RAG's Roots in IR: Tracing the Intellectual Lineage of Retrieval-Augmented Generation

A recent paper on arXiv (2608.08445) disputes the common belief that Retrieval-Augmented Generation (RAG) is a new approach stemming from the shortcomings of large language models (LLMs). The authors contend that fundamental RAG principles—such as the combination of retrieval with language generation, knowledge enhancement, answer validation, and iterative query improvement—were explored in information retrieval (IR) and question answering (QA) research as early as the 2000s, prior to the advent of LLMs. This study meticulously outlines the intellectual heritage linking contemporary RAG and Agentic RAG to earlier IR and QA frameworks, highlighting the reasons for this historical oversight, including community fragmentation and evolving research focuses. The authors advocate for recognizing these historical roots to inspire more informed advancements in the field.