The Future of Employment

Two weeks ago, researchers from Oxford published a very intriguing paper.  The Future of Employment explores which jobs in the United States might be automated, and when.  Clocking in at 45 pages + 16 pages of sorted occupations, it’s worth a read – provides a more granular perspective, industry by industry, compared to MGI’s Disruptive Technologies report from May 2013.

Headlines have publicized the percentage – 47% – of US jobs that are amenable to automation; the model suggests a > 70% chance that computers will be taking over these jobs in the “near future”, perhaps 10-20 years.

That’s not to say the paper is perfect – it eliminates 201 of 903 occupations off the bat for incomplete data, the data wasn’t collected to assess automation, and it’s unclear why only 70 occupations were deemed reasonable to hand-label for training, if an entire Oxford workshop were dedicated to the science of subjective labeling.  Such occupations as Models and Interviewers were considered 98% and 94% likely for automation in the short-term – suggesting the O*NET data is unacceptably faulty in its grasp of specific features amenable to automation.  Also, completely new & substitute occupations for AI aren’t considered, being beyond the scope of this paper.

At the same time, this paper generates discussion & debate, which are key as we peel off toward this brave new future.  What will happen to our world as a whole, and to the economies of individual cities, states and nations, as the next 20 years descend on our heads?  Which industries can and will we completely replace, what new human jobs will replace the jobs that vanish, and when will those new occupations be replaced as well?

It’s hard to imagine that Mathematical Technicians (99%) will be one of the first jobs on the chopping block, but the questions deserve thought.  The pace of occupation automation reflects a couple key factors:

1)   Which industries are least conservative, have the least regulation, and demonstrate moderate levels of concentration – with enough competition that every company strives to gain the slightest competitive advantage?

2)   Which industries are primed for replacement?  Where the automated agents (robots or machine learning-driven user interfaces) address a very large consumer or industry pain, or there exists a shortage of cost-effective human labor (whether world-wide, or in a specific region), where political and social battles create roadblocks to any other path?

3)   Which automated agents require the least variety of adaptations or advancements to existing technology?  Where can customized industrial or personal robots be repurposed, and for those eager to find opportunity, which extraordinary university AI research labs provide fewer routes to tech commercialization and more chances for exclusive licensing?

4)   Finally, in which industries and arenas can the largest social impact be measured or described?  In what situations can autonomous agents succeed in addressing the human condition, where purely human efforts have failed?  Where can you tell the heartwarming stories that all of mankind will support – jobs being done that are too hazardous for humans to consider, jobs that protect or save human life, jobs in the public interest?

There are three useful thoughts in the subtle conquering of industries:

1)   Find jobs hidden in the supply chain, buried deep in the ever-turning industrial cycle, or located far in the background – jobs already outsourced, or jobs in the domain of reasonably large competitive conglomerates.  These are occupations whose automation very few consumers and government agencies would notice.

2)   Find cost savings in industries with vast need for cost savings.  I’d say healthcare, if only it were a less conservative industry, one less tightly managed by political alliances and the trifecta of Pharma, Insurers and Medical Devices.  Look elsewhere.

3)   Raise public outcry for highly personable robots – consumer demand is a tricky but extraordinarily powerful thing.  Personable doesn’t simply mean big eyes, a preemptive smile and the remarkable voice talents of Morgan Freeman or Oprah, though each may be key – it requires many of the tips and tricks that have created rabid early adopters and fan bases from Apple to Spotify.  There’s no shortage of advice online, but think Netflix.

We’re still thinking in very human terms – one human job to one robot.  Why not wholescale replacement of entire industries?  This is a seismic shift along the lines of global industrialization – on a smaller scale, consider even Wal-mart and the gradual demise of Small Town USA.  Who says the robots need to be human-like, on-site, or exist in physical form at all?  What if the future is all machine learning-driven software, 2D and 3D user interfaces, and a few humans left over to push the buttons and act as 5^th backup for core breaches?  Think of Homer Simpson operating the nuclear plant – isn’t a fully automated software solution, as seen in Gen III and IV nuclear reactors, far safer?

Which parts of the human enterprise can be replaced in one fell swoop, in grand or mini-revolutions of the robotic variety?  When’s the imminent day we wake up and find ourselves entirely dependent on trillions of AI agents, failing to see the full ramifications of a tsunami that swept in as we lived life nearly unawares?