Can a robot do my job better than I can?

"Could a robot do my job better than I?" This is a question that gnaws at many of us as robots and computers inexorably develop the skills to work in ways once considered quintessentially human.

By Michael Osborne

Think about self-driving cars, or Amazon’s product recommendations, or Google Translate. The real question is not whether these technologies will have employment implications, but rather: how much, and to whom.

The novelty of the modern age is that machines, using advances from machine learning, computer vision and computational linguistics, are increasingly able to perform cognitive tasks. No longer are machines simply substituting for routine physical labour: as automation enters the realm of the mind, many more occupations may come under threat.

Michael Osborne

Michael Osborne at the Digitising Europe Summit 2014 in Berlin (Photo: Vodafone Institute)

In our study, we set out to answer the questions of which jobs were most vulnerable to technological change. It’s clear that the motivations above are already leading to many occupations undergoing automation: Big data analysis is automating paralegal, contract law and patent law tasks; Orchard Supply Hardware in California is introducing robotic shopping assistants; and Kiva Systems was bought by Amazon for US$775m to automate its warehousing. So, if machines can drive, serve customers, and look through data as well as humans, what are humans still good for?

We identified two human faculties which machines are unlikely to supplant in the near future: creativity and social intelligence. To succeed in either a creative or social task, a worker usually needs to have a deep intuitive understanding of other humans, and such knowledge is difficult for machines to learn. We also identified the ability to interact with complex objects in an unstructured environments as a third bottleneck.

While we’ve been hoping for a robotic housemaid for decades, in reality, the challenge of having a robot understand the difference between a pot plant containing dirt and a dirty plate in the messy and changing environment of the average home is likely to keep this a dream for a while longer. With these intuitions, we used US bureau of labour statistics data, and a machine learning algorithm, to predict which jobs were most at risk.

Some of the jobs at most risk according to our analysis were: telemarketers; tax preparers; and insurance appraisers. All have an element of routine cognitive work for which machines may be able to substitute. Some of the jobs least at risk were: occupational therapists; mental health counselors; and primary school teachers.

All these jobs require exactly the intuitive understanding of humans and human environments that machines find difficult. As an overall figure, an alarming 47% of US employment was found to be at high risk of automation. Further, the risk grew with decreasing income, suggesting only a worsening of societal inequality.

We advertise these figures as a call-to-arms. We think that never has it been more urgent for policy to ensure that education and welfare provision is sufficient to ensure that people aren’t left behind by technological change. These amazing new technologies will generate enormous wealth, but it is up to us to make sure that this is to the benefit of all.