"It Looks Sexy but it's Wrong." Tensions in Creativity and Accuracy using genAI for Biomedical Visualization

Area of Science:

• Biomedical Visualization
• Artificial Intelligence
• Scientific Communication

Background:

• Generative AI (genAI) tools can easily create visuals for biomedical content.
• However, the underlying architecture of these tools may compromise the accuracy and trustworthiness of depicted information.

Purpose of the Study:

• To analyze workflows and tensions related to genAI use in biomedical visualization (BioMedVis).
• To understand practitioners' concerns and values regarding genAI for visual data representation.

Main Methods:

• Conducted 17 qualitative interviews with BioMedVis practitioners and researchers.
• Analyzed attitudes towards genAI, ranging from adoption to avoidance, across different workflow stages.

Main Results:

• BioMedVis experts utilize genAI at various workflow stages with diverse attitudes.
• Current genAI use presents both opportunities and pitfalls for visualization research.
• Findings contrast with prior predictions on genAI's role in the visualization pipeline.

Conclusions:

• Human intervention is essential for empathetic design and accurate assessment of scientific visuals.
• Prioritizing accuracy and trustworthiness in BioMedVis is vital, especially when public trust in science is low.
• The principle of 'first, do no harm' applies broadly to science communication.