From Search to Ask to Act: The Evolution of Information Access in the Age of Large Models and Agents

Abstract

Information access has evolved from search to ask to act. For more than two decades, information retrieval made networked information usable through crawling, indexing, ranking, user modeling, and recommendation. My early work on Web page understanding, vertical search, learning-to-rank, personalized search, and recommendation followed this user-centered view. This keynote briefly revisits that trajectory as the context for a large shift recently: from retrieving documents to constructing systems that reason over evidence, generate grounded answers, and execute information-seeking tasks. I will discuss WenLan, YuLan, and LLaDA as examples of how multimodal foundation models, large language models, and diffusion language models reshape the interface between retrieval and generation. More importantly, I will focus on retrieval-augmented generation and AI search agents. FlashRAG provides a modular open-source platform for RAG research; Search-o1 and WebThinker connect large reasoning models with active search and deep research; and DeepAgent studies scalable tool use for general reasoning. Together, these works expose new IR problems: when and what to retrieve, how to align model and retriever knowledge preferences, how to refine long-context evidence, how to maintain memory and source credibility, and how to evaluate multi-step search behavior. I will close by arguing that future IR systems should not stop at answers. Large-model-driven agents can plan, search, verify, call tools, and support decisions. RAG and AI search agents turn information access into an iterative process of evidence gathering, grounded reasoning, tool use, and responsible action, linking retrieval to decision support and human-centered applications.