Health care leaders are embracing artificial intelligence. But after an extensive review of case studies and research literature, we’ve found that their AI initiatives are predominantly focused on developing algorithms that can predict a problem. Rarely are organizations devoting resources to AI efforts aimed at understanding why diseases occur. To intervene as effectively as possible, both kinds of algorithms are crucial.
In health care, the default approach to understanding causality is to conduct a randomized controlled trial. But such trials are expensive, time-consuming, not fully representative of different types of patients and often just not feasible. Causal AI algorithms can infer causal relationships from observational data, telling us how different factors interact with each other and which one is causing what. New approaches for causal AI are being developed and validated. Some areas where we especially see value are: discovering mechanisms of disease; treatment optimization; and identifying social determinants of health.
How do we make causal AI more central? First, we should understand the breadth of what it can offer. Health care leaders must take the time to identify problems in their work that would most benefit from it, partner with companies and academics with expertise in this area and conduct a few case studies.
Second, the ability of causal AI to provide trustworthy conclusions must rely on having accurate and representative data. We can train models to be intelligent only if we give them data of excellent quality that represent the right populations, can be merged with other data sets and have been compared with a good control group. Organizations must invest in building the data infrastructure needed for these algorithms.
Creating algorithms that can ask “why?” is harder than creating ones that can make predictions. But it’s not only a worthwhile endeavor, it’s a necessary one.
(Sema Sgaier is a co-founder and the executive director of the Surgo Foundation. Francesca Dominici is a professor at the Harvard T.H. Chan School of Public Health.)