*ZOOM* [PhD Thesis Presentation] - Ms. Shoko Ota "Intrinsic Motivation in Creative Activity"
Presenter: Ms. Shoko Ota
Supervisor: Professor Kenji Doya
Unit: Neural Computation Unit
Title: Intrinsic Motivation in Creative Activity
Humans are born to be creative. Humans spontaneously make an effort to create something new and meaningful even without obtaining any extrinsic rewards. Intrinsic motivation is a desire for learning for its own sake (Berlyne, 1966) and a fundamental condition for intelligence and creativity (Boden, 1998).
Researches have found important drives for intrinsic motivation (Berlyne, 1966; Csikszentmihalyi 1997; Ryan and Deci, 2000), however, little is known about which factors of learning environment are essential for promoting intrinsic motivation when humans engage in creation.
Novelty, variety, and prediction errors are hypothesized as the key factors to explain the mechanisms of intrinsic motivation in learning (Barto et al., 2019; Oudeyer and Kalan, 2007; Schmidhuber, 1991), but those hypotheses are yet to be tested in human creative activities. Here, we propose a hypothesis that intrinsic motivation in creative activity is facilitated when humans can observe higher variety of expressions using simpler rules. Typical examples are composing haiku, solving problems and proving a theorem in physics, and inventing new technologies for new experiences. To examine the hypothesis, a novel human behavioral experiment was designed and conducted with an original computer game based on a framework of the Game of Life cellular automata (Conway, 1970).
The simplicity of a rule is controlled by parameters of state transition function and quantified by the complexity measures formulated in the theory of cellular automata. The variety of expression is quantified by the number of live cells, the number of cell state transitions, the entropy of the distribution of local patterns observed by participants, and empowerment using information theory. The degree of intrinsic motivation is measured by the subjective scores of enjoyment in the questionnaire after the task, the playing time before getting bored, the frequency of touching interaction. These measurements are compared between four rules (Rule 1 to 4) with different simplicity and variety of expression and tested with 42 participants.
The results of two-way ANOVA of the scores of enjoyment showed that participants are more motivated with a higher variety of expression and the simpler rule, while there is no interaction between two parameters. The results of multiple regression analyses also support a part of the hypothesis that the entropy of local patterns is related to intrinsic motivation. We further performed canonical correlation analysis of the variables related to intrinsic motivation versus behavioral measures and the result also supports our hypothesis that a variety of expression promotes intrinsic motivation. Classification of subjects by their preference of the entropy of local patterns provides a better fit of regression analysis, which suggests subtypes of participants favoring different components of our intrinsic motivation models. Comprehensive analyses show that subjects took different attitudes, such as watching autonomously evolving patterns or actively trying to identify the rules, which points to the possibility of further analysis with the distinction of participants based on their types of curiosity using clustering techniques.