Intelligent Biosphere

A teenage interest in the emergent dynamics of self-interested, evolving, interacting agents, sparked by the Artificial Life movement, was Drew’s route into studying real ecology at Cambridge, York, and Princeton. Throughout, his focus was on developing realistic simulation models of ecological processes, something that he was able to scale up hugely during his 8 years as head of the Computational Ecology and Environmental Science group (CEES) at Microsoft Research, which developed many such models, at spatiotemporal scales from millimetres to global, seconds to centuries. CEES built the first fully data-constrained model of the global carbon cycle, and The Madingley Model, which simulates the key ecological interactions among nearly all macroorganisms on Earth. From a technical perspective, CEES specialized in Bayesian approaches to constraining esoteric nonlinear ecological models to heterogeneous data, developing new methods and software tools to facilitate such an approach, from algorithms such as Filzbach, to geotemporal software such as FetchClimate. In November 2015, after 20 years devoted to ecological research, Purves changed tack to join DeepMind’s mission to create General Artificial Intelligence.