The societal cost of (mis)predicting the automation of work

Original article can be found here (source): Artificial Intelligence on Medium

The societal cost of (mis)predicting the automation of work

While it’s understandable to view automation and AI as existential threats to the American worker, the realities of industry disruption are somewhat less provocative than the popular narrative.

By Scott Andes

There has always been a successful cottage industry for forecasting future trends, but lately, predicting the scope and pace of worker automation has become a full-blown occupation for many commentators. Thus far, our track record hasn’t been so good. In 2015, The Guardian predicted, “from 2020, you will be a permanent backseat driver,” and in 2016, Uber predicted it would manufacture 75,000 autonomous vehicles by 2019. In 2013, Oxford researchers suggested 47 percent of all workers could be automated in two decades. While the study’s authors still have time to make good on their prediction, it’s not looking great.

Accurate or not, these types of predictions have become fully embedded within the public zeitgeist. According to a recent Pew study, eighty percent of Americans believe by 2050, “most the work done by humans will be done by robots.” But just as the public’s view of the future of work is becoming more concrete, researchers and scientists that build autonomous systems are becoming increasingly uncertain.

Even in industries where most expect rapid automation, like long haul trucking, researchers aren’t sure. For example, Parth Vaishnav and Venkat Viswanathan, professors at Carnegie Mellon University’s Block Center for Technology and Society and PhD. candidate Aniruddh Mohan, who study the economic and engineering systems associated with autonomous long haul trucking, think there are simply too many complexities and unknowns to make any predictions. “Anyone who says with a high degree of confidence we will have ubiquitous autonomous trucking in the decade simply isn’t telling the truth,” says Vaishnav.

In terms of artificial intelligence replacing human workers, CMU’s Tom Mitchell believes where we’ve gone wrong is in confusing jobs as a whole with the tasks that comprise them. “While it’s definitely true many occupations have at least one task that can be accomplished by machines,” says Mitchell, “it’s extraordinarily rare for all of a job’s tasks to be automated.”

“Anyone who says with a high degree of confidence we will have ubiquitous autonomous trucking in the decade simply isn’t telling the truth.”

Overall, when asked to predict when we’ll see mass job automation, academics agree, that it definitely won’t be anytime soon. “Given what we know now,” says MIT’s Bill Bonvillian, who studies the workforce, “we think the pace of worker automation is likely to move considerably more slowly than expected. That’s at least the view of MIT’s recent ‘Work of the Future’ report.”

But all of this raises another question — who cares? Does it matter if public opinion on the pace and breadth of automation is far more pessimistic than expert consensus?

Photo by National Cancer Institute on Unsplash

Misinterpreting the pace of technological change isn’t a trivial exercise in prediction. Doing so can have detrimental effects.

First, bad information leads to bad public policy. Policymakers don’t make decisions simply for the problems of the day, but for issues that will impact their constituents in the future. I recently spoke with the mayor of a mid-sized city who told me, “we aren’t planning on building parking lots any longer. Self-driving cars will soon make them irrelevant.” Similarly, while the Bureau of Labor Statistics predicts the number of trucking jobs to grow by 5 percent over the next decade, what if a state legislature eliminated funding for trucking certificates at state-funded community colleges due to their concerns of automation?

By over-predicting looming mass automation, we run the risk of missing the fundamental reality that those workers most vulnerable in the future are the same workers facing the greatest hardship today.

Second, public sentiment on the state of the economy and one’s own prospects is actually incredibly important. Nobel Prize winner Robert Schiller writes in his new book Narrative Economics that the sense that the current economy cannot provide opportunities to many workers adversely affects consumer confidence, labor force participation and the general dynamism of the economy. Over a third of 18- to 25-year-old workers believe their job is likely to be automated. This may be one (or many) of the reasons why, while unemployment rates are at an all-time low, the number of working-age adults not looking for work is at an all-time high. There are enough headwinds facing American workers without the need to fabricate new bad news.

Third, by over-predicting looming mass automation, we run the risk of missing the fundamental reality that those workers most vulnerable in the future are the same workers facing the greatest hardship today. Low-paid service sector workers, gig workers and temporary workers may not be at risk of automation, but rightfully demand workforce interventions ranging from accessible and affordable training to worker insurance to help mitigate the risk of temporary work.

At the same time, by inaccurately judging the pace of automation, policymakers run the risk of undervaluing our ability to respond. If members of Congress really believe 47 percent of jobs are to be automated with the next two decades, where would they even begin? By taking a more grounded view of technological change, informed by scientific and academic experts, policymakers can actually pursue targeted interventions for specific at-risk occupations. In fact, advances in AI are creating incredible new opportunities for personalized tutoring that reduces the cost and duration of workforce training.

While it’s understandable to view automation and AI as existential threats to the American worker, the realities of industry disruption are somewhat less provocative than the popular narrative. By forging ongoing partnerships with the scientists and engineers at the frontier of these technologies, public and private decision-makers alike can anticipate and react to the economic truths of the future of work.

Scott Andes is the executive director of the Block Center for Technology and Society at Carnegie Mellon University. Previously, Scott was a Fellow at the Brookings Institution. His research focuses on the economic and social impact of new technology.