On the Use of Artificial Intelligence and Decision Modeling in Global Health Research and Surveillance in Arequipa, Peru

• In Event: Modeling Health Futures: Practices of Assuming, Anticipating, and Forecasting - II

Abstract

This paper examines the implementation of a reinforcement learning algorithm being tested by global public health researchers. Based on ethnographic fieldwork conducted in Arequipa, Peru, I examine the application of a Monte Carlo Tree Search (MCTS) algorithm in a mobile, decision-support tool designed to enhance field surveillance of Chagas, a parasitic disease affecting 8-10 million people living in endemic Latin American countries. One of the most striking properties of MTCS is that even its inventors admit they do not understand how it performs so well, arriving at seemingly optimal decisions with consistency. Indeed, the algorithm itself is often said to “behave mysteriously”. Similarly, my interlocuters cannot say why it works, yet they increasingly invest in this model to guide their understandings of human behavior as a variable in disease surveillance.

In this paper, I look at the algorithm-human interactions at the level of design—coders building models and epidemiologists designing studies—and at the level of implementation—inspectors using the app to make decisions and guide movements to collect epidemiological data in the field. How do different researchers manage the forms of uncertainty produced by this model? What happens to error when the code specifically does not allow for it? How is health knowledge made in a situation where, as one interlocutor put it, “we do not know what we are doing”?

I investigate the extent to which global public health can rely upon artificial intelligence for acquiring new methods of surveillance, not only for vector presence and transmission dynamics, but for the movements of humans collecting field data. I argue that it is productive for anthropologists of expertise and design in STS to engage emerging sites in which researchers test decision-based algorithmic technologies to further interventions in their fields.

Author

• Melissa Salm, University of California, Davis