In 2013, we published a paper titled ‘The Future of Employment: How Susceptible Are Jobs to Computerisation?’ which estimated that 47% of jobs in the US are at risk of automation.

Since then other, similar studies have emerged, arriving at different numerical conclusions but built on the same intuition – that the future of work can be inferred by observing what computers are capable of. There are good reasons to believe that this view is correct. Back in 2003, MIT researchers David Autor, Frank Levy and Richard Murnane highlighted the disappearance, since 1980, of jobs that were intensive in ‘routine’ tasks. 

Their findings were entirely predictable. As early as 1960, Herbert Simon predicted the decline of routine jobs in his essay ‘The Corporation: Will It Be Managed by Machines?’ He argued that computers held the comparative advantage in routine, rule-based activities that are easy to specify in computer code. Through a series of case studies from the same year, the US Bureau of Labor Statistics arrived at a similar conclusion, suggesting that a little over 80% of employees affected by contemporary technological advances would be in jobs involving filing, computing, machine operations such as tabulating or keypunching, and the posting, checking and maintaining of records.

Our estimates – particularly the 47% figure – have often been taken to imply an employment apocalypse. Yet that is not what we were saying. Our study simply looked at the susceptibility of existing jobs – 702 occupations, comprising 97% of the US workforce – to recent developments in emerging technologies such as artificial intelligence (AI) and mobile robotics. It did not predict a timeframe, and it did not explore the new sectors and roles that will undoubtedly arise in the years and decades to come.

Appropriately – or perhaps ironically – most of our analysis was carried out using AI and machine learning techniques. First, though, we gathered a group of machine learning experts to assess, in the context of current technologies, the potential automatability of 70 occupations using detailed task descriptions. Our trained algorithm was then able to assess the automatability of a much wider range of occupations, using data derived from the vast O*NET online jobs and skills database.

We argued in our subsequent report, for instance, that even many non-routine tasks, such as legal writing or truck driving, will soon be automated. Telemarketing and insurance underwriting were among the occupations deemed at greatest risk of automation; social work and many medical professions among the least. Waiting staff were found to be at high risk – an assertion our expert panel did not necessarily agree with but which was proved correct a few years later with the launch of a completely waiter-less restaurant chain. We also provided concerning evidence that jobs associated with low wages and low educational attainment have a strong relationship with potential computerisation.

What our results show is that the potential scope of automation is vast, just as it was on the eve of the Second Industrial Revolution, before electricity and the internal combustion engine rendered many of the jobs that existed in 1900 redundant. Had our great-grandparents undertaken a similar assessment at the turn of the 20th century, they would probably have arrived at a similar figure. Back in 1900, over 40% of the workforce was employed in agriculture. Now it is less than 2%.

Seen through the lens of the 20th century, our estimate that 47% of jobs are exposed to future automation does not seem extraordinarily high. Policymakers need to understand the thinking behind the numbers in studies like ours to draw their own conclusions about the scale of the changes facing us. The world of work is, once again, changing at pace, and will continue to change. We need to be able to craft appropriate responses.

Professor Michael Osborne, Dyson Associate Professor in Machine Learning, Department of Engineering Science

Dr Carl Benedikt Frey, Co-Director and Oxford Martin Citi Fellow, Oxford Martin Programme on Technology and Employment