How do Humans and Algorithms Interact? Augmentation, Automation, and Co-specialization for Greater Precision in Decision-making

The pervasive use of algorithms to increase decision intelligence raises critical questions. How to combine human insights and algorithms’ data-processing capacity? When augmented by algorithms, should humans focus more on some specific tasks in their jobs? To address these questions, we propose a model of human-algorithm interaction whereby the two agents differ in the type of information they can process (in terms of content, relevance, frequency, and cost) and can complement each other for greater precision in decision making. Using standard tools of Bayesian statistical learning, we compare the cases when humans and algorithms specialize (on certain tasks and without interacting) with the cases when they collaborate by co-specializing. Depending on the properties of information, co-specialization may be anchored to: (i) the human’s capacity, with the algorithm augmenting the human on the most routine tasks while automatizing less routine tasks; or (ii) the capacity of the algorithm, with the human focusing both on the most and the least routine tasks. We discuss the implications for job, contract, and organization design.