Researchers predict that designers at all wage levels and employees with deep industry knowledge will still be in high demand.
The robot revolution is not about jobs but tasks. That’s what MIT and IBM researchers think after analyzing 170 million job posts over seven years. The idea was to figure out how job responsibilities are shifting, from people to algorithms and among workers at different pay grades.
MIT-IBM Watson AI Lab just published “The Future of Work: How New Technologies are Transforming Tasks.”
Researchers found that people in the middle of the wage scale will get squeezed the most in the short-term as tasks shift to both lower- and higher-paid workers. Martin Fleming, chief economist at IBM and lead author of the report, said this more nuanced understanding of the impact of automation will help governments, companies, and workers prepare.
“The benefit of research of this type is that we’re able to see change is occurring, it will continue and will be an early warning signal,” Fleming said.
Thinking about automation in terms of jobs being destroyed is the wrong way to view the coming changes, he said.
“This is much more about transformation of how work is done as opposed to transformation of the volume of work or number of workers involved,” Fleming said. “We see that the tasks performed and the roles we have will change.”
Researchers analyzed changes in job role requirements during the seven-year period of early phase technology adoption. They were looking for tasks that are suitable for machine learning (SML). These job responsibilities are most likely to be automated, including tasks like credentialing and scheduling.
The researchers found that:
- Most occupations in most industries have at least some tasks that are SML.
- Few if any occupations have all tasks that are SML.
- Unleashing machine learning potential will require significant redesign of the task content of occupations, as SML and non-SML tasks within occupations are unbundled and rebundled.
The report authors also predict that designers at all wage levels will be in high demand.
“It’s actually a task that encompasses many different skills,” Fleming said.
Physical flexibility, common sense, judgment, intuition, creativity, and spoken language are some of the skills that companies will still need from humans.
Tasks that are less likely to be automated and more likely to be highly compensated include administrative, design, industry knowledge, and personal care and services tasks. Tasks that have become less well compensated include business, human resources, and sales.
Middle-wage workers squeezed
The researchers wanted to understand how job responsibilities are changing at the task level: which tasks are being automated and which ones are shifting among workers. The analysis found that about four tasks are shifting from middle-wage workers to low-wage workers and one task is shifting from middle-wage workers to high-wage workers. Both groups with new tasks see a pay increase, at the expense of middle-wage workers. The paper did not describe the shifting tasks.
Fleming said that this transformation of the workforce is an opportunity for the US government and employers to improve retraining and education programs and catch up to other reskilling efforts in other countries.
“By not doing those routine tasks, workers can focus where they themselves can create more value and earn more income,” Fleming said. “We see the demand for labor continuing to grow, not only in the US but around the world.”
He said that retraining is key to preserving middle-class wages.
“There is a short-term challenge faced by workers in midskill jobs,” Fleming said. “The key is to train and retrain and adapt and adjust as task requirements shift.”
How jobs are changing
Fleming shared an example of how automation trends are affecting the 381,000 employees at IBM. The company has started to using machine learning to automate pricing tasks. Predicting an optimal price that both meets the needs of the customer and is in line with IBM’s financial objectives as well is a good assignment for machine learning, he said. What does that mean for the people who used to calculate these figures?
“For people in selling roles, they can be more productive because they can better meet customer requests,” Fleming said. “For people in pricing roles who had done this in a manual way, they are freed up to focus on more complex pricing transactions.”
Instead of spending time on a standard bid for a common purchase request, an IBM employee could focus on a customized request or a bid for a company that has more regulatory requirements than others.
“We will see differences in how automation plays out by industry, by product cycle, by geography as well,” Fleming said.
Fleming said that the next phase of automation will be different from the first wave that was driven by robotics.
“A lot of the significant deployment in automotive and electronics production has been completed or is pretty far along,” he said. “In our current environment, the services industries make up a very large proportion of activity which is a work of a different nature than manufacturing.”
Robotics took over “if/then” tasks while machine learning and artificial intelligence is designed for tasks that require learning, reasoning and understanding.
Full automation will create more new jobs
Marcus Casey, the David M. Rubenstein Fellow in Economic Studies program at the Brookings Institution, said that the next phase of research should focus on how this transition to machine learning varies across business types. Casey is an assistant professor of economics at the University of Illinois at Chicago and and directs Automation and the Middle Class research at Brookings. He reviewed the “Future of Work” paper for the MIT-IBM team.
Casey said business leaders and government officials should pay attention to recommendations from another researcher studying automation, Carl Benedikt Frey. Frey made an initial prediction about 47% of jobs being at high risk for automation and is quoted in the MIT-IBM research. As Frey stated in his initial automation research, business process and technology investment, regulatory concerns, political pressure, and social resistance will determine how automation affects jobs and wages.
Frey’s latest thinking is that the true concern is not about automation in general but that the revolution won’t go far enough. The incomplete technology transformation will trap workers in a permanently unequal income distribution. If businesses only go so far toward automation, the full productivity benefit will not be realized.
Casey said that the goal is to get to a point in the machine-learning revolution at which technology is creating new tasks and jobs for people to do.
“What they’re worried about is we’ll get stuck at a place where there’s nothing that could be transformative enough to create new tasks and create new jobs,” he said. “What we want is sufficiently transformative tech that raises productivity enough so that new tasks emerge.”