Robots and artificial intelligence (AI) seem to be in the news all the time, and breakthroughs are announced regularly.
Last year, it was an AI programme which beat the world champion at Go, a game immensely more complex than chess. Now, in the austere American journal the Proceedings of the National Academy of Sciences, comes news of a big step forward in the task of getting AI programmes to think like humans.
Once we have learned to ride a bike or swim, we can remember how to do it, even after a lapse of many years. In the meantime, we will have learned many other skills as well.
This is straightforward for humans. But it has proved to be extremely difficult for AI. When an algorithm such as a neural network learns a new task, the challenge is to prevent it from “forgetting” how it solved previous ones, how to stop its knowledge from being overridden. A big team from Google’s Deep Mind and Imperial College London claims to have solved the problem.
Scientific progress such as this is uplifting and inspiring to read about. Yet there always seems to be a downside. On almost the same day as the Deep Mind paper was publicised, the latest in a series of gloomy reports about the impact of robots and AI in general on jobs was released by PwC.
“Up to around” 30 per cent of existing UK jobs are susceptible to automation by the early 2030s, intones the firm’s blog on the report. Many others have come up with similar sorts of numbers.
For economists, the question of the impact of AI on the labour market is not so much about the eventual impact on jobs. It is about the level of real wages at which jobs will continue to exist.
We have seen massive technological progress for over 200 years. Huge numbers of jobs have been destroyed, but many others have been created. Professor Len Shackleton of Buckingham University points out that, in the census of 1841, domestic servants made up one quarter of all jobs. Lots more were in what he calls the “horse economy”, for railways had scarcely begun. Almost all of these disappeared long ago. Now, we have behavioural pet therapists instead.
Bob Rowthorn at Cambridge and Stephen DeCanio at the University of California have both separately extended the standard model of economic growth to include a robot (AI) sector. DeCanio’s summary is almost a popular caricature of economists: “an increase in robotic labour can have either a positive or a negative effect on wages”. But both of these highly technical papers are serious attempts to grapple with trying to understand the circumstances in which AI will either raise or depress real wages. The answer is not obvious.
Apart from a brief surge of interest in the 1990s, the mainstream model of economic growth has not really been worked on since its inception in the 1950s. But it offers a powerful framework for understanding the impact of AI. Economists should start to focus on it again.