The integration of artificial intelligence (AI) into academics and research is now entering a new chapter. Despite offering high efficiency, the use of Large Language Models (LLMs) is still overshadowed by the phenomenon of 'hallucination'—the tendency of AI to generate information that sounds convincing but is factually incorrect. In the discipline of history, even the smallest error, such as chronological inaccuracies or misidentifying historical figures, can undermine the integrity of the historical narrative being presented.
To address this challenge, research conducted by information systems students at ITB attempted to implement a Retrieval Augmented Generation (RAG) approach. Unlike conventional AI that relies solely on the model's internal memory, the RAG method forces the system to perform a retrieval from a curated document database before generating an answer. In other words, the system acts like a researcher referring to credible archives and literature rather than merely guessing the answer.
Trials focusing on the 1945–1998 period of Indonesian history showed promising results. By integrating history books with academic authority into the knowledge base, the search engine was able to present narratives equipped with specific citations, including document titles and page numbers. This allows users or researchers to perform independent cross-verification, drastically increasing the credibility of the presented information.
Although RAG has proven effective in minimizing hallucinations on specific or rarely discussed historical topics, its effectiveness still depends on the quality of the available document corpus. Kresno Brahmantyo, a public historian, believes that this technology has the potential to become the backbone of future archive digitization. Further developments, such as integration with national archive documents processed through Optical Character Recognition (OCR), will create a historical learning ecosystem that is transparent, interactive, and scientifically accountable.
Ultimately, this approach is not merely a technical upgrade, but a form of AI-mediated public history. This innovation bridges the gap between the complexity of academic historical research and the public's need for fast access to historical information that remains grounded in accurate empirical facts.
