It is a great read. They provide a lot of historical background on technology adoption and its impact on job creation/destruction. And the study itself provides a vision into a possible future as machine learning (ML) and robotics are improving rapidly. Computers can now complete tasks we once thought, not that long ago, they would never be capable of. Tasks like:
- Driving – Google’s driverless car
- Reading – Symantec’s Clearwell system uses language analysis to identify general concepts and present the results graphically. Analysed and sorting more than 570,000 documents in two days
The authors use three “computerization bottlenecks” to address this and help determine which jobs would be hardest to computerize:
- Perception and Manipulation – Jobs requiring high dexterity or the ability to work in cramped or awkward positions will be harder to computerize. Surgeons should be safe.
- Creative Intelligence – Computers will have difficulty doing work requiring the ability to come up with unusual or clever ideas. Artists and engineers should be safe.
- Social Intelligence – If negotiation, persuasion, or assisting and caring for others are required, computers will be out of their element. Teachers should be safe.
Using these bottlenecks, the O*NET job description database, some hand coded training data, and predictive analytics they ranked 702 jobs by “Probability to be Computerized”. You can see the results summarized in the figure below.
(figure from the study)
So if you’re in Engineering, Art, Science, Business, Education, Legal, or Healthcare, you’re pretty safe. But those in Sales, Service, Admin, Production, and Transportation should probably be looking over their shoulders. Also, it’s interesting to note that employment has moved to the edges. Most jobs are either high risk (low physical, social, and creativity demands) or low risk (high physical, social, and creativity demands). There is not much in-between.
The following chart is one I created to investigate the relationship between average income and computerization risk. It combines the paper’s probability ranking data with BLS income data, using Python and the matlibplot module. You’ll want to click to enlarge the image so you can read the text. It shows 685 jobs plotted based on their probability of computerization and average income. The linear regression trend (green line) shows a definite trend: jobs that are more likely to be computerized are already paying less, on average. The jobs with more than 1 Million people currently employed are labeled to give an idea of who is and is not at risk.
It will be fascinating to watch as the job market transforms (as long as you’re not getting the Friday afternoon talks). Of course, we don’t know the timeframe for changes this big. It’s not all going to happen in a just year or two. And, you never know what sort of societal or regulatory hurdles could be put in the way to slow it down. The Oxford paper quotes someone named Mokyr on this point: “Unless all individuals accept the “verdict” of the market outcome, the decision whether to adopt an innovation is likely to be resisted by losers through non-market mechanism and political activism.” This is understandable, those in high risk jobs aren’t going to want to give them up. If that is your livelihood, it is a scary place to be. Even if you’re not threatened by these trends, it is strange to think about a software algorithm writing articles, doing legal research, or medical diagnosis.
If you are in the position to invest in, start, or work for companies doing the computerization of these jobs, there is a lot of opportunity. To see how much, I looked at all the high risk jobs (greater than 70% probability of computerization) and multiplied the number of people currently employed by the mean income. The results is the “Opportunity” represented by the labor cost savings of automating all current positions. The top twenty opportunities are in the table below. Naturally, not all the jobs will actually be at risk, and it may not make sense to automate many of them. But, if you look through the list, you’ll probably think of companies that are already taking advantage of these opportunities. For example, retailers have been installing self-checkout to automate the retail sales / cashiers jobs and Kiva Systems is replacing hand material movers in warehouses.
|92%||Retail Salespersons||$109.8 B|
|85%||Sales Representatives, Wholesale andManufacturing, Except Technical and Scientific Products||$90.9 B|
|96%||Office Clerks, General||$82.2 B|
|94%||Accountants and Auditors||$80.2 B|
|96%||Secretaries and Administrative Assistants, Except Legal, Medical, and Executive||$70.0 B|
|79%||Heavy and Tractor-Trailer Truck Drivers||$62.8 B|
|98%||Bookkeeping, Accounting, and Auditing Clerks||$58.9 B|
|85%||Laborers and Freight, Stock, and Material Movers, Hand||$56.6 B|
|92%||Combined Food Preparation and Serving Workers, Including Fast Food||$55.1 B|
|94%||Waiters and Waitresses||$48.3 B|
|86%||Executive Secretaries and Executive Administrative Assistants||$40.3 B|
|97%||Team Assemblers||$30.1 B|
|84%||Security Guards||$28.5 B|
|88%||Construction Laborers||$28.1 B|
|96%||Receptionists and Information Clerks||$26.1 B|
|73%||Administrative Services Managers||$23.4 B|
|96%||Cooks, Restaurant||$23.3 B|
|95%||Landscaping and Groundskeeping Workers||$21.5 B|
To sum up, this study holds a warning, an opportunity, and a vision. A warning to those with jobs in the high risk areas, an opportunity for those who can help make it happen, and a vision of what the world could end up looking like in ten to thirty years.
I’m not commenting on whether or not I think these things will make life, on average, better. A lot depends on what other opportunities are created for and by people during this time. But I do think it’s going to happen and there’s not much we would be able to do to stop it. Might as well get on board.