Robots and artificial intelligence: The next frontier for employment and EU economic policy

VIDEO & AUDIO RECORDINGS

Session 1

Session 2

SUMMARY

In the first session, the participant discussed the latest Working Paper published by Bruegel on the impact of industrial robots on EU Labour Market. One of the authors first presented his findings and then panellists made comments and suggestions for improvement. Contrary to many other papers on the topic that only consider the potential for displacement of workers by robots, this work uses a methodology that enables to capture both the displacement and productivity effect of robots. Indeed, by using a new technology, producers’ cost decrease, enabling them to produce more. This leads to greater demand for intermediary and complementary products, generating new job opportunities all along the supply chain. These are productivity effects that might offset displacement effects.

The paper finds that, on average across 6 EU countries, one additional robot per thousand workers reduces employment rate by 0.16 to 0.20 percentage point. This implies that, overall, the displacement effect dominates. Younger individual are most affected by replacement by industrial robots. Individuals with middle level of education are the only one significantly affected. Male workforce is also the one that has suffered the most. The effect is stronger than average in industrial sectors, particularly manufacturing. When looking at the impact on specific occupations, it finds that technicians and associate professionals have enjoyed a positive and significant employment effect, while plant and machine operators have suffered a negative effect. This was highlighted by panellists as a very useful result for policy-makers. These results are quantitatively smaller than the results of a similar study done for the US and this might be driven by differences in labour market policies and welfare system.

Panellists were overall positive about the contributions made by the paper, which were needed to make the debate progress. The methodology was praised as robust and the identification as robust and, in particular the use of regions rather than smaller territorial units was seen as a strength. They also made suggestions for improvement and for investigating further several aspects of the work. Some were surprised by the difference of findings with the US, especially since the macro-indicators and trends in both regions would tend to hint at another story. Also, the different impact on gender was at odd with other studies, though these were using a different methodology. One of the panellists recommended to look at other advanced economies than the EU, such as Japan and Korea, where robotization of production has been widespread, as it seems the story in these other labour markets is different. Panellists also highlighted that exploring more recent effect (up to 2015) was needed, as well as broader technologies than narrowly defined industrial robots.

In the second session, the panellists presented several implications of AI for society. One argued that it made business more data-based than before. The importance of AI in business comes from how it enables not especially big but constant structural improvements to business processes, cumulating to significant value for a company over time. The productivity gains are unquestionable, but wage or employment gains depend on demand and supply factors. Another panellist explained how EU was lagging behind in terms of AI capabilities and business digitalization, emphasizing lack of funding for start-ups and other institutional factors. A key insight was that Europeans appear to be much more concerned about privacy than other AI powerhouses, which slows down the creation and utilisation of data for developing AI.

In terms of policy, panellists recommended several initiatives to ensure ethical and safe use of AI. For example, companies could ensure their coders and suppliers abide to a set of ethical and safety principles internally. Panellists emphasized the importance of more STEM training and more investment on basic research. More broadly, general education about what AI technologies are was suggested. Regarding the European markets, fostering the Digital Single Market and competition for functioning markets were recommended. A concrete strategy on human-centric AI was also proposed. A panellist highlighted the importance of reforming the traditional patent framework if it has to be applied to AI technologies, and the possibility of considering at neural network “trained” weights as trade secret. Crucially, she explained how interpretability of AI systems by humans was key to ethical AI, and discussed some IEEE work underway to develop safety standards. Indeed, as one panellist highlighted, AI can be biased by poor training datasets, and so far, not even AI developers cannot understand why their AI systems take certain decision – it remains a black box.

Event notes by  Nicolas Moës

Presentation by Mario Mariniello