Reconstructing the biogeography of a hunter-gatherer planet using machine-learning

Abstract

Estimating the human abundance of the pre-agricultural, pre-urban, and pre-industrial planet is important to setting a baseline prediction for the density, diversity, and distribution of the human species if all energy requirements were foraged from wild populations of plants and animals, and all material requirements were met by locally available raw materials. Here we ask what the biogeography and demography of a hunter-gatherer planet would look like if populated by ethnographic foraging societies. Given ethnographic societies likely include some of the largest, densest, and most complex foraging societies to have existed, global inferences drawn from this sample can be considered maximum demographic estimates. To do this we trained an extreme gradient boosting algorithm (XGBoost) to learn ethnographic hunter-gatherer population densities from a large matrix of climatic, environmental, and geographic data and used the results to reconstruct the biogeography of a hunter-gatherer planet. We find the total estimated human abundance to be 6.1 (4.1-8.1) million with an ethnolinguistic diversity of about 8,200 (5,500-10,800) populations. The most common foraging lifestyles were coastal tropical foragers. Major model predictors included temperature and precipitation variables and so the global abundance and diversity of the human species in colder, drier periods of earth history was likely much less.

Use and Misuse of Machine Learning in Anthropology

“…[29]). Machine learning has also been applied to create a model of world population before the adoption of agriculture, which uses modern hunter-gatherers/foragers as analogs for past human groups [30].…”

“…Unfortunately complete data and code were frequently not available, so we were not always able to verify with certainty the exact ML methods that were used and the mistakes that were made. Of the papers cited in this review, the citations that share no, or incomplete data, and/or no code include [2]- [9], [11], [12], [17]- [20], [22], [23], [25], [29], [30], [40].…”