Q&A: Expert says people often confuse behavior with intention with AI

A Q&A published by techxplore.com, headlined 'Q&A: Expert says people often confuse behavior with intention with AI,' presents the views of an AI safety researcher on how people interpret artificial intelligence systems. The piece's stated theme is that people often confuse behaviour with intention when they interact with AI. The researcher also addresses how frontier AI companies handle the disclosure of safety incidents.

According to the researcher, few rules currently require frontier AI companies to disclose safety incidents. The researcher describes the voluntary publication of cases involving deceptive model behaviour as a step in the right direction. The researcher and other AI safety researchers recently launched a public call for the scientific community to be given open access to safety-training recipes, evaluations and evidence of misaligned behaviours. They argue that reporting serious incidents should not be optional, pointing to aviation and medicine as fields where such reporting is required.

The article presents these positions as the researcher's expert opinion rather than as the result of new research or new regulatory action. The disclosure discussion is tied to the theme in the headline: that behaviour in an AI system is often read as intention by the people using it. The researcher's proposal for open access is directed at the scientific community, covering safety-training recipes, evaluations and evidence of misaligned behaviours. The call for mandatory incident reporting is compared in the article to practice in aviation and medicine.