New report says physical and 'predictable' jobs are at risk...

...of automation; AI experts respond

McKinsey & Co. released a report showing which jobs are susceptible to automation. Here are the takeaways, and what some AI experts think.

With robots helping in hospitals, schools, restaurants, and even on cruise ships, it seems that there are few industries that won't feel the impact of automation. But whether human jobs will be replaced by machines is still a complex question, and it is still too early to know how human jobs will be impacted in the age of AI.

A new report released by McKinsey & Co. on Friday attempts to shed more light on the issue. The key takeaway? "Predictable" jobs are most at risk of becoming automated.

The report, which draws from data from the US Bureau of Labor Statistics, looks at more than 2,000 work activities across 800 occupations. It's a follow up from a report released in the fall of 2015.

In this report, McKinsey looks at "the technical feasibility, using currently demonstrated technologies, of automating three groups of occupational activities: those that are highly susceptible, less susceptible, and least susceptible to automation."

In addition, the research examines employment areas "where robots and other machines are most—and least—likely to serve as substitutes in activities humans currently perform." It also provides some caveats for the results, including the impact of natural language processing technology.

The "predictability" of jobs in manufacturing, food service and accommodations, and retailing, makes them most susceptible to automation "based on technical considerations alone," the report said.

However, it's important to note that "just because an activity can be automated doesn't mean that it will be," the report said. Automating a task also has to consider how cost-effective the solution is.

AI experts have been struggling with the question of which jobs will or will not be automated as technology advances. Several AI experts believe that this report, which gives some nuance to a complicated question, is a step in the right direction.

Toby Walsh, AI professor at The University of New South Wales, believes that although the report reaches similar conclusions to previous predictions on automation, it is more thorough.

"It's really hard to look into the future and predict precise numbers," said Walsh.

Walsh said he had "no confidence" in the widely-sourced Oxford report, which said that 47% of jobs will be automated. "The future is too uncertain to be able to quantify it down to single digit precision. What people should take away is that a significant fraction of jobs are at risk of automation.

Still, Walsh said it's important to remember that even if 47% of jobs become automated, "it does not mean 47% of people being unemployed," since we aren't counting new jobs being created.

Vincent Conitzer, professor of computer science at Duke University, said he "commend[s] them for their main point: That it is not a yes/no question whether each particular job gets automated; rather parts of jobs will be automated and this will change the remainder of the job."

However, he also said that dividing jobs into percentages of time spent on activities and assessing likelihood of automation "can be misleading" because there are often "important interactions between these activities."

Conitzer gave an example about a data collector realizing that something was entered incorrectly.

"Real-world data is often messy, and in artificial intelligence, common sense is still very difficult to obtain," Conitzer said. "AI that performs impressively in an environment where everything is as expected can fail spectacularly when something unusual happens."

In order for something to be automated, Conitzer said, it must reach a high level of accuracy.

"Even failure rates of a few percent can make a solution impractical—think of trying to write by dictating to voice recognition software—and it is easy to overlook the importance of the common sense and ability to deal with unexpected events that people use in their jobs, because it comes so naturally to us," Conitzer said.

"The report may be overstating the potential for automation in the near term in a number of places," said Conitzer. "But progress in technology is notoriously difficult to forecast."

Still, Walsh and Conitzer think the report's focus on predictability hits on a key challenge for automation.

"It's why robots have been put in the factory first—because you can control factory environments," said Walsh. "And it's why we still don't have house cleaning robots; the home is still quite an unpredictable environment."