New research using AI to match patent language with work shows that AI may reinvent high-skill work sooner than you think.
Breakthroughs have unexpected effects
When Amazon first arrived on the scene, books were its early focus for product selection, buyer insights, distribution and pricing. In those early days, retailers typically lamented the plight of bookstores, but relaxed in their conviction that products like clothing, food, jewelry and fashion would be safe.
“Obviously,” consumers would always want to experience such products by physically shopping, touching, smelling and trying them on. Now, we can see how the disruptive effects of Amazon-like technology transformed virtually every market, more quickly and differently than most anticipated.
Similarly, the more we learn about work automation, the more it affects work in ways beyond initial stereotypes. High-skill work, such as analytics, fraud detection, and insurance adjusting are often regarded as “safe” from automation, while materials moving, machine operations, and data verification are seen as most affected the more robotics and software become capable.
Leaders who aim to anticipate the effects of automation on their workforce often focus first on the more “vulnerable” work. However, just as Amazon’s technology affected markets in ways no one expected, so AI may affect high-skill work faster and in more ways than you think.
What does new research tell us about how to anticipate the effects of AI on work?
Anticipating work automation is vital, but AI is more difficult to predict
Automation affects work in subtle ways, at the task level. So, agile leaders, organizations and workers must be vigilant about emerging work automation patterns, anticipate work automation effects early, and reveal pivotal options to adapt. Reinvention will be constant, and leaders and workers must collaborate with trust and transparency, to anticipate–not avoid–opportunities, and embrace perpetual work “upgrades.”
Traditional estimates of work automation effects have used expert opinions about what work tasks will be automated, typically concluding that lower-skill and repetitive tasks are most automatable, focusing on better-known automation approaches, such as robot process automation through physical technology or software.
“Leaders who aim to anticipate the effects of automation on their workforce often focus first on the more “vulnerable” work. However, just as Amazon’s technology affected markets in ways no one expected, so AI may affect high-skill work faster and in more ways than you think.”
However, AI will also significantly impact work–beyond the more familiar robotics and software. AI is so new that it is very difficult for experts to predict its effects. Like the early days of Amazon, there are few examples to draw on when anticipating how AI will impact work.
How can leaders anticipate the work impact of AI, and will that impact be different?
Patents provide a new window into how AI will affect work
That’s where new research by Michael Webb comes in.
His approach to predicting the effects of work automation was different because:
- He focused on AI rather than robotics and software, and
- He used patent language compared to task descriptions, instead of opinions of economists, engineers, or others.
Webb describes his research this way: The text of patents contains information about what technologies do, and the text of job descriptions contains information about the tasks people do. These can estimate how much patenting has been directed at certain tasks.
Suppose a doctor’s job description includes the task ‘diagnose patient’s condition.’ Webb used a natural language processing algorithm to extract the verb-noun pairs from this task, which in this case would be “diagnose condition.” He then counted how many patents contain similar verb-noun pairs, such as ‘diagnose disease,’ with the tasks having more corresponding patents getting a higher automation score.
“…agile leaders, organizations and workers must be vigilant about emerging work automation patterns, anticipate work automation effects early, and reveal pivotal options to adapt.”
Webb recognized that just because a task is automation-susceptible doesn’t necessarily mean there will be fewer jobs containing that task, or occupations containing those jobs, or organizations containing those occupations. It depends on how the work is reinvented, and how labor markets react.
For example, ATM’s increased the number of human bank tellers, because ATM’s reduced the costs of bank branches so much that banks increased the number of branches by a larger proportion than the reduction in tellers per branch. Webb calibrated his method of building from task to job to occupation using past research on robotics and software, and then used that to estimate the effects of AI on tasks, jobs, and occupations.
Webb’s findings reinforce some typical beliefs: Robots mostly affect occupations like materials movers in factories and warehouses, and tenders of factory equipment. Software affects occupations like broadcast equipment operators, plant operators, and parking lot attendants.
However, the findings for AI are different.
Webb says: “Patents describe artificial intelligence performing tasks such as predicting prognosis and treatment, detecting cancer, identifying damage, and detecting fraud.”
Whereas robots perform “muscle” tasks and software performs routine information processing, AI performs tasks that involve detecting patterns, making judgments, and optimization. Most-exposed occupations include clinical laboratory technicians, chemical engineers, optometrists, and power plant operators. High-skill occupations are most exposed to AI, and those with college degrees, including Master’s degrees.
A Brookings report by Mark Nuro, Jacob Whiton, and Robert Maxim extends Webb’s research, showing that occupations such as “market research analyst and marketing specialists” with an average 2017 salary of $70,620, are three standard-deviations more exposed to AI effects than the average occupation, while restaurant cooks, with an average 2017 salary of $25,430 are 1.37 standard-deviations less exposed. That means the market research analyst is in the highest 0.13 of 1% exposure, and restaurant cooks are in the lowest 10% of exposure.
Workforce planning must include automation as “talent”
Leaders should start thinking of automation as an integral part of their talent pool. Leaders can incorporate the best emerging research on work automation in time to do something about it by following the four-step process described in “Reinventing Jobs”:
- Deconstruct jobs into tasks
- Consider the “Return on Improved Performance”
- Evaluate the automation options
- Reinvent the work
Your early warning systems to predict work automation must draw on the latest research, including new data sources such as patents. That’s how leaders can miss the Amazon effect when it comes to work automation.