Ch 1: 1. Even an AI Could Do That and
- Ch 2: Who is an “Artist” in the AI Era? Artificial Aesthetics.
We visited definition of aesthetics as it is modelled in current discussions of creative artificial intelligence, which values a restrictive romantic form of art perception and aesthetic (output) judgement based on binary metrics geared for mass generation. Authors provide a shallow engagement with the concept of aesthetics as just being limited to the sensorial, which bleeds into the tautological understanding of art that forms basis of much discussion in capabilities of computer-generated art. This means that there remain unrecognizable, uncategorizable, non-machine-readable “art,” which is left out of discussions about its automated reproduction and AI generated outputs. Perhaps, turning to questions of can AI do culture allow us to better the rich context that belie aesthetic judgement involved in visual generative media. Both the chapters offered a starting point to talk about perception, judgement, and taste in generative media.
Theme 2/March: Generating mass image/s
Readings:
- Manovich, L., &Arielli, E. (2024).
- Chapter 3: Techno-animism and the Pygmalion Effect
- Chapter 4: AI & Myths of Creativity. Artificial Aesthetics.
- Cubitt, S. (2020). Making the Mass Image. In Anecdotal Evidence. Oxford University Press. https://doi.org/10.1093/oso/9780190065713.003.0010
This session focused on the topic of generative images. Cubitt’s mass image aligned with the concepts of the operational images (Parikka, 2023) and others who devised a way to understand the mammoth of visual data generated and fed by and in social media posting. This framework helped bring forth the recognition, and extraction associated with the database, storage and capture of the “image” as data. However, it limits the image to a machine-readable version, as an effect and product for the model or infrastructure, and somehow beyond the (human/social) experience. This creates a fallacy that culture is simply supposed to be passively consumed, where artificial creativity serves and justifies the goals of prediction and imitation. Problematically, creativity comes to be externalized and mechanized.
Theme 3/April: Testing images
Readings:
- Manovich, L., &Arielli, E. (2024). Ch 5: From Representation to prediction. Artificial Aesthetics.
- Lesage, F., & Stewart, N. (2025). Testing Images. Working paper.
We mainly focused on the working draft by Fred and Nicole. Their theoretical paper seemed to offer two key contributions to the discussion of image generation. First, as test for fluency between the machine, (prompt engineered) input, and the judgement for the accomplished goal (output which is based on multiple iterations of the same). Second, the genealogy of testing for intelligence and creativity via image creation are interestingly intertwined. The discussions revealed a form of double-edged test and fluency, where both the user and the machine are tested/ judged for their ability to translate art to the desired output. In this process of training for skilled fluency (like prompt engineering), both the user and the model adapt, change, train and bound a specific type of creative artifact.
Theme 4/May: Wrapping up
Readings:
- Manovich, L., &Arielli, E. (2024).
- Ch 7: Separate and Reassemble: Generative AI Through the Lens of Art and Media Histories
- Ch 8: From Tools to Authors. Artificial Aesthetics.
We focused on arguments laid out by Arielli around skills as relevant to his argument about the extension of “bounded” creativity and aesthetics. For example, prompting AI tools to generate a creative output requires a dual learning: basic learning needed to describe the creative process (like language of cinematography or directorial instructions as needed to execute the ideation), and learning model’s prompt language structure or syntax. That is, AI tools nudge the user to develop new skills of prompt engineering, and training for aesthetic judgement throughout the different iterations of the generated media. However, thinking about creativity only in terms of its bounded or machinic/AI extensions, reinforces the idea of creativity as a uniquely human faculty. There remains multiplicity of discontinuities in the current workflow of visual generative models: from creator to creation; from imagination to bounded linguistic and computational aspects of the generative model; from the linguistic input to the binary/algorithmic form of pattern matching which leads to the visual output.
The plan for 2025-2026
The reading group is taking a break over the summer term as I transition to a new faculty position at Toronto Metropolitan University. I will reassess the situation in September, hoping to be in the position to reboot the group in Fall 2025 or January 2026. In the meantime, if you happen to be at 4S in Seattle in September, consider joining my session on precisely the issue of AI and creativity, tentatively scheduled for Thursday, September 04, 1:30 PM – 5:20 PM (Room: Second Floor: Redwood B).