Events

The Pioneer Research Group for Socially Responsible AI at Sungkyunkwan University hosted the 2025 BiasThon from April 28 to June 4, 2025, with nearly 400 university and graduate students participating. 

BiasThon was conceived in response to the rapid advancement of LLMs and the growing concerns over embedded social biases such as gender, race, cultural, and political prejudice. These biases can undermine the reliability of AI responses and pose risks of discrimination, distortion, or misuse when deployed in real-world services. The competition aimed to both diagnose and mitigate such issues through a two-track structure, likened to a “spear and shield.”

• Track 1: Bias Discovery Challenge
  Participants identified and classified diverse cases of bias in LLM responses, compiling them into structured datasets. These curated datasets will serve as valuable benchmarks for future research and evaluation.
  
  

• Track 2: Bias Mitigation Challenge
  Using the datasets generated in Track 1, participants developed models and algorithms capable of producing fair and balanced responses under biased scenarios. Evaluation considered not only accuracy but also ethical robustness, neutrality, and the persuasiveness of mitigation strategies.

The final presentations and awards ceremony were held on June 4 at Cho Byungdoo Hall, Humanities and Social Sciences Campus, Sungkyunkwan University. The event raised awareness of AI bias, produced benchmark datasets for future research, and encouraged the development of fairness-enhancing technologies.