On November 1st, in our section of the L’Altra Ràdio programme (Ràdio 4, RNE), Javier Otero and Marcos Montero, Heads of Artificial Intelligence and Digital Transformation at IThinkUPC, talked about how artificial intelligence systems have evolved in their ability to reason and solve complex problems.

During the early years, AI systems (such as the first ChatGPT) stood out for their ability to generate coherent text, hold conversations and imitate human style. But, as Javier explained, those first versions had clear limitations when solving logical, mathematical or physics problems. For this reason, throughout 2024, the so-called reasoning models emerged, such as OpenAI’s O models, which internally generate chains of reasoning before giving an answer. This qualitative leap has led these systems to achieve surprising results, such as winning the gold medal at the International Mathematical Olympiad and winning programming and physics competitions.

Marcos also presented Absolute Zero, a system created in China that generates new cases and problems to train itself without needing large external datasets. Just like AlphaZero in chess and Go, this model creates challenges and solves them autonomously, expanding its own capabilities. Despite its advantages, it also raises important challenges, such as the risk of generating new logical methods or results that humans are not yet able to validate.

As Javier and Marcos pointed out, these advances are fascinating and very useful, but prudence, critical thinking and ethical application of technology are always required. Machines may reason better every day, but judgement and responsibility remain in human hands.

If you want to learn more about how the reasoning capabilities of artificial intelligence are evolving, we invite you to listen to the podcast of the L’Altra Ràdio programme (around minute 14:40).