A more realistic, less alarmist Oxford study on automation and jobs

In my book, Silicon Collar and in several blogs I have critiqued the 2013 study by two Oxford researchers which reached this alarmist conclusion about the impact of modern day machines on jobs:  “According to our estimates, about 47% of total U.S. employment is at risk.”

I found several flaws in their study. With such a scary conclusion, it was irresponsible for them to not even lay out a timeline. They also did not appear to do a reality check for the job categories they analyzed. They calculated a high 0.79 “susceptibility to computerisation factor” (with 1.0 being the highest) to heavy truck and tractor-trailer drivers. This, when the U.S. trucking industry says driver shortages could reach as high as 175,000 positions by 2024 (even if the industry adopts autonomous trucks, regulations will likely require a driver as a backup). The professors had assigned an even higher factor of 0.84 to cartographers and photogrammetrists (who deduce measure- ments from images), which the Bureau of Labor Statistic projects as one of the fastest growing occupations over the next decade.  They had assigned a yet higher 0.94 factor to accountants and auditors, whereas hiring at U.S. public accounting firms jumped to reach record levels in 2013–2014. What about new jobs from the automation and new digital businesses? The Oxford profs did not feel that was worth quantifying.

I could go on. My dissension paled compared to citations in 500+ academic journals which parroted that Oxford study without questioning any of its assumptions. Now, four years later, not a single job has been lost and yet, Oxford has chosen to not issue a disclaimer or mea culpa.

Instead, a more recent Oxford paper (in collaboration with colleagues from Yale) is much more realistic about “When will AI exceed Human Performance?”. The abstract says

“Here we report the results from a large survey of machine learning researchers on their beliefs about progress in Al. Researchers predict Al will outperform humans in many activities in the next ten years, such as translating languages (by 2024), writing high-school essays (by 2026), driving a truck (by 2027), working in retail (by 2031), writing a bestselling book (by 2049), and working as a surgeon (by 2053). Researchers believe there is a 50% chance of Al outperforming humans in all tasks in 45 years and of automating all human jobs in 120 years, with Asian respondents expecting these dates much sooner than North Americans.”

The chart below provides more “milestones” from the paper. It is much less alarmist than the earlier Oxford study – when the dates for machines being able to match human capabilties are years and decades away.

And yet, I wish this new study had surveyed practitioners, not just AI experts. They would have heard automation is still too expensive. That their industry regulators take a long time to approve technology in the workplace.

And I wish they had considered what I called in my book societal “circuit breakers” to rapid adoption of automation. While the pace of technology supply is accelerating, the acceptance rate is not keeping pace. As my book pointed out:

• An early version of the ATM machine was introduced in 1960. Six decades later and even with mobile banking taking off, we still have over 90,000 bank branches across the U.S., employing over half a million tellers and other employees
• The UPC code and scanner were patented in 1952. It took the grocery industry two decades to start to widely adopt it. The end result was not loss of checkout jobs. It improved inventory control and led to an explosion of SKUs. Even today, with self-check out kiosks in many stores, those jobs have not disappeared.
• We have been predicting the death of “snail mail” for decades as email, texting, Skype, Facetime, and social media have become our preferred methods of communicating with each other. Unbelievably, the US Postal Service business has gone up — it sorts half the world’s paper-based mail and packages and keeps over half a million workers employed.
• With digital voice mailboxes and most of us doing our own word processing and travel arrangements, who needs secretaries or administrative assistants? Well, according to the Bureau of Labor Statistics, that category still employs nearly four million workers in the US.

As a technologist, it frustrates me that markets absorb technology so gradually. As a realistic analyst, I have learned we cannot force that pace too much. The more we hype technology, the more cynical are the adopters.

In any case, I am glad to see a more realistic Oxford paper. Hope they would publicize this more and distance themselves from the previous one.

Update: The World Economic Forum has transcribed the range of dates in this video