ELLIS Alicante wins competition showcasing AI's potential for critical thinking

ELLIS Alicante Foundation has emerged as the winner of the Shared Task Competition for Critical Question Generation, organized at the 12th Workshop on Argument Mining, held during the 63rd Annual Meeting of the Association for Computational Linguistics in Vienna. This is the most important annual meeting of the Natural Language Processing community.

The winning team was led by ELLIS doctoral student Lucile Favero, under the supervision of Dr. Nuria Oliver, Scientific Director of ELLIS Alicante, and professors Juan Antonio Pérez from the University of Alicante and Tanja Käser from EPFL, with the assistance of intern Daniel Frases.

A total of 13 international teams participated in the competition to develop novel NLP methods that would automatically understand argumentative texts from debates and generate three critical questions about them. These are not just any questions, but reflective ones that aim to probe the reasoning or logic underlying the arguments and identify flaws, such as fallacies or poorly constructed arguments.

The challenge posed to the teams is particularly relevant in the current context, explains ELLIS, where large language models (LLMs) are used to understand, interpret, and generate texts. Their widespread use by students has raised growing concerns about their potential to foster superficial learning and erode students' critical thinking skills.

In response, this challenge explores a different use of LLMs: instead of providing direct answers, the goal is to explore their value in helping users formulate better questions, specifically critical questions that challenge vague or fallacious arguments to foster deeper reasoning.

Furthermore, this approach intersects with ongoing NLP research aimed at addressing misinformation. Instead of relying on LLMs for objective information, this competition focuses on whether we can leverage their generative capabilities to identify what is missing, questionable, or unfounded in an argument.