Artificial intelligence and the future of work and skills: will this time be different?
Of course, artificial intelligence (AI) has its own special features as a technology and is obviously different than a car or a copier. At the level of an entire economy, however, the effects of different technologies on jobs and skills are often quite similar. Will this be true for AI as well?
Job destruction and creation usually go together…
New technologies tend to shift jobs and skills. New technologies bring new products, which shift jobs across occupations. With the arrival of cars, the economy needed more assembly line workers and fewer blacksmiths. New technologies also bring new work processes, which shift skills in jobs. With the arrival of copiers, office workers needed to replace ink cartridges but not use carbon paper. Economic history is full of examples of new technologies causing such shifts.
Workers often worry that new technologies will destroy old jobs without creating new ones. However, economic history suggests that job destruction and creation have always gone together, with a shift in jobs and skills that leaves most people still employed.
Will AI differ from past technologies in the way it shifts jobs and skills? To answer that question, we need to know which skills will be supplied by AI and which will be left for people. If workers have the skills AI lacks, they will be able to find new jobs if AI automates their old jobs. In that case, AI will shift jobs and skills just like previous technologies. However, if workers do not have the skills AI lacks, these shifts will break down.
…but do we have the skills to face the AI revolution?
As an example, consider literacy and numeracy. These skills are widely used in many jobs—so widely used that countries invest many years of formal education to help everyone develop them. The OECD assesses these skills in its Survey of Adult Skills (PIAAC) because they are so important for work and education (OECD, 2016).
A new OECD study uses PIAAC to assess what AI can do with these skills as well (Elliott, 2017). The study is only exploratory, but its results are sobering. Current AI techniques are close to allowing computers to perform at Level 3 on PIAAC in literacy and numeracy—at or above the proficiency of 89% of adults in OECD countries. Only 11% of adults are above the level that AI is close to reproducing.
If literacy and numeracy were the only work skills, this new study would suggest that “this time is different” with AI. As current AI techniques are applied, many workers with moderate proficiency in literacy and numeracy would be displaced and would not have the higher-level skills for the jobs that remain. The usual shift of workers between jobs and skills would break down.
Of course, if this happened, we would need to improve education. The results of the Survey of Adult Skills show what might be possible. For adults with tertiary education, 21% are above the computer level in literacy and 23% are above the computer level in numeracy (OECD, 2016, Table A3.3). And in the highest performing countries, these percentages for adults with tertiary education reach 37% in literacy and 36% in numeracy, for Japan and Sweden, respectively. These results are much better than the current OECD average of 11%.
With high-quality tertiary education, many more adults could develop literacy and numeracy skills above the current computer level. However, there would still be a serious problem if literacy and numeracy were the only work skills even with high-quality tertiary education for everyone. We do not have examples of education policies at scale that bring 80% or even 50% of adults above the current computer level. If literacy and numeracy were the only work skills, AI would cause the shift in jobs and skills to break down for many workers.
Fortunately, literacy and numeracy are not the only work skills. There are many more skills that are important for work than the ones measured by PIAAC. Many jobs involve tasks using expert knowledge like medical diagnosis. Many jobs involve physical tasks like driving. Many jobs involve social tasks like conversation.
PIAAC does not measure the other skills needed for such tasks and cannot tell us how AI might perform on them. However, computer scientists are developing AI to reproduce these other skills as well. Do AI’s capabilities in these other skills look like its capabilities in literacy and numeracy? We do not know.
The new study using PIAAC to assess AI in literacy and numeracy is only a first step. The OECD is working with the U.S. National Academies to develop a new programme to assess AI capabilities across all work skills. In the years ahead, policymakers will need this information to know whether “this time is different” with AI.
GOT A FEW MORE MINUTES?
Elliott, S.W. (2017), Computers and the Future of Skill Demand, OECD Publishing, Paris, http://dx.doi.org/10.1787/9789264284395-en.
OECD (2016), Skills Matter: Further Results from the Survey of Adult Skills, OECD Skill Studies, OECD Publishing, Paris, http://dx.doi.org/10.1787/9789264258051-en.
ABOUT THE AUTHOR
Stuart W. Elliott works at the U.S. National Academies of Sciences, Engineering and Medicine. He is the author of the OECD report Computers and the Future of Skill Demand.