Last Friday in Tunisia, Egypt was elected to chair the AI Governance Working Group of the Arab Permanent Committee for Artificial Intelligence — the body tasked with writing the region's rulebook for AI. AI is no longer a pilot project in Egyptian banking. It is moving into credit scoring, fraud detection, anti-money laundering, and customer risk — the functions that decide how fast a bank's loan book grows and how expensive its mistakes become. Joining me now is Nevine Makram Labib, Distinguished AI Advisor, Senior Technology Strategist, and Chair of the Global AI Governance and Telehealth Working Group at the International Society for Telemedicine and e-Health. Professor Nevine, welcome to Fintech TV.
Hi, thank you for having me.
Egypt now chairs the Arab Committee writing the region's AI governance framework. What does Egypt actually have to export here — and what is it still missing at home?
In Egypt we are now working on a comprehensive framework — not just IT audits, but a framework that governs the software, the data governance, the algorithm, and the results. We have a sovereign template in development. But what we are really missing is talent retention. We have many young specialists in AI governance, but we cannot retain them across institutions. The other crucial point is enforcement. Having a framework is not enough. If you only have guidelines and ethics but no clear regulations and a legal mechanism to enforce them — that is the critical gap we need to address right now in Egypt.
In Egyptian banks right now, where is AI genuinely making decisions that affect money and credit — and where is it still just a chatbot with a press release?
Generally speaking, you have chatbots — especially NLP in Arabic, which is genuinely important. But AI is not there only to substitute customer service. It is there to support decision-making. Right now we are not handing AI the final decision, but AI is there to recommend. So in credit scoring, in fraud detection, in AML — AI is making recommendations that inform the decision, not replacing the human. The shift happening now is from reactive to proactive — using predictive analytics to help institutions take proactive decisions rather than responding to problems after they occur.
Egyptian lenders are now using behavioural and alternative data scoring to reach customers with no credit history. Where is the line between expanding the addressable market and building bias into the loan book?
We need to rely on data that is not biased. If you evaluate a customer using behaviour data — how they manage their debts, their payment history — sometimes you have inaccurate or biased data sets that feed into decisions. We need to ensure proper data governance: how the data was collected, whether it is updated, how it is disseminated and used. That is how you ensure the algorithm produces good, reliable, and fair results. AI for good requires good data.
Most institutions will buy AI rather than build it. Should vendors be legally required to open their models to bank auditors and regulators, and who absorbs the compliance cost?
Definitely — the algorithm must be open. We need to be able to see how the system reaches its conclusion. If you are relying on a system that is essentially a black box, that is a fundamental problem. That is exactly why explainable AI — XAI — is now the global priority. How can you ask anyone to trust a system that cannot justify or explain how it reached a decision? We need to see the algorithm and understand how the developer approached the problem. On cost — I believe it should be shared between the institution and the developing company. When we talk about an AI project, we are talking about a full paradigm shift for the institution, not just buying a system. The cost of testing the algorithm and ensuring explainability must be specified clearly as part of the project from the outset.
Is credible AI governance a differentiator yet for Egyptian bank investors — and which institutions are treating this as a board-level capability rather than an IT project?
Many institutions still think of AI only in terms of generative AI. That is what we constantly have to explain to top management — when we say AI, we are not only talking about ChatGPT. We are talking about the full umbrella: data analytics, predictive analytics, fraud detection, credit scoring — all of it. Some executives still cannot differentiate between digital transformation and AI, or between simple automation and genuine machine intelligence. We need to make leaders in these institutions understand that distinction so they can initiate the right change management process. That is where it must start — with the top management. Once that understanding is in place, the institution can genuinely benefit from all these technologies, from the simplest chatbot to the most advanced predictive analytics system.
Thank you very much, Professor Nevine. We look forward to having you back.
Thank you very much.
