PaperlayerView exact revision on arXiv

Source-linked AI summary

"It is there, and you need it, so why do you not use it?" Achieving better adoption of AI systems by domain experts, in the case study of natural science research

Auste Simkute, Ewa Luger, Michael Evans, Rhianne Jones

arXiv:2403.16895v1cs.HCcs.AI

TL;DR

Practical barriers and misinterpreted expert needs remain underexplored in AI adoption research, despite the importance of AI for natural science work. The paper compares practitioners’ and experts’ accounts, finding barriers in communication and support; it argues that relevant capability communication and collaborative support can help balance system benefits against required effort.

  • Problem

    Practical barriers, misinterpreted expert needs, and training needs are underexplored, even though manual processing can leave valuable data unused and slow or halt scientific discoveries.

    Source: §1 · p. 2

  • Method

    The paper compares practitioners’ and experts’ accounts of AI adoption barriers, frustrations, and collaboration.

    Source: §4 · p. 6

  • Results

    Miscommunication, inadequate practical support, and irrelevant capability explanations create adoption barriers; clear communication and collaborative support can balance perceived benefits against required effort.

    Source: §1 · p. 2, §5 · p. 11

  • Takeaways & Limitations

    Guiding users through practical tasks, providing active support, and establishing ongoing collaboration are presented as steps likely to increase AI uptake.

    Source: §6 · p. 14

    The study covers one domain and two teams in one organisation; recommendations need testing across domains with varying risk, time pressure, and reliance on AI.

    Source: §5.5 · p. 14

Abstract

from arXiv · show

Artificial Intelligence (AI) is becoming ubiquitous in domains such as medicine and natural science research. However, when AI systems are implemented in practice, domain experts often refuse them. Low acceptance hinders effective human-AI collaboration, even when it is essential for progress. In natural science research, scientists' ineffective use of AI-enabled systems can impede them from analysing their data and advancing their research. We conducted an ethnographically informed study of 10 in-depth interviews with AI practitioners and natural scientists at the organisation facing low adoption of algorithmic systems. Results were consolidated into recommendations for better AI adoption: i) actively supporting experts during the initial stages of system use, ii) communicating the capabilities of a system in a user-relevant way, and iii) following predefined collaboration rules. We discuss the broader implications of our findings and expand on how our proposed requirements could support practitioners and experts across domains.

1 INTRODUCTION

Miscommunication and unmet support needs create barriers to scientists’ adoption of AI systems, despite the systems’ potential value for handling growing data. The paper examines these barriers and proposes support, relevant capability communication, and collaboration rules as requirements for adoption.

1 INTRODUCTION

  • Scientists constrained to manual data processing may leave valuable data unused, slowing or halting scientific discoveries.This concern arises as data volumes grow and AI systems remain difficult to adopt or use effectively.Source: §1 · p. 2
  • Existing work addresses usability, accuracy, explainability, and transparency, but experts’ needs, practical barriers, and training needs remain underexplored.The paper identifies limited research on how needs are misinterpreted and on emotional, contextual, and collaborative obstacles to adoption.Source: §1 · p. 2
  • The study examines practical and contextual barriers, practitioners’ perceptions and responses, mismatched assumptions, and how collaboration relates to experts’ motivation to use AI.It reports an ethnographically informed study of 10 interviews with AI developers and natural scientists at one organisation facing slow adoption and data-processing bottlenecks.Source: §1 · p. 2
  • Miscommunications and frustrations between practitioners and experts create barriers to AI adoption, while limited support can burden experts and distract them from their projects.Practitioners may overestimate scientists’ technical knowledge, and experts may reject unfamiliar systems when their capabilities are not explained in relation to the project.Source: §1 · p. 2
  • The proposed AI acceptance requirements are initial support, communicating system capabilities in user-relevant terms, and predefined collaboration rules.The paper also recommends effective workplace collaboration and suggests that the findings may inform practice beyond the case organisation.Source: §1 · p. 3

2 RELATED WORK

Prior work describes workflow disruption and algorithmic aversion, while practical adoption barriers and the effort experts face before and during AI use receive less attention. The paper positions its focus on those hurdles alongside existing adoption theories.

