Generative AI, like ChatGPT, has surged ahead in recent years, and there's no sign of it slowing down. New applications appear daily, from text production to scientific research, and the use of this technology continues to grow rapidly. Yet, it's crucial to pause and consider the boundaries of these digital "superbrains". How far can they go? What are their limitations? And how reliable are they in their current form? On 8 November, the VUB is hosting the seminar "Limits of Generative AI", as part of the Willy Calewaert Chair, held by Prof. Dr. Emeritus Luc Steels.

This event is organised by the departments of Mathematics & Data Science and Computer Science, offering a deep dive into the weak spots and limits of generative AI. The seminar will be held in Dutch and takes place on 8 November 2024 in room G1.52 on the VUB Main Campus in Etterbeek, from 14:00 to 17:30.

Register now

Programme

The programme provides a broad look at the challenges of generative AI, with talks from leading experts. Paul Van Eecke (VUB AI Lab) opens the seminar by questioning whether generative AI can truly reason like humans. Although AI systems can deliver impressive results, they still make 'strange' mistakes, especially with seemingly simple tasks. What do these errors tell us about the inherent limitations of AI?

Next, Remi van Trijp (Sony CSL Paris) addresses a key question: can AI really understand language? Despite the remarkable achievements of AI models like ChatGPT, there's ongoing debate about whether these systems can genuinely grasp meaning or if they merely follow patterns of words without deeper semantic understanding.

After a short break, Ann Dooms (VUB Mathematics & Data Science) will examine the mathematical boundaries of neural networks. Neural networks are computational models inspired by how the human brain works, used to recognise patterns and make decisions based on data. While neural networks play a vital role in generative AI, there are theoretical limits that define what this technology can achieve. Dooms will explore these boundaries through the lens of computational complexity and statistics.

Finally, Luc Steels (Emeritus, VUB AI Lab) will discuss whether AI is already capable of creativity. While AI is getting better at generating texts and recognising patterns, the uniquely human trait of creativity seems to remain out of reach. Steels will ask whether there are already examples of AI creativity and what this means for the future of human creativity.

Critical Questions in a Time of AI Hype
The seminar offers a unique opportunity to explore fundamental questions surrounding AI, such as the reliability and future prospects of generative AI. These questions are not only important for scientists but also for anyone who encounters AI in their daily life. The debate over AI's limits is more relevant than ever, as this technology penetrates numerous fields.*

*This is a machine translation. We apologise for any inaccuracies.

The world needs you