AI capabilities are increasing at full speed. What experts considered to be major technological obstacles a few years ago (mathematical reasoning, general knowledge, image and video generation) have been solved one by one with recent advances in generative AI. This progress paves the way for unprecedented opportunities to transform our relationship with work and address the major challenges of our time. From advancing nuclear fusion and improving climate prediction to developing new materials and enhancing medical diagnosis, the potential benefits are immense.
However, this rapid evolution of AI poses a number of challenges for our society. We already have concrete examples: an artist who sees his career compromised by content generation systems, a woman whose image is used without her consent to create pornographic content, or even a company being extorted of 25 millions euros from an attack based on deepfakes.
In addition to malicious uses, artificial intelligence poses new types of risks due to its nature. It is a unique technology that is not built, but is trained. How these systems work is still poorly understood, resulting in unexpected behaviors and unexpected biases in deployed systems. For example, the Bing Chat chatbot threatened its users or made declarations of love to them, while the countermeasures developed to make these systems more ethical often end up with unexpected biases, such as the model Gemini by Google that generates erroneous historical images.
The acceleration of generative AI developments, fueled by a competitive race and investments that increased by eight in one year to reach 25 billion dollars in 2023, forces us to look at the technological breakthroughs of the coming months and years. As illustrated below, we need to consider the range of possible scenarios now to better prepare for future challenges.
That's why we founded the CeSIA. Our mission: to provide technical insight into AI development trends, identify current risks and challenges, and anticipate future ones.
Our activities revolve around three axes: research and development, which includes creating open source tools to evaluate the monitoring of AI systems; information for the general public, to raise collective awareness on the uses and challenges of AI through interactive demonstrations and other content; and the training of future generations of researchers and developers, focused on the evaluation and alignment of AI systems.
