Africa risks falling into a “cognitive trap” in artificial intelligence unless it finds ways to connect the continent’s informal expertise and social networks with emerging AI systems, according to IMANI vice president and policy development expert Bright Simons.
Speaking at TEDxBerlin, Simons argued that the biggest challenge posed by AI is not simply the increasing intelligence of machines, but the weakening of the human interactions and collective problem-solving that have historically produced knowledge and innovation.
“The more intelligence we put in machines, the less we seem to co-create together,” Simons said.
As AI companies and technology investors prepare for a potentially unprecedented expansion of computing infrastructure and applications, Simons cited a Goldman Sachs projection of about $7.6 trillion in capital expenditure by AI companies and frontier laboratories over the coming years, saying the scale of investment suggests that investors are targeting applications far beyond routine tasks such as summarizing documents.
Instead, he said, the industry is effectively betting on what he calls “civilizational acceleration” — technologies capable of transforming areas including interplanetary travel, energy storage and precision medicine.
For that vision to materialize, however, AI systems will have to become significantly more capable. Simons said machine intelligence develops through three broad mechanisms: depth, range and amplification.
AI’s dependence on civilization
Simons argued that AI does not develop intelligence in isolation. Rather, it feeds on the knowledge, experiences and interactions generated by human civilization.
He illustrated the point with a hypothetical experiment in which identical large language models were trained on the knowledge available in different historical periods, from ancient Egypt and Mesopotamia to Greece, Rome and Renaissance Italy.
Although the underlying technology would remain the same, the resulting systems would think differently because the civilizations supplying their knowledge were different, he said.
The evolution of civilization also changes the nature of the information available to machines. Simons described a transition from predominantly textual and verbal forms of knowledge toward increasingly spatial, sensory and abductive forms of reasoning.
That shift has implications for the types of problems AI can solve, particularly as systems become increasingly connected to sensors, physical environments and non-AI tools. But Simons said the process is creating a paradox.
The “social edge paradox”
As AI companies increasingly deploy their technology to substitute for human knowledge work in consulting, design and other professional services, organizations risk turning collaborative work into an assembly-line process in which workers delegate critical thinking to machines.
Simons called this the “social edge paradox.”
The concern is that while AI is becoming more capable by absorbing the accumulated knowledge of society, its deployment inside organizations could simultaneously weaken the human interactions that generate new knowledge.
He cited research by Anthropic showing that people verify Claude’s outputs in fewer than 9% of instances, using the finding to argue that AI-assisted work can reduce the amount of critical examination taking place within teams.
“If you think about the fact that the AI companies are being forced today to focus on substitution work, human knowledge substitution work, because the big bet of civilization acceleration is not yet ready, then the first thing you have to pay attention to is how they are turning the knowledge industry, the consulting industry, the design industry, et cetera, into assembly line work,” Simons said.
For businesses, the implication is that AI adoption may require more than deploying increasingly powerful models. Organizations may also need to preserve the human processes through which ideas are challenged, tested and improved.
Africa’s infrastructure gap
Simons sees a particular risk for Africa.
The continent has less than 1% of global compute, data-storage capacity and sensors, according to figures he cited. That leaves African economies at a disadvantage as AI moves beyond text-based applications toward systems capable of interpreting physical environments and solving more complex industrial problems.
Yet Simons said Africa possesses another form of infrastructure that could become strategically important: social interaction.
The continent’s extensive networks of informal businesses, technicians, artisans and community-based expertise could provide a valuable source of real-world knowledge for AI systems if they can be properly captured, structured and integrated.
The problem, he said, is that much of this expertise exists outside formal institutions and therefore lacks conventional credentials.
Simons returned to the example of “Bra kofi,” a childhood Ghanaian archetype of an inventive garage technician capable of transforming scrap materials through experimentation, practical knowledge and intuition.
Such people, he argued, could become important contributors to AI-driven industrial development if systems are created to recognize and formalize their expertise.
From garages to AI training data
One model proposed by Simons involves deploying AI systems inside informal workplaces such as auto garages.
A system could observe mechanics as they debate engine diagnosis and maintenance, record and annotate those interactions, and convert the knowledge into structured video manuals in local languages.
The resulting material could then be used to train other technicians while providing a pathway toward formal certification.
That approach would allow AI to draw on forms of practical knowledge that are often absent from conventional datasets, while helping informal workers gain recognition and access to investment.
Simons said such platforms would need to operate across the boundaries of government, education, business and social entrepreneurship — areas that he argues are not naturally aligned with the business models of large technology companies.
Avoiding an AI development trap
The immediate attraction of AI for African economies is likely to remain its ability to handle relatively low-infrastructure applications, particularly text-based tasks such as contract processing, legislative editing and clerical work.
Simons warned, however, that an overreliance on those applications could reinforce the continent’s historical dependence on knowledge-intensive clerical activities while doing little to advance industrialization.
The next generation of AI, he said, could be much more relevant to areas such as precision agriculture, factory maintenance and scientific development because those applications depend increasingly on spatial, sensory and inductive reasoning.
“If you’re not careful, you lead yourself or you slipwalk in Africa into a cognitive trap,” Simons said, arguing that the continent could become locked into the uses of AI that require the least physical infrastructure while missing the technologies capable of transforming productive capacity.
His proposed solution is what he calls “cross-thinking” or “transmediation”.
Cross-thinkers, in his formulation, are people capable of moving across boundaries between human and machine intelligence, formal and informal economies, and different disciplines and systems of knowledge.
For Africa, that could mean creating new mechanisms to identify informal expertise, connect it with AI infrastructure and translate practical knowledge into scalable commercial and educational assets.
A different model of AI wealth creation
Simons said previous technology waves have tended to concentrate economic power and wealth among relatively small groups. The AI transition could repeat that pattern unless new mechanisms are created to distribute its benefits.
He believes Africa’s informal innovators could instead become part of the foundation of a more inclusive AI economy.
“The civilizational accelerants are people like Bra kofi,” Simons said.
The challenge, he argued, is building the platforms and institutions capable of recognizing those people, documenting their knowledge and connecting them with capital, education and technology.
That would require a broader definition of AI infrastructure, one that includes not only data centers, chips and sensors, but also the social networks through which people develop and exchange knowledge.
For Simons, the people capable of building those bridges, the “cross-thinkers” could determine whether the next AI wave concentrates its gains among a few technology companies or creates new opportunities for workers and businesses across emerging markets.
“The people who should inherit this new age of AI are the cross thinkers who can build the kind of platform that I talked to you about, that can empower them, that can recognize them, that can value their work and enable investment,” he said.
