Cooking the Books 2 – Prophecies about productivity
‘Sky to cut 2,000 jobs at its call centres in pivot to AI’ (Times, 28 March) is a typical headline these days as generative AI (like ChatGPT) is applied to white-collar work.
In an experiment carried out by an American firm of management consultants in 2023, call centre workers using AI handled 13.8 percent more customer enquiries per hour than those without, business professionals wrote 59 percent more documents per hour, and programmers coded 126 percent more projects per week. Averaging these, they came up with a figure of a 66 percent increase in productivity from applying AI. This they described as a ‘big deal’ because this ‘equate to 47 years of natural productivity gains in the United States’. But they were not comparing like with like.
They took into account only the work-time saved at the last stage of production, ignoring the labour previously expended, which would include in the case of their experiment labour spent on developing, installing and maintaining the AI system. If this had been taken into account, the productivity increase would be much smaller.
That developing and installing AI involves considerable amounts of labour is illustrated by a Sky spokesperson saying that ‘it was making a multi-million pound investment in its Livingston site, near Edinburgh, as part of a transformation “to deliver quicker, simpler and more digital customer service”’.
As the November 1979 Socialist Standard explained:
In everyday language a motor car is said to be ‘produced’ by the workers who assemble it, and bread to be ‘produced’ by workers in the bakehouse; but the labour of these workers is only a part of all that required to produce cars and bread. As Marx put it: “We must add to the quantity of labour last employed the quantity of labour previously worked up in the raw material of the commodity, and the labour bestowed on the implements, tools, machinery and buildings with which such labour is assisted” (Value, Price and Profit, chapter VI).
Let us assume that the ‘previous’ hours of labour needed to produce a commodity are 80, and the ‘last’ hours are 20 — a total of 100 hours. Let us further assume that without additional investment, but merely by simplifying the last operation, it becomes possible to reduce the necessary hours from 20 to 10. It then takes only 90 hours in all, in place of 100. Productivity will have risen by about 11 per cent. But if ‘productivity’ is calculated — wrongly — on the last operation only, it will appear to have increased by 100 per cent. Would anyone be so foolish as to look at it in that way? Well, yes — it is happening every day.
A news item about the introduction of a new machine operated by two men instead of the former ten will be presented as ‘two men do the work of ten’, as if the making and maintenance of the machine did not absorb additional labour. So productivity in that example will be said, wrongly, to have been multiplied by five. (…) As Marx explained, the amount of labour that is saved is not the whole saving on the last operating process, but the difference between that amount and the additional labour required for the new equipment (Capital, Vol.I. Kerr Edition, pp. 426-7). This is the true measure of increased productivity.
The mistaken theories have been responsible for a continuous enormous exaggeration of the increase of workers’ output, and corresponding false assumptions about the growth of unemployment. (…)
The consequence has been that every advance in technology — the steam engine, the internal combustion engine, computers and automation — and now the silicon chip — has given rise to prophecies that enormous numbers of workers would soon be out of work permanently.