Beyond the Desk: Barriers and Future Opportunities for AI to Assist Scientists in Embodied Physical Tasks

Paper Title

Beyond the Desk: Barriers and Future Opportunities for AI to Assist Scientists in Embodied Physical Tasks

Publication Info

• Topic area: AI applications in physical, embodied scientific workflows
• Keywords: AI in science, embodied cognition, laboratory workflows, field science, tacit knowledge, speculative design, human-centered AI, physical tasks, cognitive augmentation, scientific infrastructure

Background and Problem

• Problem / challenge: Current AI tools are primarily designed for desk-based knowledge work and fail to address the needs of scientists engaged in high-stakes, physical, and improvisational tasks in labs and field settings.
• Significance: Many critical scientific discoveries depend on embodied, hands-on work that involves tacit knowledge, physical coordination, and improvisation, which are poorly supported by existing AI systems.
• Motivation and related work: Prior studies have focused on AI’s role in computational and simulation-heavy tasks, such as data analysis, programming, and hypothesis generation. However, there is a lack of research on how AI can support the physically demanding and materially grounded aspects of scientific practice. This paper addresses this gap by exploring scientists’ perspectives on AI in lab and field environments.

Solution

• Proposed approach: Conducted on-site interviews and speculative design sessions with 12 scientific practitioners to identify barriers to AI adoption and envision future AI tools tailored to physical scientific tasks.
• Novelty:
  1. First study to examine AI adoption in embodied scientific workflows beyond desk-based tasks.
  2. Identification of three barriers to AI adoption in physical scientific work.
  3. Development of five speculative design archetypes for future AI tools tailored to lab and field settings.
  4. Reframing AI as infrastructure to support physical scientific reasoning rather than replacing human expertise.
• Procedure and key techniques:
  • Conducted situated interviews in labs and field sites to understand current AI usage, barriers, and needs.
  • Used speculative design activities to elicit participants’ visions for ideal AI tools.
  • Analyzed transcripts and sketches using thematic analysis to identify barriers and group speculative designs into archetypes.

Results

• Concrete findings:
  • Three barriers to AI adoption: (1) high-stakes experimental setups make AI errors too costly, (2) physical environments constrain AI tool accessibility, (3) AI lacks the tacit knowledge and contextual judgment of human scientists.
  • Five speculative design archetypes for future AI tools:
    1. AI as the lab’s collective knowledge keeper.
    2. AI as a distributed lab-wide task status monitor.
    3. AI as a real-time monitor for scientists’ cognitive and physical health.
    4. AI as a mobile scout for fieldwork.
    5. AI as a collaborator for hands-on physical chores.
• Advantage over baselines: Highlights the limitations of current AI tools in physical scientific work and proposes designs that integrate AI into embodied workflows, addressing gaps in memory, attention, and physical task support.
• Experiments / evaluation:
  • Interviews with 12 participants from diverse scientific fields (e.g., neuroscience, nuclear fusion, field robotics, biochemistry).
  • Speculative design sessions to generate future AI tool concepts.
  • Thematic analysis to identify barriers and organize speculative designs.
• Limitations and future work:
  • Limited to early-career scientists and specific fields; findings may not generalize to all scientific domains.
  • Participants were based in the U.S. and Western Europe, limiting geographical diversity.
  • Did not directly observe live experiments due to high-stakes nature of tasks.
  • Future work could explore AI adoption in other physically demanding domains (e.g., medicine, emergency response).

Summary

This study investigates how AI can support scientists in high-stakes, embodied lab and fieldwork, identifying three barriers to adoption: risk of AI errors, constrained physical environments, and AI’s lack of tacit knowledge. Through interviews and speculative design sessions with 12 scientific practitioners, the authors propose five archetypes for future AI tools, such as lab-wide monitors and mobile field scouts. The findings suggest reframing AI as infrastructure to support the physical conditions of scientific reasoning rather than replacing human expertise. These insights can guide the design of AI systems for other physically demanding domains.