On 25 November 2025, the Turgot club, represented by J-J.Pluchart, was invited by the Cerfrance network to lead a conference-debate on the current challenges of AI for productive companies and professional associations. The Cerfrance organisation brings together 700 agencies, 14,000 experts and 320,000 clients “committed to economic, human and sustainable performance”.
After having retraced the cycles of development of AI, successively “weak” then “strong”, the functionalities and the trades generated by AI were analysed. The AI “technology stack” now comprises three layers: systemic AI (semiconductors, networks, data centres), functional AI (recognition, data management, processing, storage, security), and agentic AI (Text: translations, article writing, summaries, scripts, automatic responses; Image: artistic creation, graphic design, illustrations; Video and audio: generation of synthetic voices, music, animations; Code: programming assistance, generation of scripts or functions).
More and more sectors of activity must adapt their work organisation, in particular health, defence, banking and insurance, cybersecurity, industrial production (cobotics), mobility, publishing, cinema and media, education, research, programming, etc. Strategic functions, crisis management and local services are a priori spared. The adaptation proposals formulated in the reports of Villani (2018), Draghi and Letta (2023), Aghion-Bouverot (2024) and the AI Summit of February 2025 (AI “power lever”) were presented.
The issues raised by the development of AI were then discussed: AI and national sovereignty, the contribution of AI to productivity and economic growth, the impact of data centres on GHG emissions, the financing of investments in AI R&D, the profitability of general AI models, the AI “stock market bubble”, the destruction and transformation of jobs, the dominant positions and cooperation agreements of GAFAM, the ethical codes of AI… It appears that the legal problems differ according to the phases of the process of value creation by AI: the collection of data by AI agents, learning by reinforcement models and the exploitation of applications thanks to user questions (prompts).
The conference also focused on the general and specific biases of AI models: technical and psychological biases (perceptual, emotional and cognitive), specific biases of generative AI (related to linguistic disparities and implicit character), voluntary biases (simulations, manipulations, falsifications, intrusions)*.
The presentation was concluded with a reflection on the strategies to be implemented in order to transform AI into a competitive advantage, and in particular, on the new approaches to ‘augmented management’, the integration of management systems, the audit of ‘high AI quotient’ functions, new decision-making aids, benchmarking actions, nudging, assistance with quotes (Relief) and troubleshooting, quality control, cybersecurity, and training in AI and ethics.
The conference was illustrated by use cases and references to the works and reports chronicled on clubturgot.com and analysed in the last book of the Turgot cub: New reflections on the wealth of nations. The lessons of Turgot and Smith.
Jean-Jacques Pluchart
*see review on AI and intellectual property