[2I1-OS-49] OS-49

10:00 AM - 10:15 AM JST(1:00 AM - 1:15 AM UTC)

[2I1-OS-49-05] AI Safety Discussions in the Journal of the Japanese Society for Artificial IntelligenceTopic Modeling and Trend Analysis

〇Fumito Takahashi¹ (1. Shinshu University)

Keywords:

AI safety,ELSI,dialogue system,recommender system

To clarify trends in AI safety discussions in Japan, text analyses were conducted on papers published in the Journal of the Japanese Society for Artificial Intelligence from 2001 to 2025. This involved extracting words and phrases from paper titles, identifying research topics, calculating growth rates from time-series paper proportions for each topic, and identifying which topics contained words related to AI safety/ELSI/alignment. The analysis showed that only 14 out of 1,302 papers (1.08%) included AI safety-related terms in their titles, indicating nascent stage for AI safety. Among the top five growth rate topics — "knowledge graph, licensing examination," "dialogue system," "agent model, evacuation simulation," "sound symbolism, onomatopoeia," and "recommender system" — only "dialogue system" and "recommender system" contained AI safety-related papers. This study offers insights into areas where further discussion and technical research on AI safety are needed.