2.1 AI in experts’ workflows

  • AI can disrupt expert workflows, increase workload and frustration, and contribute to loss of control or disengagement from preferred methods.Experts may adapt their work to ill-fitting technologies, requiring substantial time and cognitive effort.Source: §2.1 · p. 3

2.2 AI adoption

  • Algorithmic aversion and adoption barriers are shaped by system features and individual or contextual factors, while organisational studies identify environmental, organisational, and technical obstacles.Prior examples include concerns about transparency, agency, clinical value, performance, complexity, and technical skills.Source: §2.2 · pp. 3–4

2.3 Theories of technology acceptance

  • TAM frames adoption around perceived usefulness and ease of use, while UTAUT also considers social influence; practical hurdles and workflow changes remain underexplored in AI-expert interaction research.The paper argues that perceived benefits and required effort before and during system interaction warrant attention.Source: §2.3 · p. 4

3 METHOD

The study examines AI adoption in a natural science research centre where scientists and in-house practitioners work in the same organisation. It uses contextual interviews and grounded-theory analysis, with the first author’s organisational immersion informing the interpretation.

3.1 Procedure and Participants

  • The case was a large natural science research centre where in-house developers made AI-enabled tools for scientists, whose data volumes exceeded manual processing capacity despite low adoption.The setting was selected because experts were new to relying on AI and some projects required practitioner-expert collaboration.Source: §3.1 · p. 4

3.2 Data collection

  • The researchers interviewed five practitioners and five expert scientists using contextual enquiry, semi-structured questions, and demonstrations of software use.Interviews covered participants’ work practices, experiences with AI systems, and collaboration; follow-up questions were shaped by issues raised in early interviews.Source: §3.2 · p. 5

3.3 Data analysis and positionality

  • The first author transcribed and analysed interviews using grounded-theory coding, drawing on four days of workplace immersion for context.The author acknowledged that their experiences and positionality could have influenced data collection and analysis.Source: §3.3 · p. 5

4 RESULTS

Experts encountered practical barriers when starting unfamiliar software, while ongoing collaboration helped teams understand users’ needs, communicate software capabilities, and align expectations.

4.1 Using unfamiliar software

  • Experts struggled with command-line use and stressful installation, spending time on setup instead of research and leaving biological data unanalyzed.Some experts said software issues took more time than their projects and that they had been unable to analyze data containing biological information.Source: 4.1.1Barrier 1: Not knowing how to get… · p. 6

4.1.2Practitioners’ assumption: “They must know this much.”

4.2 Effective collaboration and communication as a solution

5 DISCUSSION

The discussion identifies practical, human-level barriers to AI adoption that can persist beyond system capabilities. It recommends reducing initial effort, explaining user-relevant benefits, and planning ongoing collaboration.

5 DISCUSSION

  • Adoption barriers are practical and human-level, often rooted in inaccurate assumptions about users’ needs rather than software capabilities.The authors argue that these barriers must be addressed before experts can explore and benefit from system features.Source: §5 · p. 10

5.1 Barriers to AI adoptions during the introductory period

  • Experts may reject systems after unsupported installation or running problems; the authors recommend assigned help, clear documentation, time to adjust, and practice.The proposed initial support also includes feedback between experts and practitioners.Source: §5.1 · p. 11, 5.1.1Recommendation: Offer active… · p. 11

5.2 Barriers to investing time to learn and explore AI systems

  • Experts were willing to invest time when they could see a system’s project-specific benefit, but practitioners often presented capabilities in technical rather than relevant terms.The authors recommend domain-specific explanations and demonstrations using similar projects or datasets.Source: §5.2 · p. 12, 5.2.1Recommendation: Clearly… · p. 12

5.3 Barriers to AI adoption due to effective collaboration between experts and practitioners

  • Planned, continuous collaboration can align expectations and reduce frustration, while early user involvement can build confidence and encourage testing and feedback.The authors recommend agreeing on collaboration type, frequency, timeframe, and goals for different stages.Source: §5.3 · p. 12, 5.3.1Recommendation: Plan the… · p. 13

5.4 Further Implications

  • Practitioners cite limited time and resources, yet a technology-first approach can leave software unused or abandoned; the authors argue early support can reduce later troubleshooting.The discussion also links adoption pressure and inadequate preparation with emotional burden and less time for experts’ projects.Source: §5.4 · p. 14

5.5 Limitations and Future Directions

  • The study involved one domain and two teams in the same organisation and environment; testing the recommendations across different risk and time-pressure settings remains necessary.The authors also call for testing across settings where experts vary in how much they rely on AI.Source: §5.5 · p. 14

6 CONCLUSION

The conclusion argues that practical workflow hurdles receive less attention than users’ responses to AI outputs or software features. It proposes active support, relevant capability communication, and predefined collaboration rules to improve adoption and contextual fit.

6 CONCLUSION

  • Active support, user-relevant capability communication, and predefined collaboration rules could increase experts’ motivation, confidence, and ability to use AI responsibly.The authors also connect these recommendations to participatory design and improving technology’s contextual fit and usability.Source: §6 · p. 14
Loading 2403.16895v1…