Absa is holding off on the use of AI agents that can make decisions on behalf of staff and customers, saying it is still working to vet the technology, an effort to reduce potential risk.
In April, Visa, one of the world’s two major players providing card and payments services to all major banks in SA, announced a new suite of technologies that will allow AI systems to make payments on behalf of human customers in e-commerce transactions.
This is part of a growing trend around AI powered agents, known as agentic AI, taking on more tasks from human beings. For example, an agent that works like a personal assistant — making bookings, creating meetings, summarising notes and other important information.
For a bank, this includes AI agents that can handle customer queries, give financial advice, or even approve a credit or loan application.
Johnson Idesoh, group chief information and technology officer at Absa, says the bank is at present sceptical about giving AI agents such autonomy.
This will likely change in future, but more testing is needed for the full set of risks to hold up to the muster of SA’s highly regulated and often conservative banking industry.
“For now, we’ve taken a stance that we are not going to use these agentic AI solutions to make final decisions,” Idesoh told Business Day.
“The first reason is most of these models are still trained in English, in the US. There’s an increasing number trained in China. And we think it’s important that more of these models get trained in Africa, in African languages, on African data.
“Second, we need to understand and be able to demonstrate how they [the agents] are making decisions. Certainly for an organisation like ours, that’s really important.”
Even then, the technology chief is not naive to the fact that in future AI agents will be a normal part of life.
“In time, agentic AI will be like a human employee,” said the former Old Mutual technology chief.
“For example, if you’re subject to credit card fraud, the rate at which these [agents] could process that and reverse it, is much quicker than I could as a human. So there are areas like that where you absolutely can see these agents will come into them very quickly.”
Until then, “customers can be assured they’re not going to have agentic AI making decisions, a colleague [bank staff] will intervene and be the ultimate decisionmaker,” said Idesoh.
Absa is not alone in its stance. SA banks are known for being some of the best in the world, with a reputation of being risk averse.
Khomotso Molabe, group chief information officer for personal and private banking at Standard Bank, recently said the bank is experimenting with various AI use cases, such as AI-powered financial advice. But such features will be brought to the public only once fully tested to limit or eliminate error.
Agentic AI signifies a paradigm shift in AI, moving beyond reactive systems to autonomous entities capable of perceiving their environment, setting goals, making independent decisions and taking actions to achieve those objectives without constant human oversight.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.