Cheaper artificial intelligence tends to create more demand for intelligence, not less, according to Sam Clarke, chief executive of South African speech-AI company Saigen. Writing on TechCentral, Clarke applies the Jevons paradox to AI. His firm sells the kind of technology he describes, so the argument is one to weigh rather than accept whole.
His first example is the contact centre. Quality-assurance teams have typically audited about 5% of calls, because checking every conversation by hand was too expensive. Speech AI has made it economically feasible to analyse every call instead of spot-checking a sample, so managers are moving from sampling to reviewing every interaction, from billing errors to customer sentiment.
## What the Jevons paradox says
The term comes from William Stanley Jevons’s 1865 book The Coal Question. Jevons observed that England’s coal consumption surged after James Watt’s improvements to the steam engine, even though the engine used less coal for each unit of work. Cheaper steam power became viable in thousands of new applications, and total coal use rose. He argued that it was “a confusion of ideas to suppose that the economical use of fuel is equivalent to diminished consumption”, a line quoted on Wikipedia’s Jevons paradox entry.
Economists call the effect a rebound: when the unit cost of a resource falls, total demand for it can rise. It does not always. The same reference notes that farm employment in the United States fell from about 40% of workers in 1900 to under 2% by 2024 without total food consumption rising, because demand for food is inelastic.
Applied to AI, Clarke’s point is that when the cost of processing information drops, organisations do not simply replace workers and stop. They apply intelligence to problems that were previously too costly to tackle.
## Radiology as the test case
Clarke cites radiology. In 2016 Geoffrey Hinton compared radiologists to a coyote that is already over the edge of the cliff but has not yet looked down, and predicted AI would surpass them within five to ten years. The prediction did not hold. Fortune reported in May 2026 that the US radiologist workforce grew by roughly 10% over the decade, with 7,469 open radiology positions and average pay of $571,000 in 2025. Hinton later said he had been “speaking purely about image analysis” and that radiologists would work with AI to be more efficient.
Meanwhile the US Food and Drug Administration’s list of AI-enabled medical devices runs to more than a thousand entries, roughly three-quarters of them in radiology. The Mayo Clinic’s radiology staff has grown by 55% since 2016 to around 400 doctors, Clarke writes. Fortune points to other reasons for the shortage as well: US Medicare and Medicaid rules require a physician to sign off on imaging studies, and imaging caseloads rose 25% between 2018 and 2025.
## Where South African speech models fail
Clarke says the effect is most visible where global models stumble. Speech models trained in the Global North assume monolingual, high-resource conditions and falter when a conversation moves between English, isiZulu, isiXhosa and Afrikaans against background noise. Saigen’s models, he says, need only about 100 hours of domain-specific audio, whereas global models are trained on hundreds of thousands of hours. That is what makes call-centre, bank and insurer deployments affordable in local languages.
Saigen has made similar claims elsewhere. Its co-founder, Dr Etienne Barnard, has said the company’s models are designed to adapt in real time to multilingual dialogue, according to a MyPR report, so these are company statements rather than independent tests.
## Limits of the argument
Clarke concedes that the paradox has limits. It holds where demand is elastic. Where demand is fixed, cheaper intelligence can simply mean fewer people doing the same work, which he calls a real risk for the entry-level roles on which many young South Africans depend. The task for South African companies is to identify where demand is elastic and AI can widen the scope of work, and to reskill staff for the higher-value work that the technology uncovers.
For firms weighing a call-centre or compliance tool, the practical test is whether a cheaper process creates work that was previously skipped, such as reviewing all calls instead of five in every hundred. More on this in our Tech & Telco section, and on how firms are funding such projects in Funding.
