The robo-job apocalypse is being delayed

A sprawling new report makes the case that automation and AI won’t lead to widespread job destruction anytime soon.

Why it matters: Technological advances in AI and automation will have an enormous impact on the workforce, but it may take decades for those effects to be fully felt. That gives business leaders and politicians a last chance to change labor and education policies that have left too many workers locked in low-quality, low-paying jobs.

What’s happening: On Tuesday, MIT’s task force on the Work of the Future released its final report, coming to the conclusion that the immediate effects of automation on work were overhyped and overshadowed by longer-term political trends that have allowed an increasingly unequal share of economic growth to be captured by the well-off.

• “If we deploy automation in the same labor market system we have now,” says David Mindell, an engineer at MIT and one of the report’s main authors, “we’re going to end up with the same results” — an ever-expanding divide between the haves and the have-nots.

Details: In contrast to more pessimistic predictions, the task force found automation and AI currently had the same effect on total job numbers as past technological shifts — some jobs were being destroyed while others were being created, even as overall employment generally kept rising.

• The result is a picture of jobs that looks very different than it did decades ago — 63% of jobs in 2018 didn’t even exist in 1940 — but ultimately one where work remains available.
• It’s possible that one day truly general AI and highly capable robots might be able to do whole classes of jobs more efficiently than human beings, but “the adoption and deployment of these technologies take time, and we’re just at the beginning of a 30- to 40-year cycle,” said Elisabeth Reynolds, the executive director of the task force, at an event today.

Be smart: The reality is that the human brain — and just as much, the human hands — are still far more adaptable and flexible than any machine, no matter how narrowly intelligent.

• For now, the effect of automation and AI — and digital technology more broadly — has been one of augmentation rather than replacement, making individual workers more productive by automating routine tasks.

Yes, but: The task force argues the fruits of that productivity growth have not translated into broad income increases not because of technology but because of a fundamentally broken labor market.

• Between 1948 and 1978 productivity and average wage growth for production and nonsupervisory workers rose in near lockstep, but since then, median wages have stagnated even as technology-aided productivity has continued to rise.
• While workers around the world grapple with the effects of technology, Americans fare worse — the task force found adjusted gross hourly earnings of lower-skill workers in the U.S. averaged $10.33 in 2015, compared to $24.28 in Denmark, $18.18 in Germany, and $17.61 in Australia.

The response to these trends should be less technological than political, the task force urges.

• The U.S. needs to modernize its labor policies: upgrading unemployment insurance, improving collective bargaining rights and raising the minimum wage.
• Workers — especially the more than 60% of adultswho lack a four-year college degree — need help in reskilling and retraining for the future, including on the job.
• The federal government should reverse the long-term declinein its contribution to R&D in order to “continue the innovation cycle,” says Mindell.

The bottom line: The future many American workers face as automation filters in is not one where work is scarce, but where work is low paying and unsatisfying.

-AI needs to be personalized for real-world business

A pioneering AI scientist and entrepreneur argues that technology needs to be specialized to work effectively in manufacturing.

Why it matters: AI has been slower to make a difference in many forms of business because it still takes expertise and investment to use it effectively. For now, that means models will need to be trained individually to be effective on the factory floor.

What’s happening: Landing AI, a company started by former Google and Baidu AI pioneer Andrew Ng, last month launched LandingLens, a visual inspection platform for manufacturers.

• Visual inspection — which usually involves human workers checking a product for defects as it’s made — is laborious and repetitive, which should make it a perfect use case for AI in manufacturing.
• But visual inspection demonstrates what Ng calls the “customization problem” of AI in manufacturing. “Every factory and every unique product needs its own trained AI model to check for defects.”

How it works: LandingLens works as a visual interface through which companies can train the model to understand the inspection needs of individual products.

• Manufacturers take pictures of a successfully finished product, as well as images of various defects. Once human inspectors have created labels for defects, the model can run experiments that will over time refine its ability to identify even minor defects.

The catch: LandingLens points to both the benefits and limitations of AI at its current level of development. Manufacturers can get real value out of AI platforms, but it requires investment in time, as well as skills many executives don’t have.

• While 84% of C-suite executives believe that they must leverage AI to achieve their growth objectives, according to a surveylast year, 76% report they don’t really know how to do it.

“AI has transformed consumer software, but if you look at the impact it has had on the broader economy, then candidly we are just beginning the path of transformation.”

— Andrew Ng, Landing AI

Πηγή: axios.com