Machine Learning Processes as Sources of Ambiguity: Insights from AI Art

Core Concepts

AI artists engage with the ML process to create ambiguity, challenging traditional HCI design perspectives.

Abstract

The paper explores how AI artists incorporate Machine Learning (ML) in their creative work to create ambiguity. By analyzing nine AI artworks, it reveals how artists manipulate the ML process, including dataset curation, model training, and application, to evoke ambiguity. The analysis highlights the importance of considering ambiguity of process alongside traditional forms of ambiguity in understanding AI art. The findings suggest a shift towards a process-centered conceptualization of ML as a design material in HCI.

Stats

"Ongoing efforts to turn Machine Learning (ML) into a design material have encountered limited success." "Our analysis shows that the current conceptualization of ML as a design material needs reframing." "The main contributions include revealing how AI artists engage with the entire ML process and identifying new types of ambiguity relevant to ML experiences." "Artists worked closely with the ML process and developed techniques to evoke ambiguity in novel ways." "The paper challenges widely accepted desiderata of ML in HCI and advocates for a process-centered perspective."

Quotes

"Artifacts using ambiguity support multiple interpretations by users." "Our investigation indicates alternative approaches to desiderata of ML in HCI." "The artists work closely with the entire ML process to foster rich interpretations by audiences." "Ambiguity can be engaged with when dealing with ML in artistic practices." "The wider implications reflect on commonly held assumptions about ML uncertainty and dependability."

Key Insights Distilled From

Machine Learning Processes as Sources of Ambiguity

by Chri... at arxiv.org 03-15-2024

https://arxiv.org/pdf/2403.09374.pdf

Deeper Inquiries

How can embracing errors and uncertainties in AI systems enhance user experiences?

Embracing errors and uncertainties in AI systems can actually enhance user experiences in several ways. Firstly, by acknowledging and highlighting the limitations of AI systems, users are encouraged to develop a more nuanced understanding of how these systems work. This transparency can lead to increased trust and confidence in the technology as users become more aware of its capabilities and shortcomings. Furthermore, embracing errors and uncertainties can also foster creativity and innovation. By intentionally introducing glitches or imperfections into AI-generated outputs, artists and designers can create unique aesthetic experiences that challenge traditional notions of perfection. This approach allows for the exploration of new artistic possibilities that may not have been considered otherwise. Additionally, by exposing users to the inherent uncertainty of AI systems, they are prompted to engage more critically with the technology. Users may be encouraged to question assumptions about AI's infallibility and consider alternative perspectives on how these systems operate. This critical engagement can lead to a deeper appreciation for the complexities involved in developing and using AI technologies. In summary, embracing errors and uncertainties in AI systems has the potential to enrich user experiences by promoting transparency, fostering creativity, encouraging critical thinking, and ultimately enhancing overall engagement with artificial intelligence technologies.

What are the implications of recognizing ambiguity of process for HCI design practices?

Recognizing ambiguity of process within Human-Computer Interaction (HCI) design practices has significant implications for how designers approach working with machine learning (ML) technologies. By acknowledging that ML processes themselves contain elements of ambiguity - such as dataset curation choices, model training decisions, or application contexts - designers are prompted to consider these factors as integral parts of their design considerations. One key implication is that HCI designers need to shift their focus from solely designing around tangible outcomes or artifacts produced by ML models towards considering how these models were created. Understanding how artists creatively use ML processes within artworks highlights opportunities for incorporating similar techniques into interactive designs. Moreover, recognizing ambiguity of process emphasizes the importance of collaboration between technical teams responsible for developing ML models and design teams tasked with creating user interfaces or experiences. Designers must work closely with engineers throughout all stages of an ML project to ensure alignment between technical decisions made during data curation/model training/application phases and desired user interactions. Overall, integrating an awareness of ambiguity within ML processes into HCI design practices encourages a holistic approach where both technical intricacies and creative decision-making play essential roles in shaping meaningful interactions between users and intelligent systems.