Future of Work: AI predicts what could and should be automated in healthcare

Using AI, Oxford researchers have developed a tool to help policymakers and organizational leaders harness the positive power of AI while reducing unwanted effects on jobs

Its first application has identified numerous specific work activities that are highly automatable using today’s technology and that healthcare practitioners actually desire to automate. On the other hand, there are many highly automatable activities that practitioners desire to continue doing themselves. Overall, healthcare practitioners desire a surprisingly high proportion of their own work activities to be fully automated.

From speech recognition to medical diagnosis, AI is having a life-changing impact on society. At the same time, there is strong public concern that AI and other automation technologies will lead to the undesirable replacement of human labor.

In healthcare in particular, automation technologies might lead to enormous progress in the quality and efficiency of products and services, but they may have unwanted consequences for personnel in the field.

How can automation technologies like AI best support hospital staff and independent healthcare practitioners, who are often at the edge of their capacity?

An Oxford study has, for the first time, captured empirical evidence of not only what healthcare work could be automated today, but also what should be automated. To this end, researchers have trained AI models on thousands of ratings provided by both healthcare practitioners and automation experts.

In the first part of the study, the ratings of the healthcare practitioners were used to develop a probabilistic machine learning model to robustly predict the desirability of automating individual healthcare work activities. The results represent the very first detailed quantitative empirical evidence of healthcare practitioners’ preferences regarding the automation of their own work activities.

“We found that healthcare practitioners desire a surprisingly high proportion of their own work activities to be fully automated. This is potentially due to their high workloads and time pressure,” says Dr Wolfgang Frühwirt, Co-Principal Investigator of the study, who works with the Oxford Machine Learning Research Group.

In the second part of the study, the potential of automating healthcare work activities was investigated. The researchers compared the results of an AI analysis based on the newly obtained automatability ratings of healthcare professionals to those of a previous study by the Oxford Machine Learning Research Group, which was based on ratings of experts in AI and machine learning.

“Our findings show that healthcare practitioners, people actually working in the field, are less optimistic than technical experts about the current automation potential of healthcare work. At the same time they would desire many of their work activities to be fully automated,” explains Dr Paul Duckworth, who is a postdoctoral research assistant at the Oxford Robotics Institute and Principal Investigator of the study.

To support governments and organizations with practical strategic advice on automation, the AI models for automatability and desirability of automation were combined.

“We are presenting a succinct four quadrant model, the Automatability-Desirability Matrix, based on our findings. It can be used to support policymakers and organizational leaders in developing practical strategies on how to harness the positive power of AI, while accompanying change and empowering stakeholders in a participatory fashion,” says Dr. Wolfgang Frühwirt.

To demonstrate the actionable insights that can be gained, below are highlighted four extreme examples of particular work activities – all based on the AI models’ predictions:

• A difficult to automate activity that practitioners strongly desire to automate is ‘Developing treatment plans for patients’.
• A highly automatable activity that practitioners strongly desire to automate is ‘Entering data into computers’. Until this study there has been a lot of speculation, but little empirical evidence about the desirability of automating clerical activities.
• A very hard to automate activity that healthcare practitioners don’t perceive as very desirable to automate is ‘Encouraging patients to develop life skills’.
• An example of an activity with an already high level of automatability that physicians strongly desire to continue doing themselves is ‘Operating on patients’. This might be in contrast with what patients want and why the team is planning to include the patient’s view in their AI-based models in the future.

Overall, there was a positive association between automatability and desirability of automation of work activities.

Precise quantitative estimates of all work activities in healthcare can be found in the recently published journal article, ‘Towards better healthcare: What could and should be automated?’, in Technological Forecasting and Social Change.

Dr Paul Duckworth concludes, “We hope to contribute to a more accurate understanding of the challenging field of healthcare automation, considering more of its stakeholders, with the ultimate goal of managing technology for better healthcare outcomes.”

Further information

Oxford and The Future of Work

Oxford is one of the world-leading centers for Future of Work research. Since an Oxford University study (Frey & Osborne, 2017) predicted that 47% of US jobs are at high risk of automation in the near future, Oxford has received immense global media attention on this topic. Only recently, Oxford has announced its biggest ever cash donation: A US billionaire donated £150m to set up a new institute researching the societal impact of AI.

Dr. Wolfgang Frühwirt is an Associate Member of the Oxford-Man Institute (University of Oxford, Department of Engineering Science) where he works with the Machine Learning Research Group under Professor Stephen Roberts. Wolfgang holds two PhDs in the healthcare domain and a master’s degree in business. Operating at the intersection of multiple complex domains, he combines technical and business know-how (Managing Partner at object a GmbH) with psychological expertise (Private Practice in Psychotherapy and Executive Coaching). His current research interests lie in the Future of Work and AI's impact on the human subject and society in general. He is also interested in probabilistic machine learning and applied neuroscience.

Dr Paul Duckworth is a Postdoctoral Researcher at the Oxford Robotics Institute (University of Oxford, Department of Engineering Science). Prior to joining, he was a Principal Researcher in the Machine Learning Research Group with Professor Mike Osborne. He received his PhD (2017) from the University of Leeds, working on machine learning for autonomous mobile robots. His current research interests lie in probabilistic machine learning, mobile robotics, and sequential decision making problems. He is also interested in the impact machine learning technologies have on society and the future of human work.