“The Future of Employment: How Susceptible Are Jobs to Computerisation?” is a 2013 working paper by Carl Benedikt Frey and Michael A. Osborne of the Oxford Martin School. It became one of the most widely cited and most argued-about pieces of automation research of the decade, largely because of a single headline number: the authors estimated that about 47 percent of total US employment was in the high-risk category, meaning those jobs could potentially be automated over the following one to two decades.
The method was novel for its time. Frey and Osborne had a panel of machine-learning researchers hand-label 70 occupations as automatable or not, then trained a Gaussian process classifier on features describing 702 occupations from the US Department of Labor’s O*NET database. The features captured “engineering bottlenecks” they argued were still hard for computers: perception and manipulation, creative intelligence, and social intelligence. Occupations heavy in those skills scored as safe; routine, codifiable work scored as exposed. They also found that the probability of computerisation was strongly and negatively correlated with wages and educational attainment.
The paper drew heavy criticism for treating whole occupations rather than individual tasks as the unit of automation, which many economists argued inflated the estimate; later task-based studies, including OECD work, produced much lower figures. Even so, “The Future of Employment” reframed the public debate and seeded a generation of follow-on research. For a business reader it is the origin of the “47 percent of jobs” talking point, and a useful caution that exposure to a technology is not the same as actual job loss.