Yasunori Morishima

The functionality of human-machine collaboration systems - mind model and social behavior

As machine evolution continues, human-machine collaboration systems are increasing their importance. In this paper, a social and intelligent agent is discussed as a strategy for improving the effectiveness and efficiency of human-machine collaboration. The proposed approach uses an embedded mind model, which enables agents to interact with the human user consistently, context-dependently, and socially. The relationship between the mind model, personality, and social behavior of the agents is studied through implementation of a prototype system as an e-learning application.

The Role of Affect and Sociality in the Agent-Based Collaborative Learning System

As computer systems are evolving and coming to be regarded as social actors, the importance of social intelligence that enables natural and socially appropriate interactions is gaining a growing interest among the human-computer interaction researchers. This article discusses the definition, importance, and benefits of social intelligence as agent technology. It then describes a collaborative learning system that incorporates agents that are equipped with a social intelligence model. We argue that socially appropriate affective behaviors provide a new dimension for collaborative learning systems. The system provides an environment in which learning takes place through interactions with a coaching computer agent and a co-learner, an autonomous agent that makes affective responses. The social intelligence model that handles affective responses is based on psychological theories of personality, emotion, and human-media interaction, such as appraisal theory and the Media Equation. Experiments conducted with this collaborative learning system to examine the effect of the social intelligence model suggested that users had more positive impressions about the usefulness, the application, and their learning experience when the co-learner agent displayed social responses with personality and emotions than when it did not express them. It should be noted here that the co-learner agent did not provide any explicit assistance for the learner, such as giving clues and showing answers, yet it influenced the user’s evaluation on the usefulness of the learning system. Experimental data also suggest that the co-learner agent contributed to the effectiveness of the learning system.

Allocation of Limited Cognitive Resources during Text Comprehension in a Second Language.

For native (L1) comprehenders, lower-level language processes such as lexical access and parsing are considered to consume few cognitive resources. In contrast, these processes pose considerable demands for second-language (L2) comprehenders. Two reading-time experiments employing inconsistency detection found that English learners did not detect an inconsistency unless contradicting parts were adjacent. This suggests that the text memory that L2 comprehenders can access online is an immediately preceding sentence. This makes a sharp contrast with the finding of previous research that L1 comprehenders appeared to retrieve an earlier portion that was several sentences away. A third experiment with probe verification suggested that L2 participants did not reactivate prior text information when reading the contradicting sentence following an intervening sentence. These results provide evidence for the limited resource allocation for discourse-level processes in L2 comprehension. The results are discussed based on the two-stage model that consists of the resonance and integration processes.

An Implementation of Socially-Intelligent Agents providing Emotional Support and its Application

In this paper, we propose a software platform for realizing applications using socially-intelligent agents. Socially-intelligent agents are software agents which employ social intelligence. Although there exist many types of social intelligence, we focus on a type of social intelligence for providing emotional support. Applying the type of social intelligence, our socially-intelligent agents behave like people by simulating social behavior of people. We show how we realize the socially-intelligent agents through an explanation on the conceptual architecture of our software platform. We also show the mechanism for generating social behavior which makes our software platform unique. According to the theory of media equation, we expected socially-intelligent agents will cause good effect for users. To justify our expectation, we implemented an application of our socially-intelligent agents for eLearning. We give explanation about the application and how socially-intelligent agents work in it. We also discuss our ideas on social intelligence and show our direction for future works.

Elaborations for the Validation of Causal Bridging Inferences in Text Comprehension

The validation model of causal bridging inferences proposed by Singer and colleagues (e.g., Singer in Can J Exp Psychol, 47(2):340–359, 1993) claims that before a causal bridging inference is accepted, it must be validated by existing knowledge. For example, to understand “Dorothy took the aspirins. Her pain went away,” one first computes a mediating idea RELIEVE [ASPIRIN, PAIN]. Then, the truth of it is validated on the basis of existing knowledge. The present study examined the hypothesis that a causal inference would be drawn and validated even when validating knowledge is not familiar or available because elaborations are made to retrieve or construct such knowledge. Experiment 1 showed that people tend to judge naturalness of a text based on causal relations and that causality was indeed recognized in those texts in which the antecedent sentence and the consequence sentence are not linked by familiar knowledge. Experiment 2, which measured sentence reading times, showed that while sentence processing times were longer for such texts than texts involving familiar knowledge, there was no difference between these texts in reading the subsequent sentence describing the validating idea. These results provided evidence supporting elaboration as well as validation of causal bridging inferences.

MIND MODEL AND SOCIAL INTELLIGENCE IN HUMAN-MACHINE COLLABORATIVE SYSTEMS

Abstract Even though social intelligence has not been clearly defined yet, consideration of this new type of intelligence should be important for realizing a new generation of human-machine collaborative systems based on human-centered system design policy. In this article, social intelligence and mind model for implementing socially intelligent agents are studied. At first, emotional and affective aspect of social intelligence is mainly discussed because the aspect is considered as more important in contrast with goal-oriented intelligence. Then, implementation of the mind model and its application as an experiment system is discussed with considering the experiment results.