1. INTRODUCTION
1.1 Background and workshop goals
In today's rapidly evolving digital landscape, the field of information retrieval (IR) is at the intersection of traditional search methodologies and cutting-edge machine learning and artificial intelligence (AI) technologies. As we witness the proliferation of generative AI-driven models, such as diffusion and large language models (LLMs), it has become increasingly evident that the boundaries of IR research are expanding. The ways users search for information and the ways systems recommend information to users are already being impacted by generative AI technologies. Conversely, information retrieval technologies can have substantial impact on generative AI applications in terms of efficiency, effectiveness, robustness, and trustworthiness. Retrieval-augmented language models are just one example of direct impact that has recently attracted considerable attention in both academia and industry.² A few of these areas have been discussed in two visioning perspective papers on “Retrieval-Enhanced Machine Learning” (Zamani et al., 2022) and “Large Language Models and Future of Information Retrieval” (Zhai,
- and multiple workshops including “Search Futures” (Azzopardi et al, 2024) and “Task Focused IR in the Era of Generative AI” (Shah & White, 2024). These areas are becoming increasingly important every day. This growth of activity is why we believe this is the right time to discuss major challenges and opportunities in shaping the Future of Information Retrieval Research in the Age of Generative AI. This has motivated us to organize a visioning workshop on this topic with the support of the Computing Community Consortium (CCC). To explore and navigate this dynamic landscape, we gathered a group of 44 experts across academia, industry, and government in the fields of information retrieval, natural language processing, human-computer interaction, machine learning, and broadly artificial intelligence to Washington, D.C. for this visioning workshop.³ The group came together to outline the future of IR research, with generative AI playing a central role in reshaping how we discover and interact with information. This report is a reflection of the themes, recommendations, and ideas that emerged during the visioning workshop. The workshop used the Chatham House rule,⁴ so the ideas are not attributed to specific participants. This report is a synthesis of the discussions during the workshop and does not include citations as in a traditional journal article. From the visioning workshop discussions, two major themes emerged: (1) enhancing generative AI models and applications using information retrieval techniques, and (2) enhancing information retrieval models and applications using generative AI techniques. We acknowledge that it is sometimes difficult to draw the boundaries between IR and generative AI systems and these two systems are intertwined from many aspects. The ill-defined boundary between the two areas was brought up and thoroughly discussed during the workshop and the participants acknowledge and stress that information retrieval is not (and never has been) limited to current search engine systems, but broadly addresses how users or machines In the last couple of years, Google Scholar indexed over 6,000 papers related to the phrase “retrieval-augmented generation” (as of July 2024). Although we have not counted the number of citations those papers generated, the impact of this area is clearly substantial. See Appendix A.2 for the complete list of workshop participants. https://www.chathamhouse.org/about-us/chatham-house-rule
look for, find, present, access, discover, and interact with information. This creates a large number of challenges that information retrieval and other AI communities urgently need to tackle in the next five to ten years, and corresponding research opportunities that are available. Important real-world applications that can benefit from research in this area include, but are not limited to, search engines, recommender systems, question answering systems, dialogue systems, and intelligent assistants.
We want to stress that this workshop focuses on challenges and opportunities that arise at the intersection of IR and generative AI fields and is not intended to cover the union of the two fields. There are additional issues that arise within generative AI independent of IR and within IR that are unrelated to generative AI. This report makes no attempt to discuss those issues.