Information Ethics (IE), which has been developed since the late 1990s by Luciano Floridi as a comprehensive normative ethics for present-day advanced information society, is becoming increasingly influential in various discussions on ethical issues in information-related fields. In this study, we examined the theoretical framework of IE and considered possible methodologies for implementing it in applied fields such as libraries. First, we identified three theoretical issues to be resolved in IE from the perspective of application: (1) the difficulty of directly applying its fundamental principles, (2) the difficulty of quantitative evaluation, and (3) the ambiguity of the specific implications in the concepts of good and evil. Next, focusing on utilitarianism among the existing normative ethical theories, we suggested that the above issues could be resolved by adopting the idea of rule utilitarianism based on the utilitarian dimensions of IE. In conclusion, we proposed the possibility of implementing the ethical principles of IE to practice by introducing: (1) secondary rules, (2) a quantitative perspective, and (3) a perspective on distribution, after positioning the ethical principles of IE as the primary rules.

　It is not easy for the general public to understand the contents of the Diet proceedings because they contain many technical terms and a large amount of text. If a language model (a technology that allows computers to understand and generate human language) can be used to summarize the Diet proceedings, the understanding of the general public would be facilitated, but currently there are challenges in selecting an appropriate language model. This study aims to select an appropriate language model for summarizing the minutes of a Diet session by comprehensively evaluating summaries produced by multiple language models through a combination of automatic and manual evaluation. First, 20 paragraphs were randomly extracted from the Diet proceedings in 2022 (about 61.1 million letters), and their summaries were created using the language models and by human. Then, the similarity of these summaries and the original sentences were evaluated through the five automatic evaluation tools or manual evaluation. The BERT language model generated most natural sentences with appropriate selection of words, and received high evaluation through both of the automatic and manual evaluations. We conclude that the BERT language model is most suitable for summarizing the Diet proceedings at this moment.