Machine learning in geography–Past, present, and future

This paper concentrates on the different meanings of machine learning (ML) from its origins to the present and potential future, focusing on contributions within the discipline of geography. Understanding the history of ML is important both for understanding current trends and predicting future applications. The 1950s saw the first idea of ML with the Turing Test, which was described as a ‘learning machine’ that could acquire information and become artificially intelligent. This thought prevailed until the 1960s with the release of Perceptron I, which caused a shift to attempt to code computers to learn from input data in order to run like a human brain. The 1980s and 1990s saw the advancement of computing technologies which enabled ML within artificial intelligence to prosper. The development of artificial neural networks during this time shifted ML from a knowledge‐driven approach to a data‐driven one. The 2000s saw ML becoming more widespread within geography and other disciplines as unsupervised methods enabled the analysis of large data sets. More recently, deep learning emerged to enable processes to be integrated into many software services and applications. Present day ML has been defined by Stanford University (2017) as “the science of getting computers to act without being explicitly programmed”. Academic research alongside industry and government investments in ML now account for some of the most important technological advancements. The use of ML within geography has been increasing at a rapid rate paving the way for future advancements in the discipline.