Why AI’s spoils will pass South Africa by

“Once you start thinking about growth, Nobel Prize-winning economist Robert Lucas wrote, “it is hard to think about anything else.” Artificial intelligence seems to have a similar effect. Once you start to think about AI, everything else looks like something of a sub-plot. 

This is not merely because it is in vogue, academically. Data-centre capital expenditure is now effectively the last engine still driving the US economy and a substantial one for China’s. Few questions matter more than what the long-term economic and social impacts of AI are, particularly for countries such as South Africa which are a long way – in every sense – from the AI deus ex machina.

The pessimistic take has become something of an orthodoxy, at least for those accounts coming from inside the AI companies themselves. Dario Amodei, chief executive of Claude maker Anthropic, has estimated that half of entry-level white-collar jobs will be eliminated within five years.

His argument runs as follows: as models approach general intelligence, a narrow elite of capital owners will control the platforms, the compute power, and the electricity which powers them. Agentic software and advanced humanoid robots will be a substitute for white-collar and blue-collar labour respectively. Most people will hold few assets and find that what remains of their labour has negligible value.

A scent of investor persuasion

An element of scepticism is needed. These are the views of parties with very specific interests; it is informed conjecture offered by people whose companies’ valuations depend on their being believed by investors.

Why would they puncture the myth that AI will be structurally transformative? Deployment across the economy has barely begun, but even so, labour-market data offers scant and inconclusive evidence of disruption. Daron Acemoglu, another Nobel Prize-winning economist, is one of the more sober voices and a more neutral one; he puts the total factor productivity gain from AI at well under 1% over a decade, concentrated in a handful of sectors.

But there could be an element of truth in the pessimistic take of Amodei. If the productivity gains of AI accrue to capital and a thin stratum of expert workers, the effects could be socially disastrous and exacerbate inequality.

David Autor and Neil Thompson have shown that automation cuts both ways; it can raise wages where it strips out low-expertise tasks and depress them where it substitutes for high-expertise ones. Whether a given worker wins or loses depends on which part of the job the machine takes.

Past example

The distribution of past innovation can also shine a light on potential effects. A study of US patents filed in the early 2000s showed that for every dollar of value added by a new invention, about 30 cents accrued to workers, while as much as 70 cents became profit, to be distributed to shareholders.

Economists usually assume that labour and capital shares of output are stable over the long run. AI could be about to break this assumption, further concentrating the returns in the hands of capital.

Furthermore, second-order effects could compound the first. Workers displaced from cognitive work will not vanish; they are likely to crowd into those occupations that machines are yet to disrupt, pushing down wages for those already there. In this way, even workers in sectors supposedly “safe” from AI – construction, social care, logistics – could see their wages shrink.

‘Age of abundance’

AI optimists reject this notion. They retort that productivity gains from the technology will lower prices, with consumers benefiting. Elon Musk recently told the Economist that he believed that, thanks to AI, we were entering an age of abundance which would make money superfluous (he neglected to address how this would benefit anyone investing in his companies, but that might be beside the point). 

But even if AI does lower the production cost of certain goods, it depends on what becomes more affordable. For example, AI might produce abundant, nearly free software. It will do rather less for the supplies of housing, electricity and food, where elasticity is low and the constraints on supply are physical. Cheap code is thin gruel for someone who can’t afford a house because tech bros have invested their winnings in real estate.

SA exposed

South Africa’s economy is woefully unprepared and especially exposed to these changes.

The country doesn’t need to imagine a permanent AI underclass; it already has one. Roughly a third of the labour force is unemployed on the narrow definition and above 40% on broader measures. Among the young, the figures are worse still. The relevant question is therefore not whether AI creates a bifurcated society, but whether it will widen the already gaping divisions, and whether South Africa’s democracy will be able to withstand the consequential socioeconomic pressures.

Three channels of effects matter. First is outsourcing; business process services – such as call centres and back-office operations in Cape Town and Durban – represent one of the few genuinely successful job-creating export industries of the past decade.

More than 70,000 jobs have been created in the Western Cape alone, with many of those employed being young. Sadly, this industry is at risk of being made completely redundant. Voice and text handling is precisely the kind of thing that large language models do competently and at negligible marginal cost. India’s exposure might be larger in absolute terms, but South Africa’s is arguably more concentrated.

Second is that the South African capital markets have almost no exposure to AI. The principal counter for the AI trade on the JSE is Naspers and Amsterdam-listed Prosus, given their investment in Tencent. A national hedge that rests on one stock and a Chinese platform is not much of a hedge. Finally, there is infrastructure. Data centres require electricity and South Africa’s constraint needs no elaboration.

The fact remains that this is all pure conjecture. It is still unclear as to what the actual effects of AI will be on productivity.

Perhaps the clearest signal of how little anyone knows is the behaviour of investors themselves. Many are buying AI exposure not from conviction and profitability but as a hedge; if it does change everything, and you are not in it, then you are left with nothing.

That is not a forecast; it is a statement of uncertainty. And uncertainty is a challenge more easily dealt with by those with assets than by those without. DM