Artificial intelligence (AI) is increasingly framed as a technology whose consequences will depend on the choices we make today.
Governments, companies and individuals face difficult questions about how AI is developed, regulated and used, from its effects on employment and inequality, to concerns about misinformation, cyber security and potentially dangerous uses of increasingly capable systems.
But some of the most consequential choices may not be about AI at all. They are about the systems surrounding it.
AI does not arrive in isolation. It enters schools and hospitals, companies and governments, electricity grids and data centres, labour markets and financial systems. These systems determine not only who gets access to AI, but what can actually be achieved with it.
The inequality beneath the technology
Much of the optimism around AI rests on an appealing idea. Capabilities that were once expensive and scarce can now become widely available.
A student can ask an AI tutor to explain a difficult concept. A small business can analyse markets or automate administration without employing a large team. A junior employee can access analytical support that once required considerable experience.
These are meaningful advances. But access to increasingly capable software is not the same as distributing intelligence.
Intelligence in practice is relational. It emerges through people, knowledge, institutions, infrastructure, capital, experience and the ability to act.
Access to increasingly capable software is not the same as distributing intelligence.
Perhaps much of the current AI debate ascribes too much agency to the technology. Too often, capability is treated as if it resides mainly within the machine, even though outcomes are produced by the wider system in which that machine operates.
Consider two businesses using the same AI model. One has skilled employees, reliable infrastructure, clean data, capital and managers capable of redesigning processes around the technology. Another has the same model but fragmented systems, limited capital and little capacity to turn its outputs into action.
They have access to the same AI. They do not have the same possibilities.
More intelligence does not guarantee action
There is a second assumption worth questioning. We often talk about AI as though greater intelligence is all that stands between existing systems and better outcomes.
South Africans should be particularly wary of that assumption.
A school may have access to an AI tutor, but that does not solve overcrowded classrooms, device shortages, weak connectivity or the wider conditions in which children are expected to learn.
A clinic may have access to advanced diagnostic support, but knowledge does not create medicines, specialists, functioning referral systems or additional beds.
A small business may have access to sophisticated AI tools, but those tools do not provide working capital, reliable electricity and water, or access to customers.
We have seen this gap between knowledge and action before.
Climate change is one of the clearest examples. For decades, scientists have improved our understanding of the causes, risks and likely consequences of a warming planet. We have increasingly sophisticated models, more data and greater analytical capability.
Yet knowledge has not translated automatically into action.
Energy systems, political interests, industrial dependence, public behaviour and economic trade-offs continue to shape what societies are willing and able to do.
The lesson for AI is uncomfortable but important. More intelligence does not guarantee better outcomes. Societies can know more and still struggle to act.
These examples point to the same underlying reality. AI can reduce certain information and analytical constraints, but it does not remove the physical, economic and institutional constraints within which people and organisations operate.
That is one of the ironies of the AI boom. We are becoming better at producing knowledge, predictions and recommendations, while many of our hardest problems remain problems of coordination, incentives, infrastructure and collective action.
We saw something similar during the ERP era. Organisations invested heavily in integrated systems intended to standardise processes, improve coordination and create a more coherent view of the enterprise.
But implementing an ERP system did not automatically mean that an organisation suddenly had better processes. Technology could impose structure and improve visibility, but weak processes, poor incentives and organisational complexity often remained.
The big data era produced a related lesson. Organisations accumulated more data and built more sophisticated analytics, but better visibility did not automatically lead to better decisions or outcomes.
In many organisations, the problem was not simply a lack of information. It was whether they had the capacity and willingness to act on what that information revealed.
Higher education offers another example. When massive open online courses emerged, they were often presented as a way to dramatically widen access to high-quality education.
They did expand access to lectures, course materials and expertise that had previously been difficult or expensive to reach.
But access to educational content did not automatically translate into learning. Students still needed time, motivation, prior knowledge, support and the ability to persist through difficult material. The surrounding educational system still mattered.
That distinction feels especially relevant to AI. Making powerful tools widely available may remove one barrier, but it does not remove all the other conditions required to turn access into capability.
South Africa's AI question
For South Africa, the strategic question is not simply whether our schools, businesses and public institutions gain access to increasingly capable AI. The more important question is what surrounds that access.
AI will give people and organisations capabilities that were previously expensive or difficult to obtain. That is significant progress.
But AI will enter an economy shaped by unequal access to capital, infrastructure and education. It will enter organisations with existing hierarchies, incentives and institutional weaknesses. It will operate within a digital environment already vulnerable to cyber crime and other security risks.
Its impact will also depend on infrastructure, digital networks, education, private-sector capability, regulatory capacity and public institutions whose strengths and weaknesses long predate AI.
These systems will help determine what South Africa is actually able to do with the technology.
The choices we make about AI are important. But some of the choices we neglected in the past may now prove even more consequential.
AI may expand what is technically possible. But technological capability does not automatically translate into a deeper talent base, broader innovation, entrepreneurial opportunity, more capable institutions or social and ethical progress.
The difficult part of the AI transition may therefore be less about what the technology can do and more about whether the systems around it can turn new capabilities into broadly shared progress while managing the risks that accompany them.
AI can widen the boundaries of what is possible. What societies do with that possibility is another question.
Share
