The second edition of the interdisciplinary HumanCLAIM Workshop brought together scholars from various European universities with experts at the VU to explore the human perspective on multilingual AI models. The workshop started with a social event to explore how the artistic installations at the festival of lights reflect the role of AI for the digital society.

On the next day, the invited speakers Miryam de l’Honeux and Arianna Bisazza shared their knowledge on fairness in cross-lingual NLP and language learnability in polyglot models. They showed multiple examples of how current AI models disadvantage low-resource languages and thereby reinforce economic inequalities.

After this motivating kick-off, all participants engaged in poster sessions and discussion groups facilitated by the intimate and focused setting at the 3D room. We shared and developed innovative ideas about identifying biases in multilingual models, and enhancing their cognitive plausibility and generalizability. The workshop focused on the current shortcomings in how AI models represent words and explored strategies for more human-like and efficient approaches.

Organization committee: Lisa Beinborn, Richard Diehl Martinez, Urja Khurana Support: Network Institute

Thanks to Lisa for sharing her account of the humanCLAIM workshop.