North African public sector experiences confirm the EU’s phased approach to AI adoption

The growing popularity of Large Language Models (LLMs), such as ChatGPT, has shaped public perception of artificial intelligence, accelerating AI adoption worldwide. This enthusiasm, driven by LLMs' accessibility and ease of use, risks oversimplifying the complexities of implementing AI systems in public sector contexts, writes Oussama Tlaghi.

The EU is among the few public entities that has championed a measured, infrastructure-first approach rather than rapid implementation with its 2024 AI Act, which emphasised human oversight, transparency and institutional readiness foremost.

This approach is particularly relevant in North Africa, where water scarcity, energy limitations, and institutional capacity constraints directly impact technology implementation. Evidence from successful regional digital transformation initiatives demonstrates how systematic, infrastructure-first approaches deliver more sustainable improvements.

Clashing private-public sector objectives

The past decade provides instructive lessons on the adoption of new technologies in public sectors. The blockchain boom of 2016-18 saw governments worldwide announcing initiatives from digital identity systems to land registries. Dubai's blockchain strategy, while attracting substantial private sector involvement, often struggled to demonstrate advantages over traditional database solutions in terms of efficiency or cost-effectiveness. The Web3/metaverse wave of 2021-2022 followed a similar pattern. South Korea's $187 million investment in "Metaverse Seoul" and Singapore’s misadventures with digital assets faced challenges in user adoption and practical utility.

This pattern of initial excitement about transformative potential often overshadows practical considerations. At their core, there is a fundamental disconnect between private sector investment models and public sector requirements.

Firstly, Venture Capital-backed investment operates on a portfolio model that accepts high failure rates. Government tech initiatives, however, must demonstrate success across all implementations, as they affect essential public services and involve taxpayer funds. Secondly, Tech startups’ rapid and disruptive development fundamentally conflicts with government requirements for service continuity and universal access. Whereas start-ups frequently pivot, government services must maintain operational stability and continuity. Finally, private sector companies can operate with high burn rates and negative cash flows based on future potential. Public sector investments require immediate accountability.

Resource constraints in North Africa

The infrastructure demands of AI systems intersect with North Africa's existing environmental challenges. The region faces the world's highest rates of water stress, with resources nearing critical thresholds. Each newer, more complex AI model comes with increasing water demands for data centres and any AI-related infrastructure. Similarly, they are contributing to the growing global electricity demand.

Beyond physical resources, capacity building is essential. Government systems must maintain resilient internal capabilities, which cannot be delegated to external actors. E.g. the African Union’s Continental AI strategy emphasizes the need for capacity building at the civil servants’ level to ensure successful public sector adoption of AI. International cooperation through the EU Global Gateway also offers potential pathways for capacity building.

The identification of worthwhile AI applications requires evaluation beyond technical feasibility. Just as blockchain technology often proved to be an over-engineered solution, many proposed generative AI applications may offer minimal advantages over existing systems. The experience of the COVID-19 response in Africa indicate that the most efficient tech solutions can be locally driven, building on existing systems.

LLMs' vulnerability to errors and their non-deterministic nature make them problematic for public services. Often, these models struggle to discern the relevance of inputs when answering long queries, as they are vulnerable to data contamination and red herrings. Accordingly, the EU’s AI Act stresses the importance of human oversight and deterministic outcomes in its AI risk mitigation guidance.

The implementation of AI systems may also face increasing stakeholder resistance, including from labour unions who are apprehensive of the automation-fuelled job loss, institutional actors who prefer that the aforementioned resources be allocated to other development projects, and environmental groups who spotlight the water and energy requirements.

Digital transformation first

Digital transformation demonstrates fundamental advantages over rapid AI adoption in public sectors. Following its 2021-25 digital strategy, Tunisia has incrementally ensured that internet and data penetration reach its “94 underserved areas.” Tunisia’s internet penetration rose from 66.7% to 79.6% in three years. The digitization of its national ID now provides Tunisians with paperless and remote access to numerous government services. This success can be traced to the project’s alignment with predictable implementation patterns, clear accountability frameworks, and measurable service improvements.

With two decades of e-Gov projects, Morocco’s digital transition offers more success cases, starting with its 90.7% internet penetration (compared to neighbouring Mauritania and Algeria’s internet penetration of 44.4% and 72.9% respectively), its award-winning biometric ID program, and its DigiTPME program that assists SMEs with their digital infrastructure. The alignment of Morocco’s digital transformation with its green manufacturing, specifically in terms of stakeholders and funding, is also notable.

Morocco's financial inclusion strategy (2019-22) illustrates how systematic digital transformation can achieve substantial improvements without resource intensity. By 2022, mobile payment accounts increased by 26%, merchant payment terminals grew by 17%, and financial access points expanded by 17.7%. The program also delivered meaningful social inclusion benefits.

This implementation leveraged existing bank and payment institution networks, expanding digital payment acceptance through merchant terminals, and building on established mobile penetration. Secondly, the incremental deployment allowed for institutional adaptation and capacity building. Rather than attempting a wholesale transformation, the initiative began with basic digital payment infrastructure, gradually expanding from urban to rural areas, eventually reaching 34% of rural communities. In tandem with systematic coordination between financial institutions, payment providers and government agencies, this approach created a stable foundation for growth.

Conclusion

Evidence from Morocco's financial inclusion implementation, Tunisia's e-government upgrades, and broader regional experiences illustrates how digital transformation can deliver substantial improvements without the resource intensity of advanced AI systems.

In line with the EU’s phased approach to AI adoption, governments should first consider identifying high-impact use cases for AI, ensuring proper integration with legacy systems and involving various stakeholders. Public sector agencies must also account for the need for human oversight. First and foremost, capacity building and limiting dependence on private sector expertise and external actors in day-to-day operations should precede the wholesale adoption of generative AI models.

Oussama Tlaghi is a Fellow with the Cambridge Middle East and North Africa Forum's Young Leaders' Initiative. He is a Political Economist with a background in International Relations and Computer Science. Oussama has a Master's degree from the University of Cambridge, and has worked as a Data Engineer.

He is particularly interested in how digital transformation is influencing economic development in the MENA region, and what impact emerging technologies have on regional politics and innovative approaches to sustainable development.

Photo by Click Smith | Nick254 Media Ltd on Unsplash

Share this article: