Re: Using artificial intelligence to assess clinicians’ communication skills.
Collaboration between engineering and the behavioural sciences is critical for the advancement of artificial intelligence in medicine
We read with interest the paper by Ryan et al.  which explores the potential applicability of artificial intelligence (AI); in particular, automatic speech analysis technology, in assessing and improving clinicians’ communication skills. The paper also explores the potential of speech analysis technology to identify patient distress, with the overall goal of improving healthcare interactions.
We would like to add some additional comments, informed by our current experience of a research collaboration between engineering and behavioural sciences, working together to conduct paralinguistic analysis of distress in families with a history of cancer.
Interdisciplinary Collaboration is Essential
The development of AI requires access to high-quality human data. As Ryan et al. explains, for AI systems to ‘learn’ they require ongoing high-quality data input (i.e. audio files of real speech). Behavioural scientists working in healthcare settings often accumulate hundreds of hours of audio-recorded interviews with patients, consumers and healthcare professionals collected in a research setting that, after securing ethical approval and participant consent, may be used for this purpose.
Researchers working to improve patient healthcare experiences may not be aware of the concurrent development of AI technologies that share their goals. Further, research groups developing AI technologies may not be aware of the extent of the high-quality human data available in behavioural research settings that may help to facilitate machine learning. Collaboration between these groups is key and can be mutually beneficial. We argue for increased access to funding schemes that support this kind of truly interdisciplinary collaboration.
The Development of a ‘Shared Language’
One of the key challenges of interdisciplinary collaboration is the process of defining and describing problems so that all groups have a shared, rather than parallel, understanding. We have found that this process requires patience, reflexivity, openness to learning, and a commitment to abandoning rigid adherence to the terminology and assumptions of one’s own field. Ultimately we have found the development of this ‘shared language’ to be one of the most rewarding aspects of our collaboration between engineering and behavioural sciences, and it will be interesting to witness the likely convergence of language in the AI technology-medicine space.
Stakeholder Engagement is Key to AI Uptake in Medicine
The development of AI technologies for medical settings also requires collaboration with potential end-users of the technology – healthcare professionals and patients. Early engagement with stakeholders is critical to allay understandable concerns.
Ryan et al. conclude their article by wondering whether medical professionals will be willing to accept AI technologies aiming to enhance their clinical communication into their practice. Medical professionals may be willing to accept AI if it can automate and improve efficiency and accuracy of disease diagnosis [2, 3]. However, applying AI to human interactions will come with new challenges. Key concerns for clinicians will likely include considerations of the implications of AI for patient confidentiality, apprehension about the accuracy/validity of the feedback, and a need for reassurance about how data from the analysis of their clinical performance will be used.
In sum, we agree with Ryan et al. that AI is likely to transform healthcare-patient interactions in the near future. The key to success will be ensuring that multidisciplinary collaboration is optimised and engagement with relevant stakeholders is done well, and early.
Brittany C. McGill (with Claire E. Wakefield, Vidhyasaharan Sethu & Julien Epps).
 Ryan P, Luz S, Albert P, Vogel C, Normand C, Elwyn G. Using artificial intelligence to assess clinicians' communication skills. BMJ. 2019 Jan 18;364:l161.
 Jha S, Topol EJ. Adapting to Artificial Intelligence: Radiologists and Pathologists as Information Specialists. JAMA. 2016;316(22):2353–2354. doi:10.1001/jama.2016.17438
 Yu, K. H., Zhang, C., Berry, G. J., Altman, R. B., Ré, C., Rubin, D. L., & Snyder, M. (2016). Predicting non-small cell lung cancer prognosis by fully automated microscopic pathology image features. Nature communications. 2016;7: 12474.
Competing interests: No competing interests