RespMultimodal 2026: Responsible Multimodal Foundation Models for Knowledge Discovery

作者：Seungbae Kim, Jinyoung Han, S. Shyam Sundar, Wei Wang, Haewoon Kwak, Jisun An, Sungeun Hong, YunSeok Choi, Soyoung Park, Seunghyun Lee · 年份：2026 · DOI：10.1145/3770855.3818262 · 研究领域：Computer science、Data science、Knowledge management、Engineering ethics、Management science

As multimodal large language models (MLLMs) increasingly act as intermediate reasoning layers in data mining pipelines, they fundamentally reshape how latent structures, frequent patterns, and causal hypotheses are discovered across data streams. Enforcing responsible AI constraints—such as fairness requirements, transparency demands, and regulatory considerations—creates an emerging tension where responsibility protocols may alter, obscure, or invalidate discovered correlations. Rather than presenting isolated model-level debiasing techniques, this workshop frames bias, interpretability, and robustness as core data mining challenges that dictate what can be reliably discovered from data in high-stakes settings. By bridging advanced multimodal fusion with human agency, ethical trade-offs, and policy constraints, this curated program explores the fundamental validity of mediated patterns and establishes a shared research agenda for trustworthy knowledge discovery.