AI will destroy entry-level jobs – but lead to a basic income for all

A summit on AI and business is told starter roles in many industries will be automated by machine learning systems, while other experts urge caution against such predictions.

By Nick Heath, May 5, 2016,

Young people will bear the brunt of AI-fuelled job losses as smart systems undercut entry-level roles in everything from marketing to retail.

Machine learning and expert systems will not destroy jobs wholesale, predicts George Zarkadakis, digital lead at advisory firm Willis Towers Watson, but will remove the need for many tasks that employees have traditionally cut their teeth on at the beginning of their careers.

Zarkadakis cited a study by consultants McKinsey, which found that just under one third of activities that make up 60 percent of existing jobs will be automated.

Unfortunately for new entrants to job markets, the bulk of these activities will be concentrated in starter roles, said Zarkadakis.

“We’ve done some research ourselves and looked at the impact on entry-level jobs. Jobs that graduates get once they leave university. We found that many of the entry-level jobs are very susceptible to complete obliteration,” he told The AI Summit in London.

“But look at the impact. What will happen to the world when young graduates will not be able to enter the job market? There will be major disruption in the labor market.”

More about Innovation

The automation of these activities will cut about $1tn from business’ wage bills, predicts Zarkadakis, good news for companies he said, but potentially bad for workers.

In the future, increasing numbers of people will work in the so-called ‘gig economy’, Zarkadakis claimed, where companies contract individuals to perform small tasks on demand. This type of work differs from full time employment in that there are no guarantees of long term ties between the employer and employee – with fewer obligations on both sides.

“This sort of disruption in the job market is already happening because of the gig economy and if we want to have an idea about what will happen with AI, we only need to see what’s already happening with digital platforms,” he said.

“Increasingly companies are reducing full time employees and using those digital platforms to acquire the desired skills on a contract basis. Maybe that’s telling of what the future will be like.”

This shift away from steady employment and income could see governments assume a greater role in preventing individuals from slipping into poverty.

“We’ll have people being responsible for their financial well-being, as opposed to companies looking after them. The role of government will be increased, probably through universal income,” he said, citing experiments with schemes to guarantee everybody a basic income in various countries worldwide.

Yet Zarkadakis’ is only one point of view, Microsoft’s chief envisioning officer Dave Coplin, expressed frustration at talk of technology destroying jobs, pointing out that technology generally complements human labor, rather than replaces it.

“We’re locked in this endless cycle of pointless rhetoric of humans vs machines,” said Coplin, whose employer offers a range of machine learning services via Microsoft Azure cloud and is pushing the idea of smart bots powered by its Cortana virtual assistant as the future of customer relations.

“‘Machines can beat us at chess, they can beat us at Go, they’re going to steal our jobs’. Hang on. Stop. When was this ever the dialogue for what we did with technology? Technology is here to augment what we do.”

Roadblocks to widespread automation

Beyond the question of how AI will affect society, there is the more practical consideration of how long it will be before the technology is mature enough to affect such changes.

Companies are beginning to look at using machine learning and expert systems to further automate manual roles in service industries, in areas ranging from handling helpdesk calls to training shop assistants.

But Harrick Vin, chief scientist at Indian outsourcer Tata Consultancy Services, highlighted the significant obstacles to training machine learning systems that, until solved, will hamper the use of such systems to replace manual labor.

Training machines using supervised learning can take six to 18 months for every new domain of knowledge – an issue when each business can have many different domains they are seeking to automate – he said.

“If you really want to see benefits, like the ones that are being projected, then you’ve got to figure out how to scale and not take six months, one year, 18 months to train an engine to perform one task, because a typical large business performs hundreds, if not thousands, of different activities.”

Without reducing this upfront training time, “it is going to be impossible to scale”, he said.

And not only does it take too long to train systems, once ready to use such systems will likely need new training as businesses change the data they collect and the way they do business.

“You have to build these systems to be inherently adaptable,” he said.