For a French AI geared towards the future

On March 13th 2024, the "Commission de l'IA", led by Anne Bouverot and Philippe Aghion, released its first report entitled “AI: our ambition for France” to the President. In the process of societal dialogue around AI that this report wants to initiate, we wanted to make our contribution.

‍

First of all, we salute this effort for its vision and ambition, it precisely highlights the colossal potential of AI and the need for France and Europe to invest in responsible innovation.

‍

In particular, we want to highlight three points:

1. Focus investment on specific segments of the value chain. It is not about running after the progress of American giants, but about exploiting our comparative advantages.
2. The proposal for a Global AI Organization, to define globally harmonized standards and independent evaluation bodies.
3. Preparing society for the wave of generative AI through a plan to raise awareness and train the nation, to train future AI development talents and prepare society for new uses.

‍

The report provides a relevant and detailed overview by focusing mainly on the current paradigms of generative and specialized AI. Analyzing the societal and economic impact of current AI systems is essential to understand the challenges of the present and the near future. However, to have a more complete vision, we propose to strengthen this study by extending the analysis to innovations that are on the horizon, such as systems with a high level of autonomy.

‍

Current generative AI tools such as Stable Diffusion or ChatGPT simply generate content, text, or images in response to a user request. However, this paradigm is changing. We are witnessing the emergence of AIs characterized by greater autonomy, which from a given goal are capable of executing long series of actions with very little human supervision. To do this, these systems break down a task into sub-goals, observe the effects of their actions and adapt continuously.

‍

These systems, whose first generations are illustrated by projects such as AutoGPT or Devin, differ from simple tools by their ability to act independently to achieve a given objective. They raise societal questions, and risks that are qualitatively different from generative AI tools, such as those “related to models that are able to replicate or train other models”, as mentioned in the European AI Act. Let's not replicate the mistake made with the climate issue with AI: let's not let the prospect of immediate gains overshadow the importance of preparing for future challenges, even if they are uncertain.

‍

The future of AI is uncertain; it is essential to explore the extent of possible technological breakthroughs in order to anticipate systemic impacts.

‍

To complement the report with a proactive approach to future technological developments, we make four recommendations:

1. Prospective work on technological advances in AI over the next 5 years. The aim of this report would be to explore a wide range of plausible scenarios and assess their societal impacts, with particular attention to the analysis of emerging risks (as illustrated in the diagram above).
2. Investing in the evaluation of AI systems. The report recommends only 0.05% of the total investment budget to create a French evaluation ecosystem, insufficient in view of the importance of evaluating models, the challenges of emerging risks, and the economic potential of the sector, which could become one of our comparative advantages for Europe.
3. Focus research and development efforts on specialized AI technologies with a strong positive impact such as the creation of new materials or medical diagnosis.
4. Investing in the research and development of technical countermeasures to address AI-related risks such as cyber defense, protection against deep fakes, or the supervision of autonomous AIs.

‍

To read our complete analysis and detailed recommendations, download our response note to the report “IA: our ambition for France” by clicking the link below.