Is Artificial Intelligence Evolving Gradually or Experiencing a Breakthrough?
This question was at the center of discussions between Jean-Paul Haton, emeritus professor and one of the European pioneers of artificial intelligence, and Karim Beguir, co-founder of InstaDeep, during a panel on the future of AI in Tunisia. Although their analyses differ on the pace of this transformation, they converge on an essential point: Tunisia has the necessary skills to participate in this technological revolution, but it must now invest in infrastructure, computing power, and research to transform this potential into economic value.
Historical Perspective on Artificial Intelligence
For Jean-Paul Haton, born in 1948 and involved in AI research since the 1970s, it is essential to place current progress in a historical perspective. According to him, AI did not begin with ChatGPT. He recalls that the first systems, in the 1980s, relied on what is called symbolic artificial intelligence. Knowledge was integrated manually in the form of rules developed by experts. This approach is still used in certain sectors like banking or industry, but it remains limited by the difficulty of maintaining and enriching these knowledge bases.
The Advent of Data-Driven Learning
The arrival of learning from data has profoundly changed the situation. Thanks to computing power and the explosion of available data volumes, systems are now capable of learning on their own. This evolution already opens up concrete applications in medical imaging, scientific research, or precision agriculture. In Tunisia, Jean-Paul Haton mentioned work done in Sfax around intelligent imaging applied to the agricultural sector.
The Next Stage: Agentive Artificial Intelligence
For the researcher, a new stage is now opening up with agentive artificial intelligence. These systems will be capable of executing complete missions rather than just responding to a request. They will be able to search for information, organize a purchase, manage an administrative process, or coordinate several tasks autonomously.
Limitations of Current Models
Jean-Paul Haton estimates, however, that current models reach an important limit. They produce impressive results but do not truly understand the world around them. They function essentially on probabilities without possessing the common sense that allows, for example, understanding spontaneously that an object dropped will fall to the ground. For him, the next advances will come from models capable of integrating a better understanding of the physical world.
Rapid Evolution According to Karim Beguir
Karim Beguir shares a much faster reading of this evolution. According to him, November 2022 marked the arrival of generative artificial intelligence for the general public, but the real turning point occurred in 2023. Since then, models have progressed at an unprecedented rate and constantly push the limits that seemed unbreakable just a few years ago. The InstaDeep leader believes that the reasoning capabilities of new models are progressing rapidly and that debates about their limits are evolving as quickly as the technology itself. In his view, the real question is no longer whether artificial intelligence will be able to perform certain tasks, but how companies, researchers, and administrations will exploit this potential.
Strategic Challenges for Tunisia
Both speakers also emphasized a strategic challenge for Tunisia. Training engineers is no longer enough; the country must also have its own computing infrastructure. Several speakers highlighted that investing in graphics processing unit (GPU) clusters, essential for developing artificial intelligence models, would be a major lever to strengthen the country's technological autonomy. These infrastructures could also contribute to the development of national solutions in cybersecurity and the protection of critical infrastructures. This ambition, however, requires addressing another challenge: energy. Artificial intelligence models require considerable computing power, and their energy consumption is increasing rapidly. Jean-Paul Haton insisted on the importance of developing more energy-efficient artificial intelligence to reconcile technological performance and sustainability.
Conclusion
The discussions finally highlighted a shared observation. Tunisia already has a significant pool of engineers, researchers, and talents recognized internationally. The challenge now is to give them access to the necessary technical means to create, train, and deploy their own solutions from Tunisia, rather than depending exclusively on foreign infrastructures. These discussions took place within the framework of the third edition of the Tunisia Global Forum (TGF), organized on July 21, 2026, in Tunis by the Association of Tunisians from Grandes Écoles (Atuge). The panel, titled "Building the Foundations of Artificial Intelligence that Creates Value," is part of the preparatory work for the future Tunisian white paper on artificial intelligence, which will define the country's major orientations in terms of AI, digital sovereignty, and innovation.
