As the global artificial intelligence race becomes increasingly dominated by massive computing infrastructure and billions of dollars in investment, South African researchers believe the country can still carve out a meaningful place in the industry — by focusing on smarter and more energy-efficient AI systems.
According to Benjamin Rosman, head of the Wits MIND Institute at the University of the Witwatersrand, South Africa is unlikely to compete directly with technology giants in the United States or China when it comes to scale.
Instead, the country’s opportunity lies in developing alternative approaches to AI that require less computational power and consume less energy.
Rosman said current large AI systems demand enormous infrastructure, including thousands of GPUs, massive data centres and extensive water usage for cooling systems.
By comparison, he noted that the human brain can perform highly sophisticated reasoning tasks using remarkably little energy.
“We can’t compete with OpenAI by simply trying to build bigger models,” Rosman explained in an interview with MyBroadband.
“We don’t have unlimited access to GPUs or the same scale of infrastructure. So the question becomes: can we discover entirely different ways of building intelligent systems?”
Interdisciplinary research driving new AI ideas
The Wits MIND Institute was established as a research and policy think tank focused on artificial intelligence and governance in Africa.
However, Rosman described the institute as something broader — a space where researchers from very different fields collaborate to rethink how intelligence itself works.
Only a portion of the institute’s academics focus directly on computer science or AI engineering.
Researchers and fellows also work in areas such as neuroanatomy, behavioural psychology, philosophy, ethics, humanities, health sciences and game design.
The institute regularly hosts seminars, workshops and collaborative sessions designed to generate unconventional research ideas.
Topics explored range from AI safety and machine consciousness to neuroscience and biological intelligence.
Rosman said one of South Africa’s strongest competitive advantages is its academic talent and ability to foster collaboration across disciplines.
This model has also helped the institute overcome financial and infrastructure limitations.
In 2024, the institute secured R16.5 million in grant funding from Google to support collaborative AI research projects.
The funding has since been used to support experimental interdisciplinary initiatives that researchers describe as “crazy and different.”
From elephant communication to AI sleep deprivation
Some of the projects underway at the institute explore ideas rarely seen in mainstream AI development.
One study attempts to simulate sleep deprivation in large language models by comparing AI behaviour to how the human brain functions under fatigue.
Researchers hope this could reveal hidden failure patterns in AI systems and improve long-term reliability.
Another project involves analysing animal brain structures to identify biological mechanisms that current AI systems do not yet replicate.
The institute has also explored the use of AI models to study elephant communication, believing that understanding complex animal communication systems could inspire more efficient AI architectures in the future.
Rosman believes these kinds of unconventional studies may eventually contribute to the development of AI systems that are significantly more efficient and less dependent on massive computing resources.
He added that the institute measures success differently from commercial AI companies.
Rather than focusing purely on profitable outcomes, researchers prioritise generating new knowledge — even if experiments fail.
“In research, failure still teaches you something valuable,” Rosman said.
“If an idea doesn’t work, you’ve still learned something important about intelligence.”
Source: MyBroadband, Wits MIND Institute
