The debate about artificial intelligence (AI) has settled into a tedious and often polarized equilibrium. On one side, executives are warned of an imminent cataclysm where AI will displace jobs on an apocalyptic scale. On the other side, there are promises that this innovative technology will finally deliver the operational efficiencies it has long been touted to provide. Occasionally, both perspectives are presented simultaneously, creating confusion and apprehension around the true potential of AI.

The shared unexamined premise between both camps is that the enterprise demand for intelligence—essentially, the capacity to solve problems—is fixed. However, historical precedents suggest a different narrative. When the cost of a vital resource experiences a significant decline, demand typically does not remain static; rather, it often surges dramatically. In the contemporary landscape, AI is effectively revolutionizing the cost of processing information, leading to an increased appetite for intelligence from organizations.

Also Read503 Service Unavailable Error Causes Frustration for Users

In 1865, William Stanley Jevons, an English economist, published The Coal Question, wherein he made a counterintuitive but crucial observation. He highlighted that advancements to the steam engine, particularly those made by James Watt, had drastically reduced the amount of coal necessary to accomplish a specific amount of work. While one would reasonably assume that Britain would consume less coal, the exact opposite occurred; as the cost of work decreased, steam power became a viable option in a multitude of new applications—from textile mills to extensive mining operations and railways. The result? An exponential increase in coal consumption.

This phenomenon became known as the Jevons paradox and serves as a crucial economic engine for understanding AI today. Contrary to popular fears, AI is not diminishing enterprise demand for intelligence; it is reducing the costs associated with it. With the advent of cheaper intelligence solutions, organizations are confronted with a vast reservoir of latent demand—numerous challenges that existed but were previously deemed too costly to solve.

The coyote hasn’t run off the cliff

Reflecting on past predictions, a decade ago, the field of radiology appeared to be a prime candidate for an AI-driven extinction. With the advent of image-recognition technologies capable of analyzing thousands of scans efficiently, esteemed figures in the AI community, including Geoffrey Hinton—who later received a Nobel Prize—predicted it would swiftly outperform human radiologists within five years. In 2016, Hinton articulated a stark warning, likening radiologists to “the coyote that’s already over the edge of the cliff but hasn’t yet looked down,” suggesting that organizations should stop training radiologists.

Fast forward to today; it is clear that his dire predictions were overly alarmist. Rather than facing redundancy, radiology has become the leading target for AI integration within the medical field, with approximately three-quarters of over a thousand AI applications approved by the US Food and Drug Administration focusing on imaging. Astonishingly, the demand for radiologists has continued climbing, with the staff at leading institutions like the Mayo Clinic increasing by a staggering 55% since Hinton's prediction, now employing about 400 specialists. In fact, Hinton himself has conceded that he was premature in his assessment, noting that AI would not eliminate the profession but rather enhance radiologists’ efficiency.

This increase in demand is attributable to factors far removed from AI, such as the global trend of aging populations and the corresponding rise in medical imaging needs. However, the Jevons paradox illuminates why AI has not curtailed the burgeoning demand: the development and utilization of AI technologies make each analysis faster and more affordable, enabling more cases to be processed and, consequently, necessitating more skilled professionals to handle the increased workload.

Graph illustrating the relationship between AI and operational efficiency

To further illustrate this concept, let’s take a look at the contact center industry. Traditionally, quality assurance teams would audit approximately 5% of calls—a figure not due to a lack of need, but rather a ceiling imposed by the prohibitive costs of human reviewers. The economics of audio evaluation significantly shifted with the introduction of speech AI. Transcribing, indexing, and auditing each interaction have become so affordable that contact centers are now performing in-depth analyses of every call rather than just spot checks. This transition has allowed organizations to identify systemic issues in billing, track customer sentiment more effectively, and, in insurance, verify that verbal disclosures align with what is documented in the Customer Relationship Management (CRM) systems.

In true Jevons fashion, the demand to analyze the remaining 95% of calls was always present; AI simply transformed what was previously an unaffordable luxury into an essential component of operational infrastructure.

This shift in dynamic is particularly evident in South Africa, where the economic landscape presents unique challenges. Cities and regions across the country, such as Stellenbosch, Soweto, Springs, and Saldanha, do not mirror the tech environments of places like Seattle. Here, organizations face the stark reality that off-the-shelf global AI solutions often falter. The speech models designed in the Global North generally presume monolingual, high-resource contexts. They struggle in South Africa, where conversations fluidly interchange among languages, including English, isiZulu, isiXhosa, and Afrikaans, often set against a backdrop of noise and real-world conditions.

To effectively lower the cost of voice intelligence and enhance operational insights, we require speech models customized for local dialects and specific contexts, which often necessitate training on smaller, targeted datasets. For instance, while global models may rely on hundreds of thousands of hours of diverse audio, efficient models that address local speech patterns can function effectively on as little as 100 hours of relevant recordings. Companies like Saigen are at the forefront of this effort, developing models that are capable of handling code-switching—a linguistic feature prevalent in South African communication that demonstrates the necessity of technology adapting to cultural contexts rather than requiring people to conform to existing frameworks.

The better question

In recent years, South African executives have frequently posed the question: Which workforce capabilities will AI replace? However, a more constructive query would be: Where has demand for operational intelligence been suppressed by unmanageable costs?

While the Jevons paradox holds considerable validity, it is essential to recognize its limitations. The principle primarily applies in scenarios where demand is elastic; thus, when a cheaper input is made available, it unlocks new possibilities that were previously unfeasible. However, in situations where demand remains fixed, the decreasing cost of intelligence can translate into fewer individuals performing the same tasks, which poses a genuine threat to entry-level positions that many young South Africans rely on for employment.

Executives must navigate this delicate balance, discerning which type of demand they are encountering, and then explore what becomes achievable when the cost barrier begins to diminish. As history has shown, the landscape of work is constantly evolving, and AI will not extinguish work; rather, it will perpetually uncover new avenues of value that businesses could not afford to identify before. The challenge lies not in fearing automation but in harnessing it to expand capabilities and drive innovation.