Link Search Menu Document
Bibliography

The future of employment: How susceptible are jobs to computerisation?

pdf link

sciencedirect link

BibTeX 

@article{frey2017future,
  title={The future of employment: How susceptible are jobs to computerisation?},
  author={Frey, Carl Benedikt and Osborne, Michael A},
  journal={Technological forecasting and social change},
  volume={114},
  pages={254--280},
  year={2017},
  publisher={Elsevier}
}

Abstract

We examine how susceptible jobs are to computerisation. To as sess this, we begin by implementing a novel methodology to estimate the probability of computerisation for 702 detailed occupations, using a Gaussian process classifier. Based on these estimates, we examine ex pected impacts of future computerisation on US labour market outcomes, with the primary objective of analysing the number of jobs at risk and the relationship between an occupation’s probability of computerisation, wages and educational attainment. According to our estimates, about 47 percent of total US employment is at risk. We further provide evidence that wages and educational attainment exhibit a strong negative relation ship with an occupation’s probability of computerisation

Notes and Excerpts

They do acknowledge that computers can do quite complex cognitive tasks. That’s the point of the study.

First, our analysis builds on the labour economics literature on the task content of employment (Autor, et al., 2003; Goos and Manning, 2007; Autor and Dorn, 2013). Based on defined premises about what computers do, this literature examines the historical im pact of computerisation on the occupational composition of the labour mar ket. However, the scope of what computers do has recently expanded, and will inevitably continue to do so (Brynjolfsson and McAfee, 2011; MGI, 2013). Drawing upon recent progress in ML, we expand the premises about the tasks computers are and will be suited to accomplish. Doing so, we build on the task content literature in a forward-looking manner.

They use O∗NET for occupational work activities. But Also:

Our analysis builds on the task categorisation of Autor, et al. (2003), which distinguishes between workplace tasks using a two-by-two matrix, with routine versus non-routine tasks on one axis, and manual versus cognitive tasks on the other. … Following recent technological advances, however, computerisation is now spreading to domains commonly defined as non-routine.

Production is Cobb douglass with inputs of (non-susceptible labor and susceptible labor + computers)

\[Q = (L_s + C)^{1-\beta} L_{NS}^\beta\]

Drawing upon the ML and MR literature, and a workshop held at the Oxford University Engineering Sci ences Department, we identify several engineering bottlenecks, corresponding to three task categories. According to these findings, non-susceptible labor in puts can be described as,

\[L_{NS} = \sum_i (L_{PM,i} + L_{C,i} + L_{SI,i})\]

where LPM, LC and LSI are labour inputs into perception and manipulation tasks, creative intelligence tasks, and and social intelligence tasks.

Despite the paper being “about” the increasing capacity of machines, it still looks like they were blindsided by what the bottlenecks actually are. EG:

For a computer to make a subtle joke, for example, would require a database with a richness of knowledge comparable to that of humans, and methods of benchmarking the algorithm’s subtlety

They identify nine )net characteristics related to these supposed ‘bottlenecks’ and then fit a model to to a small set of hand-labelled binary 0,1 automatibility labels. Then they apply that model to all 700 occupations.

I think there is some merit to replicating their paper with different labels assigned to some occupations.

Section II is about history of technology and employment. Potentially good teaching example.

Predicted least vulnerable:

  1. 0.0028 29-1125 Recreational Therapists
  2. 0.003 49-1011 First-Line Supervisors of Mechanics, Installers, and Repairers
  3. 0.003 11-9161 Emergency Management Directors
  4. 0.0031 21-1023 Mental Health and Substance Abuse Social Workers
  5. 0.0033 29-1181 Audiologists
  6. 0.0035 29-1122 Occupational Therapists
  7. 0.0035 29-2091 Orthotistsand Prosthetists
  8. 0.0035 21-1022 Healthcare Social Workers
  9. 0.0036 29-1022 Oraland Maxillofacial Surgeons
  10. 0.0036 33-1021 First-Line Supervisors of Fire Fighting and Prevention Workers

Predicted most vulnerable:

  1. 0.99 43-4141 New Accounts Clerks
  2. 0.99 51-9151 Photographic Process Workers and Processing Machine Operators
  3. 0.99 13-2082 Tax Preparers
  4. 0.99 43-5011 Cargo and Freight Agents
  5. 0.99 49-9064 Watch Repairers
  6. 0.99 1 13-2053 Insurance Underwriters
  7. 0.99 15-2091 Mathematical Technicians
  8. 0.99 51-6051 Sewers, Hand
  9. 0.99 23-2093 TitleExaminers, Abstractors, and Searchers
  10. 0.99 41-9041 Telemarketers

Seems like the model mostly picked up on the social aspects as being the barrier to computerization.

Top and bottom seem plausible to me.

Some iffy ones:

    1. 0.38 27-3091 Interpreters and Translators
  • Economists are hand-labelled as 0, lol.
    1. 0.94 0 35-3031 Waiters and Waitresses
    1. 0.042 15-1132 Software Developers, Application