Speaker: Daniel Susskind, Balliol College, Oxford
Chair: Adam Bennett, St Antony’s College, Oxford
These days, it is hard to imagine a topic with more significant implications for the future than the impact of technology on the labour market. It is also hard to find people who have combined policy work with research on this topic, such as Daniel Susskind. In this seminar, he presented his insights, particularly regarding the role of ‘advanced’ capital in overturning traditional understandings of the relationship between technology and labour market skills.
Susskind begins by looking at historical data on the labour market. Thus, in the 20th century a ‘skills wage premium’ emerged based on increasing returns to college education. Strikingly, this premium remained even as more and more people were graduating from college due to the skill-biased nature of technological change, which meant demand for skilled labour to operate this new technology outpaced increasing supply. By contrast, in the 19th century, the industrial revolution appeared to have a bias toward increasing the demand for unskilled workers, displacing skilled workers (such as the luddites) with machines that could be operated by unskilled workers. Initial data from the 21st century suggest that today both things are happening at the same time, leading to more employment at the bottom and the very top of the skill distribution, but also to the ‘hollowing out’ the semi-skilled middle classes, which have thereby seen their wages stagnate.
The current consensus in explaining these observations is based around the work of Autor, Levy and Murnane (ALM). They suggest that computer capital can accomplish routine tasks and complement workers in carrying out problem solving in non-routine tasks. The key assumption behind this is that tasks can be thought of as routine when they rely on explicit knowledge and non-routine when they rely on tacit knowledge. Thus, if human beings cannot explain certain tasks, they cannot be performed by a machine, which in turn puts a boundary on the substitution of labour with computer capital.
Whereas this appeared true before the late 1990s, since then we have been able to automate ever more tasks such as truck-driving, medical diagnosis, and others previously considered to rely on ‘tacit’ knowledge. One explanation, supporting ALM, would be that it is just the boundary between routine and non-routine that is moving as we ‘uncover’ tacit rules.
However, Susskind challenges this hypothesis. He argues the routine vs. non-routine distinction is no longer compelling when thinking about constraints on automation. In particular, technology allows machines to perform tasks in very different ways from humans, especially relying on databases and pattern-recognition technologies. In his view, ALM present a special case when the only way to perform a task is to replicate how humans do it; however, there are potentially many other cases where tasks can be ‘routinised’ in a different way than how humans would do them.
At the core of Susskind’s argument is that ‘advanced capital’ can ‘routinise’ tasks that traditionally only human beings have been able to do. Instead of complementing complex tasks, such advanced capital is directly competing with labour, essentially pushing labour into a shrinking set of tasks. This is in sharp contrast with ALM, who focus on ‘traditional capital’ that complements skilled labour, thus raising productivity. Instead, in this model advanced capital erodes the set of tasks in which labour has a comparative advantage.
This argument has significant implications. Theoretically, this shifts the focus away from the old race between technology and education, in which the goal was to train as many as possible so they are able to put systems to good use. Instead, here labour is pushed away from whole sets of tasks. This leads to a race between labour being replaced in tasks by advanced capital and the creation of new tasks elsewhere in society not already performed by advanced capital. A key empirical variable here is the idea of ‘routinisability’, or what makes it possible to make a certain task routine. Indeed, many tasks relying on tacit knowledge and split-second decisions are actually easier to automate thanks to machine learning.
Finally, Susskind addressed head on the policy questions that arise from this account. The skills bias of the 20th century labour market relying on educating people to put machines to good use is likely to be eroded. Therefore, the traditional college education might no longer be the route to top-end skills, requiring a rethinking of education. Furthermore, with ‘advanced capital’ eroding the value of labour (as opposed to ‘traditional capital’, which complements it), the capital share of income is likely to increase further. Therefore, inequality between the owners of machines and workers is likely to spike without changes in taxation that shift the burden more onto capital.
Overall, Susskind paints a somewhat pessimistic picture of the future of the labour market. However, if his distinction between advanced and traditional capital holds, and technological progress continues, society would be well advised to grapple with the implications for education and inequality sooner rather than later.
Ivaylo Iaydjiev (St Antony's College, Oxford)