Controlling the effects of AI on work and inequality

This series is a partnership with the Weizenbaum Institute for the Networked Society and the Friedrich Ebert Stiftung

Christian Kellermann and Mareike Winkler open a Social Europe series on artificial intelligence, arguing that regulation will be needed to ensure prosperity for all.

What will be the effects of the digital transformation on jobs? Job creation outnumbering digital job destruction is part and parcel of standard artificial-intelligence (AI) prophecy. But the extent to which work tasks are upgraded—rather than downgraded or even replaced—is determined by at least two dimensions: the technical side and the work aspect.

Today, in the production and service sectors ‘digitalisation’ in most cases means the use of smartphones and tablets. These devices undoubtedly are operated by complex technology—such as AI. But full automation is not yet the main reality.

Nevertheless, the robot—another smart device—is already replacing human work, which has negative effects on wages. Middle- and low-skilled jobs in particular have been affected by information and communication technology (ICT) and robots since the 1970s and 80s. The consequences are decreasing wages on the one hand, and productivity growth and rising ‘digital dividends’ on the other. These dividends, however, are mainly received by the capital owner and explain (in part) the shrinking wage share.

In a country such as Germany, robots are certainly common but in industry they are very concentrated, especially in automotive manufacturing. The vast majority of studies therefore conclude that digitalisation drives the automation of work tasks in certain domains, but also creates much—or even more—work in other, less automated areas, primarily in the service sector.

Daring assumption

Believing that digitalisation must have automatic positive effects on total employment, however, would be quite daring. It depends on the assumption that demand for work lost is (over)compensated by new demand for work elsewhere.

The more precisely this presumed multiplier effect is broken down, the more pronounced the doubts about the associated technology optimism become. The promise is that sectoral productivity gains through digitalisation lead to ‘prosperity for all 4.0’. Yet not only have such ‘trickle-down’ claims gone through a credibility crisis in the last 30 years; they also present a very demanding scenario when it comes to digitalisation.

On the one hand, the assumption is correct that demand for services—or, put more generally, for manual tasks—will increase if some employees receive higher wages because they benefit from digitalisation. On the other hand, these tasks are relatively price-inelastic, so if their price falls due to the use of technology, demand for them will not grow to the same extent.

Technology will not automatically lead to a general increase in prosperity. Instead of focusing on the side of technology and associated investments, a social technology assessment is required, in which the distributional effects of digitalisation are carefully considered. Without controlling AI’s differential effects on the labour market, inequality will continue to rise.

Tacit knowledge

Luckily, the scenario of a highly automated industry remains a vision for the future, mainly because of the complexity of even simple work. Each job comprises a whole bundle of experiences—no matter how routinised the tasks may be. The capacity to work generally requires tacit knowledge about how to deal not only with complexity but also uncertainty, which is out of reach for ‘tool’ or special-purpose AI.

Today, so-called ‘world knowledge’ can be formalised in simple individual cases in AI models, but it is expensive, resource-consuming and always reductionist. The marginal utility of today’s AI is still very limited and does not justify scenarios of massive job losses. These assumptions are usually based on a simplistic understanding of routine work and the production process.

When it comes to regulation, one of the most urgent issues is thus to counter the digital anxiety of many workers with a realistic assessment and an appreciation of their individual working abilities. Practical, including technical, co-determination is also needed in the digitalisation of operational processes.

This requires the strengthening and extension of co-determination structures and rights. Co-determination serves here not only to control the technology but can also be a supportive factor in investment decision-making, which often is not properly recognised by management alone.

Prosperity for all

Finally, a forward-looking policy has the responsibility to correct potential, excessive and unequal distribution effects—so that eventually prosperity for all is in fact created. In the short term, redistributive measures are essential to pursue a social ‘Pareto optimum 4.0’; in the long run, a transition plan is needed towards a world of work which tames advanced AI.

Such shared prosperity will be largely material in nature. But it can also be increasingly immaterial—including a reduction of working time.