cs.HCMar 19, 2026
SaveBeyond the Desk: Barriers and Future Opportunities for AI to Assist Scientists in Embodied Physical Tasks
Organizations: UC San Diego, La Jolla, CA, USA
Abstract
More scientists are now using AI, but prior studies have examined only how they use it 'at the desk' for computer-based work. However, given that scientific work often happens 'beyond the desk' at lab and field sites, we conducted the first study of how scientific practitioners use AI for embodied physical tasks. We interviewed 12 scientific practitioners doing hands-on lab and fieldwork in domains like nuclear fusion, primate cognition, and biochemistry, and found three barriers to AI adoption in these settings: 1) experimental setups are too high-stakes to risk AI errors, 2) constrained environments make it hard to use AI, and 3) AI cannot match the tacit knowledge of humans. Participants then developed speculative designs for future AI assistants to 1) monitor task status, 2) organize lab-wide knowledge, 3) monitor scientists' health, 4) do field scouting, 5) do hands-on chores. Our findings point toward AI as background infrastructure to support physical work rather than replacing human expertise.
Figures & tables
Figure 1. Via interviews with 12 scientific practitioners working in lab and field settings, we found three barriers to adopting AI for embodied physical tasks: (A) Neuroscientists like P1 have experimental setups (e.g., surgeries to implant sensors in rodent brains) that are too high-stakes to risk AI making errors. (B) The challenging environments of lab and field sites, like a biochemist’s clean-room lab bench with delicate equipment (P12), make it hard to access AI tools. (C) Scientists feel AI cannot match the tacit knowledge and contextual judgment of human experts like a field roboticist improvising materials to get their experimental rig working at a remote volcano site (P2). (Image credit: all illustrations were drawn by human artist Lauren Cheng without AI assistance.) Three illustrations labeled (A), (B), and (C) depict scientists working in a lab or in a field setting. Image (A) shows two scientists sitting at a table performing rodent brain surgery, with one appearing exhausted as the clock shows 8:00pm. In (B), a cellular biologist stands at a lab bench preparing reagents with a test tube in her left hand and a pipette in her right hand. She appears confused as she relies on scattered sticky notes and analog protocols. In (C), a scientist working at remote terrain near a volcano is stuck fixing an unresponsive robot with limited tools and no support. He holds the broken robot leg in his right hand as the quadrupedal robot lies next to him with tangled wires.
Figure 2. To find barriers and opportunities for AI adoption, we visited workplaces and performed on-site interviews with 12 scientific practitioners in a range of lab and field settings such as (A) behavioral neuroscience labs with surgery and bench areas (P1, P5), (B) field site for a primate behavioral scientist (P8), (C) cellular biochemistry lab (P12). Four images of various scientific settings are labeled with (A), (B), (C), and (D). Image (A) shows a behavioral neuroscience lab bench with a microscope, tools, and scattered notes for soldering a neurobiology implant device. In (B), multiple scientists in cleanroom suits stand around computers and machinery in a nuclear fusion laboratory. In (C), a macaque in a dry landscape interacts with an apparatus made out of acrylic panels. Image (D) provides a closer look at a lab bench, featuring many reagent bottles and sticky notes on the shelves and a notebook on the table top.
| ID | YoE | Gender | Field | Institute | Lab Description | Role | AI Use |
|---|---|---|---|---|---|---|---|
| P1 | 3 | M | Behavioral Neuroscience | R1 Public | Neural activity of rodent behavior | PhD student | Sporadic |
| P2 | 5 | M | Field Robotics | R1 Private | Robot locomotion and navigation dynamics | PhD student | Daily |
| P3 | 2 | F | Field Robotics | R1 Private | Robot locomotion and navigation dynamics | Undergrad researcher | Daily |
| P4 | 5 | M | Nuclear Fusion | Industry | Inertial fusion materials design and development | Scientist | Sporadic |
| P5 | 3 | M | Behavioral Neuroscience | R1 Public | Neural activity of rodent behavior | PhD student | Sporadic |
| P6 | 10 | M | Nuclear Fusion | Industry | Inertial fusion materials design and development | Scientist | Sporadic |
Table 1. We interviewed 12 scientific practitioners at research institutions across multiple fields about the barriers and opportunities for AI tools in their daily working environment. The ‘YoE‘ column indicates how many years of experience they have in their respective field. ‘R1 Public/Private’ refers to U.S. high research output (R1) universities ( , 2025 ) . ‘AI Use’ is self-reported frequency of AI usage.
Figure 3. Thematic maps summarizing (top) the three categories of barriers that discourage lab and field scientific practitioners from adopting AI, and (bottom) five speculative AI assistant archetypes organized into mid-level conceptual clusters. The figure two thematic maps, one titled ``Barriers that Discourage Lab and Field Scientific Practitioners from Adopting AI in their Workflows," and the other titled "Five Speculative Designs for Imagined Future AI Scientific Assistants.'' Each column represents a major theme and includes the relevant codes within.
Figure 4. Five archetypes that are representative of speculative designs that our 12 participants envisioned for how future AI could help lab and field scientists: (A) as the lab’s collective knowledge keeper, (B)-(C) distributed lab-wide task status monitor, (D) real-time monitor for scientists’ cognitive and physical health, (E) mobile scout for fieldwork, (F) collaborator for hands-on physical chores. (Image credit: all illustrations were drawn by human artist Lauren Cheng without AI assistance.) A larger illustration shows an overhead view of a lab space with a distributed AI system across multiple rooms labeled (A), (B), and (C), while three illustrations underneath labeled (D), (E), and (F) demonstrate scientists using AI to assist them in their work. In the chemical reagent room (A), a screen located on the lab bench wall houses relevant protocols, notes from colleagues, reminders, images, videos, observations, and warnings specific to the lab. A camera on the device is directed at the work surface. In room (B), cell incubators are highlighted to signify detection of a drop in temperature. In another room (C), a scientist focused on a separate task is alerted about the cell incubators by the screen at her lab bench and by an audio warning. Illustration (D) depicts a scientist wearing a cleanroom suit facing a wall-mounted screen displaying “scanning…,” health information, and suggested tasks. Next to them is a door with a biohazard symbol. In (E), a field scientist working in a remote area deploys stimuli to far away primates via a drone, monitoring their behavior on his device. Illustration (F) shows a woman directing a humanoid robot to pick up a fifty kilogram glass panel used in a muddy terrain tank in her robotics lab.