Ambient Artificial Intelligence Scribes in Ambulatory Care: Patient Acceptability and Trust

BACKGROUND: Ambient artificial intelligence (AI) scribes may improve clinician-level outcomes (e.g., documentation burden) and patient experience. However, patients must agree to allow clinicians to use ambient AI scribes for these benefits to materialize. What motivates patients to agree to these tools' use and what factors improve their acceptability remain underexplored. It is also unclear whether patients' current trust in AI tools influences which factors affect ambient AI scribe acceptability.

OBJECTIVES: This secondary, convergent mixed-methods analysis had a quantitative aim of describing patients' trust in AI tools, a qualitative aim of exploring correlates of ambient AI scribe acceptability, and an integrated aim of exploring whether factors associated with acceptability varied across patients with differing levels of general trust in AI tools.

METHODS: Survey and interview data originally collected to assess the patient experience after implementing ambient AI scribes were re-analyzed (n = 20). For survey data, we descriptively summarized the trust in AI items and categorized patients into low, moderate, and strong trust groups. For interview data, we used deductive (guided by the Theoretical Framework of Acceptability) and inductive coding to draw themes pertaining to acceptability. For data integration, we qualitatively examined differences in acceptability themes by level of trust in AI.

RESULTS: In our sample, patients' trust in AI ranged from low to high, and most patients (60%) had moderate trust in AI. Ambient AI scribe acceptability was generally influenced by minimal ethicality concerns, supportive affective attitudes, minimal patient burden, substantial perceived or realized benefits, and strong intervention coherence. There were minimal differences in acceptability themes by level of trust in AI. Patients recommended developing robust patient education initiatives and, to allow patients to make informed decisions, providing advance notice that these tools will be used.

CONCLUSION: These findings highlight patients' suggested strategies for improving ambient AI scribe acceptability. Our findings also generate hypotheses for future research examining the explanatory processes that underlie patients' perceived acceptability of ambient AI scribes.