The ethics of using biased AI in infectious disease outbreaks

Artificial intelligence (AI) is already being used in healthcare to increase efficiency and sometimes accuracy of clinical diagnosis, risk prediction, and administrative tasks. There are also a wide variety of possible uses of AI in the public health context, particularly when preparing for and responding to pandemics. Generative AI may be used to conduct contact-tracing and deliver personalised public health guidance, through tracing back interactions with infected individuals; neural networks may be used to analyse pathogen genetics and assist in the design of vaccines and therapeutics; and agent-based scenario modelling may be used to forecast the shape of an outbreak and recommend non-pharmaceutical interventions to mitigate effects among the most vulnerable populations. But the current pace of AI development means there is great uncertainty over what will be possible to do with AI if, and when, we face the next pandemic. 

But any use of AI will pose risks alongside potential benefits. For instance, biases in training data mean that machine learning (ML) models may make more accurate predictions regarding risk of developing diseases, projections of outbreaks, etc., for some individuals or groups rather than others. What should we do with various uses of AI in a pandemic? What are the parameters for ethically appropriate use, and potential specific good practices or use cases? And how might these issues vary across LMIC and HIC contexts? This project will address these questions through a series of transdisciplinary workshops, involving both technical and ethical AI experts, in order to arrive at recommendations for the ethical use of AI in pandemics.

Project team

Dr Jamie Webb, University of Oxford (UK)

Dr Tess Johnson, University of Oxford (UK)

Kadija Ferryman

Ruth Faden