Why business leaders are jittery over AI models
Business
By
Graham Kajilwa
| Nov 16, 2025
Concerns about bias in artificial intelligence (AI) systems have become a major priority for chief executives, who view ethical issues as the biggest obstacle to adopting new technologies in their businesses.
According to the 2025 KPMG Africa CEOs Outlook report, ranking at 63 per cent, ethical challenges top the list of worries for business leaders in the continent, compared to 58 per cent for CEOs globally.
African CEOs also highlighted issues around data quality, availability, and accessibility at 54 per cent versus 52 per cent globally, implementation timelines at 58 per cent compared to 48 per cent globally, and environmental sustainability at 38 per cent against 34 per cent reported by global CEOs.
These ethical challenges were discussed during the launch of the report where biases in the AI systems being pushed in the market were highlighted.
The argument is that, the data sets used to train the AI systems in the market are not familiar with the context of the continent. As such, the quality of data from these systems, when adopted by businesses, may not be accurate.
READ MORE
Somali banks on Port of Mogadishu to push 24hr-economy plan
Inside Sharjah's 'Poetry Pharmacy', where healing comes in verses, not pills
South Africa announces plan to bid for Olympic Games
Trump signs bill to end record-breaking US shutdown
Wangai, Wafula lead Kenyan charge at Sunshine Development Tour
Government moves to restore learning in universities after 49-day strike
AG, MPs want court to lift freeze on cybercrime law enforcement
Nairobi among 33 cities picked to fight extreme heat
Kihara Maina, Group Chief Executive of the I&M Bank pointed this out, saying some of these technologies being adopted in the market come from other regions.
“For example, we have to appreciate that a lot of large language models are being trained on data sets that are not necessarily from Africa. We have to think of the biases that then exist,” he said.
He gave an example of an AI system trained on facial recognition but using individuals of Caucasian descent from the United States. Such system, he said, may not be useful if it is adopted as a security tool in the African setting.
“It is important we are clear about how to make sure that we are watching carefully the kind of biases we might be getting through adoption of some of these technologies,” said Maina.
The KPMG report notes that when assessing AI development and adoption, the nature and quality of the data used to train AI systems is often overlooked.
“Much of the data that fuels today’s leading AI models originates from Western economies, where digital behaviour, language structures and socio-economic patterns differ significantly from those in Africa,” reads the report from the audit and tax advisory firm.
It adds that comparing Africa’s AI readiness to global benchmarks without accounting for these dataset disparities introduces systemic biases that distorts both performance and applicability.
“Models trained primarily on Western datasets often struggle when applied in African contexts. They can mis-identify faces, misinterpret intent or produce outputs that reflect foreign social norms,” says the report. “This is not a failure of the technology itself, but a reflection of bias in the underlying data.
Tackling this issue, the report says, requires African organisations, research institutions and governments to invest in local data curation, labelling and open-data partnerships.
Gerald Kasimu, Partner and Head of Advisory at KPMG noted that the report shows CEOs are not only engaging with external experts to help in adoption and also tailoring of AI training models to ensure inclusivity across different generations.
He pointed out that the assumption is that matters on AI are predominantly for the younger generation in businesses' workforce.
“The reality of the matter is we need to have human in the middle. Technology cannot work on its own and the mature employees have the experience and knowledge to determine whether the results coming out of these AI models are valuable or not,” he said.
Kasimu said results from AI systems need to be validated by humans for them to make sense to the business.
“You need a level of experience to be able to deploy AI productively within the organisation,” said Kasimu.