10 August 2018

Why human-AI collaboration will dominate the future of work

By Alison DeNisco Rayome

We are in the midst of an "AI awakening," as artificial intelligence technologies can now match or surpass humans in fundamental skills like image recognition, Erik Brynjolfsson, director of the MIT Initiative on the Digital Economy, said in a panel discussion session at the 2018 MIT Sloan CIO SymposiumArtificial General Intelligence (AGI)—the point when machines will be able to perform all intellectual tasks that humans can—is still a long way off, Brynjolfsson said. But machine learning has reached superhuman capabilities in certain areas, and can offer enterprises a number of benefits. In two papers recently published in Science and the American Economics Association, Brynjolfsson and colleagues developed a rubric of 23 questions to identify tasks that AI is now adept at, and applied those to the O*NET database of 964 occupations in the US.


Most jobs involve 20 to 30 distinct tasks, the research found. In most cases, machine learning could perform some tasks better than humans in a given occupation. However, it could never perform all tasks needed for the job better than its human counterpart.

"Most jobs will be partly affected by machine learning, but there will also be things humans need to do," Brynjolfsson said in the session. Instead, the future will likely involve partnerships between humans and machines (known as collaborative robots, or co-bots) to more efficiently get work done. "Rarely will we completely wipe out entire job categories," he added.

Only 5% of workers will be displaced by AI, said panel participant Elisabeth Reynolds, executive director of MIT's Work of the Future Task Force, citing McKinsey research.

"The introduction of the co-bot is allowing us to replace routine work and allow workers to do something else," Reynolds said. "You do have to deal with displacement, but it is a small percentage of the growth we see." This echoes Gartner research, which predicted that AI will eliminate 1.8 million jobs by 2020, but will create 2.3 million in that same timeframe.

Take the example of FedEx, Reynolds said: When the company introduced robots that moved freight around to its North Carolina facility, it was projected that they would replace about 25 jobs in the warehouse of 1,300 people. But the hub will still create about 100 new jobs every year. "I think there is more opportunity than we are understanding at this point," Reynolds said.

However, you also have cases like those in some Amazon Fulfillment Centers, which introduced robots but made human tasks less varied and mobile, Reynolds said. "We need to think about how humans are advantaged and the skills they bring to a job when designing technology," she added.

AI workforce challenges

The US currently has about 6 million unemployed people, and 6 million job vacancies. This could have something to do with a skills gap, said Iyad Rahwan, the AT&T career development professor and associate professor of media arts and sciences at the MIT Media Lab. But to win a higher-pay job, a person usually needs more education and analytical skills, which may not be easily attainable, Rahwan's research found.

"We have a skills mismatch problem in this country," Reynolds said. "There's a lot of growth in high-skilled jobs, and we don't have people in regional labor markets filling them." Part of this is due to geographic restrictions, as less than 2% of the American population moves across a state border each year, she added.

"We really control the future of ways of AI and machine learning will be built into work," said Jason Jackson, assistant professor in the MIT Department of Urban Studies and Planning. "We can think about ways machine learning can be used to complement existing work, and make it even better."

Healthcare has a number of strong applications for AI and robotics, the panelists agreed. Physical assistance robots can provide services like lifting patients out of beds that humans may struggle with, Jackson said. And the ability to compile data on diseases can help better diagnose patients, Reynolds said.

In preparing for the future of work, CIOs should look to hire workers that are flexible, and open to learning, as automation may change the nature of their job, Reynolds said.

One of the greatest challenges of implementing AI is ensuring that your data is up to date, and actually reflects some underlying process, Rahwan said.

"Sometimes you have predictive models from data and let it go wild, but then things change," he added. For example, if you optimize something to do with transportation or logistics, and a regulation changes, there could be indirect but important impacts on your business. "If you train machine learning models on one set of data that is historical and then deploy it and the world changes because of something you haven't thought would impact your business, you could be losing out on further opportunities to optimize the business," Rahwan said. "Algorithms have to continuously learn."

One thing is clear: Digital technologies will continue to accelerate, and our current skills, organizations, and institutions are still lagging behind, Brynjolfsson said. "Business as usual won't solve this problem," he added. "We need to reinvent our skills, organizations, institutions, and metrics to keep up with accelerating technology."

No comments: