Sean Ammirati is a partner at Birchmere Ventures and the former COO of ReadWrite.
It’s 2019, and Rosa steps into the law firm she worked at five years ago as a paralegal. It’s a far smaller company than the one she remembers—just a few partners and a small amount of support staff.
Back then, she saw the writing on the wall: Her employer had started using computer programs to automate the grinding work of reviewing documents she and her colleagues had been hired to do.
Rosa recalls a moment of fear, and then determination: She decided to learn how to program. She got a new job as a software developer, then a product manager.
While not all of her colleagues had chosen to pursue a career in software, most today were freelancing on more creative and enjoyable work than the routine scouring of documents they used to do.
Now she was back at her old firm to talk about launching a startup based on an idea she’d been working on for the last few months. She was a client now, potentially a founder of a company—and she realized she had never been so excited to step foot in that office.
The Robot Workforce
Take this scenario across all kinds of white-collar jobs—accounting, human resources, and administration. Any office job that involves drudgery is a candidate for automation. The next generation of robots may not have mechanical limbs, but they’ll do work that humans used to do in much the same way they’ve taken over our factories and warehouses.
This future state is much more likely than most people realize. Over the next decade, I believe sophisticated algorithms running in the cloud and referencing huge data sets will replace most routine professional tasks.
You can actually look back over the last few decades and see a similar transformation in routine manual labor brought about by industrial robotics. The National Bureau of Economic Research published a fascinating paper earlier this year on “The Micro & Macro Disappearing of Routine Jobs.” In the research, the paper’s authors classified occupations into four categories based on the combinations of the work being either routine or non-routine and cognitive or manual. They matched these occupations against statistics from the US government’s Current Population Survey from January 1976 through the end of 2012.
I found it particularly interesting to look at the disappearance of “routine manual” per capita employment over the 36-year period studied. Consider the explosion of the industrial robotics industry over that same time period. The chart below is from the NBER paper:
General Motors installed the first industrial robot in a factory in Trenton, New Jersey, in 1961. According to the International Federation of Robotics, in 1973, there were 3,000 industrial robots in operation. Ten years later, there were 66,000. That exploded to 1.1 million by 2011.
GM installed a spot-welding robot in its Lordstown, Ohio, facility in 1969. Here’s how the IFR described GM’s productivity gains: The robot “boosted productivity and allowed more than 90 percent of body welding operations to be automated vs. only 20 percent to 40 percent at traditional plants, where welding was a manual, dirty and dangerous task dominated by large jigs and fixtures.”
Did this robot destroy jobs? Perhaps not, as Cynthia Breazeal has argued:
“With any new technology, they take over the jobs that people don’t necessarily want to do anyway, and they create new jobs. They empower people to do more interesting work.”
Consider the IFR and NBER data together. The industrial robotic revolution changed the nature of routine manual labor. It also created a new, multibillion-dollar industry growing by double-digit percentages annually. It also, as Breazeal points out, has empowered people to do more interesting work.
Rosa’s tale is a bit of speculative science fiction. But when I look at the progress in computer science, I believe over the next handful of years we’ll see big data and machine learning have a similar transformative impact on professional occupations that industrial robots have had on routine manual occupations.
Looking at a specific field like natural language processing illustrates the point. There is a long history of individuals theorizing about using machines to understand and translate words, even before the first computer. After World War II, researchers began to explore developing techniques used for code breaking for translation in general.
In 1954, Georgetown University and IBM collaborated on the first working demonstration of a computer translating Russian to English. This immediately created national headlines and optimistic predictions about the pace with which progress would be made over the next few couple years.
In 1970, Terry Winograd at the MIT Artificial Intelligence Lab created SHRDLU, sometimes known as Block World. The idea was that in a very constrained environment—a world of simple shapes which would not be unfamiliar to a kid playing Minecraft—a computer could “understand” what a user meant when talking about that environment.
For the next two decades, most natural-language-processing systems focused on building sets of more and more complex rules. However, in the late 1980s, the field evolved as researchers started to leverage machine-learning algorithms to deliver on some of the promises first made about computer translation in the 1950s.
Thanks to the incredible processing-power gains from Moore’s Law and an exploding amount of digital content on the Web, these algorithms now include the use of unsupervised learning algorithms, which beat “supervised” systems—ones given a set of rules in advance—in tackling a wider range of languages and documents.
This transition is where the impact of NLP on industries becomes even more interesting. For example, as the Wall Street Journal reported last year, the US Department of Justice has approved the use of this technology for reviewing documents during legal discovery. The first example of this being used in a DOJ case was the merger of Constellation & Crown Imports. It’s estimated this saved the companies 50% of the cost of manual review.
A report by the Rand Institute For Civil Justice, “Where the Money Goes: Understanding Litigant Expenditures for Producing Electronic Discovery,” suggests that machines can do work of higher quality than humans, at lower cost:
Because this is a nascent technology, there is little research on how the accuracy of predictive coding compares with that of human review. The few studies that exist, however, generally suggest that predictive coding identifies at least as many documents of interest as traditional eyes-on review with about the same level of inconsistency, and there is some evidence to suggest that it can do better than that.
When you think about it, this is the real catalyst for the coming automation of routine professional tasks—delivering higher-quality results much faster and cheaper. At my firm, Birchmere Ventures, we think electronic discovery is just the beginning of the legal profession’s transformation by these technologies.
One way to think about occupations ripe for robots is to look at different professional tasks with a knowable problem and solution—even if it’s really complex to figure out that solution.
When I think about this trend the societal impact will be dramatic. I teach at Carnegie Mellon University. My students come from a variety of different graduate programs there, from computational biology to business. They’ll all have a very different career path over the next few years because of this.
I believe our educational institutions need to prepare our students to be able to navigate this transition to “more interesting work.” To start with, they need to work on developing the skills necessary to solve unknown problems with unknown solutions. That’s why I argue even those who aren’t interested in doing a startup right out of school may benefit from my Lean Entrepreneurship course at CMU.
The good news is that I believe everyone wants to be creative. There should be a lot more opportunity for this as routine work is minimized. Finally, we also need to be preparing students to create their own careers and in many cases be their own boss as a freelancer (something 34% of the workforce is already doing) as “organizational” infrastructures shrink.
If you’re interested in reading & thinking more about this evolution, I can’t recommend The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies strongly enough.
Photo by Matthew Hurst; charts via NBER; SHRDLU image by Terry Winograd