Inducing perspective sharing between a user and an embodied agent by a thought balloon as an input form SUZUKI Satoshi V. DCISS, IGSSE, Tokyo Institute of Technology / National Institute of Informatics TAKEDA Hideaki National Institute of Informatics / The University of Tokyo http://www-kasm.nii.ac.jp/~ssv/ ssv@ntt.dis.titech.ac.jp This Study Is About... Embodied Agent What is he thinking? A blank thought balloon as a prompt of perspective sharing between the user and the agent User CIA2006 (2006-09-11) ― p. 2 Agenda User's learning about agents Blank speech balloons as a prompt to let a user learn about the agents Experiment Discussion CIA2006 (2006-09-11) ― p. 3 User's Learning about Agents Human-agent mutual adaptation (Yamada & Yamaguchi, 2004) Not only learning by an agent with machine learning, user modelling, etc. But learning by a human interactant is important How can the user learn about the agent? Peer relationship between the user and the agent is important to maintain smooth interaction (Reeves & Nass, 1996) Body-orientation correspondence (Suzuki & Takeda, 2006) Perspective sharing with blank thought balloons CIA2006 (2006-09-11) ― p. 4 Balloons in User Interface History of balloon expression European religious pictures in the Middle Ages (Blanchard, 1974) Comics (Harrison, 1981) Speech balloons and thought balloons Common in expression of utterance/reflection of a character over cultures (cf. Wikipedia:en, ja, fr, de) Balloons in user interface Comic chat (Kurlander et al, 1996) ComicDiary (Sakamoto et al, 2001) Function of blank thought balloons has not been focused on in previous studies! CIA2006 (2006-09-11) ― p. 5 Function of Thought Balloons Containing self-disclosure (Derlega et al, 1993), i.e. what the person is really thinking What is deep self-disclosure? According to Altman & Taylor (1973), information about self including General topics rather than specific topics Original Invisible topics rather than visible topics Weak sides of a discloser Socially undesirable response Strong emotion CIA2006 (2006-09-11) ― p. 6 Experimental Setting No Balloon Input (NBI) Speech Balloon Input (SBI) Thought Balloon Input (TBI) CIA2006 (2006-09-11) ― p. 7 Experiment Participant 39 Japanese undergraduate and graduate students Mean of age: 22.9 (SD3.56), 20 males and 19 females Experimental Environment Interface Laptop PC, USB mouse Macromedia Flash (full screen mode) Participants solved object searching tasks (Wick, 2003), i.e., they searched 4 items indicated from 4 pictures with advice from 2 agents: partner agent (PA) and non-partner agent (NPA) In the NBI condition, they read NPA's response; in the other conditions, they filled in the blank balloon for each picture They distributed 15 points as reward for agents' advice CIA2006 (2006-09-11) ― p. 8 Demo CIA2006 (2006-09-11) ― p. 9 Hypothesis Hypothesis A user should share perspective with the agent which emits blank thought balloons to be filled in, and change attitude so that he/she accept the agent's perspective Prediction P1 The participant should fill in the blank thought balloon longer and deeper than the blank speech balloon P2 The participant should evaluate the NPA higher than PA after filling in the blank thought balloon Measures Length (number of words) of filled in the blank balloons Depth of content filled in the blank balloons (mean value of two judges' evaluation in 5-point scale based on the definition of Altman & Taylor (1973)) Reward for the NPA CIA2006 (2006-09-11) ― p. 10 Result Median (quartile deviation in parentheses) of length/depth of balloon content Pictures SBI cond. (n = 13) TBI cond. (n = 13) Wilcoxon rank sum test length depth length depth length depth 1st 5.0 (1.50) 2.0 (0.25) 9.0 (2.00) 2.5 (0.50) 30.0 36.0 2nd 5.0 (1.00) 2.0 (0.50) 5.0 (1.50) 3.5 (1.00) 80.5 38.5 3rd 6.0 (0.50) 2.0 (0.25) 6.0 (1.50) 3.0 (1.00) 78.5 37.0 4th 6.0 (1.00) 2.0 (0.25) 6.0 (1.50) 3.0 (1.00) 89.0 41.0 Bold: p < .01; Oblique: p < .05 Significant difference in length and depth of balloon content was observed between the SBI condition and TBI condition (SBI < TBI) P1 was supported CIA2006 (2006-09-11) ― p. 11 Result Mean (SD in parentheses) values of reward for the NPA Pictures NBI cond. (n=13) SBI cond. (n=13) TBI cond. (n=13) F(2,36) 1st 8.38 (3.07) 7.00 (3.08) 8.08 (2.47) 0.825 2nd 7.38 (2.06) 7.85 (2.23) 7.15 (3.26) 0.244 3rd 7.77 (2.20) 8.54 (2.30) 6.92 (2.90) 1.37 4th 7.23 (1.36) 8.23 (1.42) 6.77 (1.59) 3.40 Oblique: p < .05, Magenta: p < .05 in multiple comparison Only the significant difference between the SBI condition and the TBI condition was found with one-way ANOVA Moreover, reward for the NPA in the SBI condition was higher than that in the TBI condition P2 was contradicted with this result CIA2006 (2006-09-11) ― p. 12 Lessons Learned Content filled in the blank thought balloons was longer and deeper than that filled in the blank speech balloon It is suggested that the blank thought balloon functioned as a prompt to share perspective between the user and the agent Influence of verbal instruction (answer/think) should be considered However, participant's attitude toward the NPA got worse after filling in the blank thought balloons The participant should became to regard the NPA as his/her “enemy” Social relationships between a user and an agent which emits blank thought balloons should be considered CIA2006 (2006-09-11) ― p. 13 Conclusion It is suggested that filling in blank thought balloons emitted from an embodied agent prompted perspective sharing between the user and the agent Detailed cognitive mechanism of how and why the user accept and reflect agent's perspective is still to be discovered Construction of human-agent social relationships (current work) Application to real problem solving should be attempted CIA2006 (2006-09-11) ― p. 14 Thank You! If you have interests in my study, do not hesitate to contact me (ssv@ntt.dis.titech.ac.jp) Further investigation of influence of human-agent social interaction Embodied agents in mobile/ubiquitous environment (e.g. inter-device agent, mobile display) Perspective-taking environment to "Think different" Possible application with human-agent social interaction Computer-supported learning environment to keep learner's motivation Decision making environment to let a user deeply consider information CIA2006 (2006-09-11) ― p. 15