Avaliação do uso de modelagem qualitativa com apoio de agentes aprendizes virtuais na compreensão da dinâmica de sistemas por alunos do Ensino Fundamental

This Dissertation presents a pedagogical intervention proposal based on the use of qualitative modelling, supported by virtual teachable agents (or TA) functionality available in the workbench DynaLearn (www.dynalearn.eu), a tool designed to facilitate understanding of the systems dynamics by students in a class of 8th grade of elementary school at a public school in the Federal District. The virtual teachable agent takes the figure of a Pet whose main function is to make the students feel as if they are 'teaching him' new knowledge. The student learn while teaching the pet during the construction of qualitative simulation models. These models are built in accordance with techniques developed in an area of Artificial Intelligence known as Qualitative Reasoning, in which mathematical functions and knowledge about quantities are represented qualitatively, whitout using numerical data. More specifically, the modelling approach follows the Qualitative Process Theory by Forbus, to represent systems structure and functioning. During the modelling activity, the student can call a second agent, presented as an adult animal similar to the Pet, featured as a quiz show animator (Quizmaster), which asks questions to the Pet about the model contents. At the end of each interaction with the Quizmaster, a table is generated to reporting correct and wrong answers given by the student. From the interaction with the Pet and with the dynamics of the quiz, and the analysis of reports generated by DynaLearn, students are encouraged to reflect and return to reading texts to reconstruct their models. This way, they improve the 'performance of their Pets' (that, after all, is the students’ performance themselves), in an iterative process until all the answers of the Pets are correct. Our hypothesis is that the use of qualitative models supported by virtual teachable agents increases motivation and consequently the performance of students in understanding the system structure and function compared to situations in which students learn by reading texts without the use of models, and build models without the support of TA. The methodology consisted of qualitative and quantitative approaches, respectively a case study in which opinions of students and their Science teacher were recorded in interviews, questionnaires and observations, in analyses of quiz reports, and by means of experimental studies with statistical analyses of the performance of students in pre and post-tests. The results show that the use of the virtual agents significantly improved the students' understanding of the dynamics of the systems of interest, compared to reading texts without the support of models and modeling without TA. Students felt more motivated to study science concepts, and the teacher considered very productive the use of virtual agents for her teaching work. The activities and experiments described here led to the formulation of a proposal for a pedagogical intervention, presented as supplementary material to this dissertation.