How Can We Live with Overconfident or Unconfident Systems?: A Comparison of Artificial Subtle Expressions with Human-like Expression

How Can We Live with Overconfident or Unconfident Systems?: A Comparison of Artificial Subtle Expressions with Human-like Expression Takanori Komatsu (tkomat@shinshu-u.ac.jp) Faculty of Textile Science and Technology, Shinshu University, 3-15-1 Tokida, Ueda 386-8567, Japan Kazuki Kobayashi (kby@cs.shinshu-u.ac.jp) Faculty of Engineering, Shinshu University, 4-17-1 Wakasato, Nagano 380-8553, Japan Seiji Yamada (seiji@nii.ac.jp) National Institute of Informatics/ SOKEDAI/ Tokyo Institute of Technology, 2-1-2 Hitotsubashi, Tokyo 101-8430, Japan Kotaro Funakoshi (funakoshi@jp.honda-ri.com) Honda Research Institute Japan Co., Ltd, 8-1 Honcho, Wako 351-0188, Japan Mikio Nakano (nakano@jp.honda-ri.com) Honda Research Institute Japan Co., Ltd, 8-1 Honcho, Wako 351-0188, Japan Abstract Expressing the confidence level of a system’s suggestions by using speech sounds is an important cue to users of the system for perceiving how likely it is for the suggestions to be correct. We assume that expressing the levels of confidence using human-like expressions will cause users to have a poorer impression of a system than if artificial subtle expressions (ASEs) were used when the quality of the presented information does not match the expressed level of confidence. We confirmed that this assumption was correct by conducting a psychological experiment. Keywords: Artificial Subtle Expressions (ASEs), Human- like Expressions, Confidence, Users’ Subjective Impressions Introduction Human-machine communication using speech sounds is becoming more common (Cohen, Giangola, & Balogh, 2004; Nass & Brave, 2005) because users can obtain information while engaging in their primary tasks without facing nor manually operating the information providing systems (e.g., intelligent home appliances or car navigation systems). However, due to various reasons (for example, Benzeghibaa et al., 2007), such as noise in the sensors, the incompleteness of data, immaturity of technology, and the complexity of tasks, the reliability of such systems is often limited. Cai & Lin (2010) experimentally showed how expressing the levels of confidence for such systems to indicate whether the system’s represented information is accurate or not to users plays an important role in improving both the user’s performance and their impressions. When intending to express a system’s level of confidence, one can easily have the idea of using human-like verbal expressions such as “probably,” “definitely,” or “83% confident.” However, expressing levels of confidence using such human-like expressions might frustrate users when the quality of the presented information does not match the expressed level of confidence. For example, users might feel frustrated with systems (like car navigation systems) that express a higher level of confidence like “you should follow my suggested route” or “I am 80% confident,” but the represented information was wrong (this is the case of being “overconfident”). Since human-like expressions make users expect higher human-like abilities from the systems (for example, Sholtz & Bahrami, 2003; Kanda et al., 2008), such inconsistent behaviors eventually make them deeply disappointed (Aronson & Linder, 1965; Komatsu & Yamada, 2010; Komatsu, Kurosawa & Yamada, 2011). Related to the above issue, we have proposed artificial subtle expressions (ASEs) as an intuitive methodology for notifying users of a system’s internal state. Actually, the ASEs only have a complementary role in communication and should not interfere with communication’s main protocol. This means that the ASEs themselves do not have any meaning without a communication context. In particular, we showed that ASEs implemented as beep-like sounds succeeded in accurately and intuitively conveying a system’s confidence to the users (Funakoshi et al., 2010; Komatsu et al., 2010a; Komatsu et al., 2010b;). Therefore, we assume that our proposed ASEs are suitable for expressing levels of confidence in comparison to human- like expressions.