"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

2024-03-26

Abstract: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.

Human-Computer Interaction,Artificial Intelligence

What problem does this paper attempt to address?

This paper discusses the low acceptance rate and low efficiency of human-computer collaboration in the use of artificial intelligence (AI) systems in natural science research, despite the widespread belief that AI can improve efficiency. Through in-depth interviews with 10 AI practitioners and scientists, the study reveals three key factors that hinder the effective adoption of AI systems:

1. Lack of proactive support during the initial stages of system usage, where experts face difficulties in installation and operation. These technical issues add pressure and divert attention from the research project.
2. The communication of system functionalities fails to align with the users' needs, leading experts to reject the use of AI due to their lack of understanding on how AI can adapt to their requirements.
3. Lack of predefined collaboration rules, resulting in misunderstandings and frustration during the collaboration process.

The paper proposes recommendations to improve the adoption of AI:

1. Provide active support to experts during the initial stages of system usage.
2. Communicate the functionalities of the system in a user-oriented manner.
3. Follow predefined collaboration rules.

The study also highlights that existing technology acceptance models such as TAM and UTAUT theories overlook the impact of practical operational challenges and workflow changes on expert adoption of AI systems. The paper calls for more attention to the technical background limitations of experts, as well as the emotional and practical consequences, emphasizing the need to consider these factors in human-computer interaction design to facilitate effective adoption and collaboration of artificial intelligence.