In this paper, I describe Two subjects. First subject is modeling of ego state applicable to the Artificial General Intelligence (AGI). As the reference model of AGI's ego is Human's ego, I investigated the ego state of Transaction Analysis comparing American lifestyle against Japanese lifestyle described in "The Chrysanthemum and the Sword". I propose the standard human's ego state are Personal Child Ego(PC), Personal Adult Ego(PA), Personal Parent Ego(PP), Social Child Ego(SC), Social Adult Ego(SA) and Social Parent Ego(SP). The mother of Artificial intelligence(AI) is knowledge, and AI has same nature of the knowledge. Knowledge is intellectual common property of all people and is social assets. AI is understood as social assets and SA ego state is applicable to AI. AGI has to have an autobiography memory to keep Identity. My second subject is Swarm Intelligence. Swarm Intelligence can be formed on Mutual understanding among members. First vision and first audio are fundamental information for mutual understanding. Using such information based on common ego state, AI can organize Swarm Intelligence. AI and human can have SA ego state and share their first vision and first audio, consequently they can make their swarm intelligence society.

Intuitive inference and logical inference are two types of inference for human, and many models have been proposed. However, most of them supposed the intuitive inference and the logical inference are realized by different processes. However, no brain area is known yet for the logical inference and few neural model are proposed that can clearly explain its macroscopic process. So, in this study, we propose an integrated model in which the intuitive inference is represented as a search process of in a continuous and distributed associative memory, and is switched to a symbolic inference mode that biases an associative gain when it find values during the intuitive inference search. In this study, we discuss its computational model by an associative memory and show its simulation results.

Episodic memory is the important function of hippocampus, and we can't lack it for our wholesome life. But its theory and role in decision making is not clear yet. In this study, we analyze and discuss on its model that we proposed at the WBAI hackathon. As the result, an association of the episodes and the value enabled rapid action learning with small number of experiences, and takes complementary role with the strong but slow feature of Deep Learning.

To understand the human brain function, qualitative reasoning approach seems quite effective as the brain is so complicated dynamic system. In this paper, we proposed qualitative neuron model with which we described the neural substrate subserving saccadic eye movement. As a results, the simulation showed the burst discharges in the pontine reticular formation that should play an important role in controling the saccadic eye movement. Thus it was confirmed that our qualitative neuron model has high expression power to the phenomenon related to the saccadic eye movement.

大脳皮質の言語野における、意味解析機構の形式的モデルを提案する。このモデルは、計算論的神経科学における大脳皮質ベイジアンネット仮説と、理論言語学における組み合わせ範疇文法の理論を融合させたものである。文の意味は、アドレスと意味表現からなる組の集合で表される。この表現方法のおかげで従来意味論の記述に用いられてきたラムダ計算が不要になり、可変長の値を持つ変数を一切用いずに、統語素性の単一化の機構だけを用いて意味解析することが可能になる。このため、提案モデルは神経科学的に妥当なベイジアンネットでの実現がおそらく容易である。文法と意味の処理は局所化されているため、モデルの一部を破壊することで、ブローカー失語とウェルニッケ失語に似た発話を再現できる。これらの結果は、大脳皮質ベイジアンネット仮説に対する、新たな状況証拠を与えている。

MLSH (Meta Learning Shared Hierarchies) is a meta learning method that divide a policy into multiple sub-policies and a master-policy that picks one of the sub-policies to be actually used. By training sub-policies in advance, master policy can rapidly adjust to given environments. However, MLSH is not suitable for complicated environments. In complicated environments, a number of sub-policies are required, and it is very difficult to train them properly. We propose a method to effectively prune excessive sub-policies to give better chance the other sub-policies to betrained. We demonstrate that we can prune 40 % of sub-policies while preserving the performance.

Reinforcement learning using deep learning and Monte Carlo tree search has been reported to be extremely effective as an artificial intelligence algorithm that is used in AlphaZero etc. and is widely applicable to various games. Since this method is essentially an algorithm that solves the search problem efficiently, it is possible to solve a general combination optimization problem as well as a game. Therefore, in order to deepen the understanding of this method, experiments were applied to combinatorial optimization problem, and the results are reported.

This paper discusses a bottleneck of development of Artificial General Intelligence (AGI) and proposes a hypothesis that data acquisition issues can be the bottleneck of the AGI. To address the data acquisition issues, this paper proposes General Supervisor Database (GSD). Moreover, this paper introduces new intellectual property to realize the GSD and proposes General Supervisor Database by Intellectual Property (GSDIP).