There is a large gap between current AI systems and human intelligence. We study intelligence with a focus on metacognitive activities while following the recent development of deep learning, and eventually aim at their concrete implementation through three approaches: Introspection, Constructivism and Neuroscience.
Metacognition and Curiosity for Efficient Exploration
Many tasks have sparse rewards in the world. Random exploration for sparse rewards, however, is computationally expensive. A human agent explores the world by seeking sparse rewards efficiently. We focus on metacognition and curiosity as the crucial factors for such efficient explorative strategies. We proposed experience sampling methods to approach metacognition [1,2]. We demonstrated our experience sampling methods to collect introspective experience effiently. We also investigate neural mechanisms of curiosity for such an efficient explorative algorithm and its relation to metacognition.
 Niikawa, T.*, Miyahara, K., Hamada, H.T., Nishida, S. (2020) A new experimental phenomenological method to explore the subjective features of psychological phenomena: its application to binocular rivalry. Neuroscience of Consciousness, 2020 (1), niaa018. https://doi.org/10.1093/nc/niaa018
 Miyahara, K.*, Niikawa T., Hamada, H.T., Nishida, S. (2020) Developing a Short-term Phenomenological Training Program: A Report of Methodological Lessons. New Ideas in Psychology. 58.
Logical Tasks for Measuring Extrapolation and Rule Comprehension
Logical reasoning is essential in a variety of human activities. A representative example of a logical task is mathematics. Recent large-scale models trained on large datasets have been successful in various fields, but their reasoning ability in arithmetic tasks is limited, which we reproduce experimentally. Here, we recast this limitation as not unique to mathematics but common to tasks that require logical operations. We then propose a new set of tasks, termed logical tasks, which will be the next challenge to address. This higher point of view helps the development of inductive biases that have broad impact beyond the solution of individual tasks. We define and characterize logical tasks and discuss system requirements for their solution. Furthermore, we discuss the relevance of logical tasks to concepts such as extrapolation, explainability, and inductive bias. Finally, we provide directions for solving logical tasks.
Fujisawa, I., Kanai, R. (2022), Logical Tasks for Measuring Extrapolation and Rule Comprehension, arXiv. https://doi.org/10.48550/arXiv.2211.07727