At the Opening of 2020

News from WBAI

New year’s greetings from the Whole Brain Architecture Initiative (WBAI)!

While going through transitions including the office relocation, WBAI continued its activities with its basic ideas of high public interest, with the support and collaboration of its supporting members, WBA seminar executive committee (previously WBA supporters), SIG-WBA, advisors, and affiliated organizations and individuals in 2019.

As major activities last year, WBAI held the 4th WBA Symposium (in Japanese) on the theme of “questioning integration by learning from the brain,” conferring the WBAI Incentive Awards to Naoto Yoshida, who showed the emergence of homeostasis from maximizing survival probability, and Haruo Mizutani, who contributed to connectome informatics and Nico2 AI School, and the Merit Award to Takeshi Sakurada, who has contributed to the financial and legal aspects of WBAI’s management since its foundation.

In 2019, WBAI had a booth at the 3rd AI Expo in Tokyo, held the 2nd Aphasia Hands-on, and WBA seminars with the themes: “Computational Psychiatry,” “Free Energy Principle,” “Probabilistic Graphical Model and Brain,” and “Recognition of Sociality” (all in Japanese).  WBAI also sponsored the Animal-AI Olympics, a contest to reproduce versatile animal intelligence (hosted by the Leverhulme Centre for the Future of Intelligence), held two related meetups in Japan (first, second) and awarded the WBA Prize at NeurIPS 2019 to Guillermo Barbadillo, who produced a highly biologically plausible work.  In related projects in the government grant “Brain Information Dynamics,” a model based on our neocortical framework was developed [1] and a robot simulator PyLIS was released.  We also continued to be engaged in the development of software platforms for brain-inspired AI as HPCI research.  Yamakawa, our representative, participated in Beneficial AGI 2019 and discussed AI society to support humanity [2]. (see the FY2019 policy).

Last year, as deep learning applications advanced, the perception that the next target for AI will be its generality came to be common.  Research in continual and lifelong learning to respond to a wider range of tasks advanced, and the research to realize intelligence by combining deep learning technologies as architecture seems increasing.  However, little epoch-making technology has emerged in the machine learning field.
As for the relation between AI and neuroscience, the convolutional neural network for object recognition was associated to the visual area of the brain a few years ago, and models related to natural language processing (such as BERT) proposed two years ago were associated with the language areas of the brain last year.  In this way, the trend of associating models of human-like computational functions with brain activities for the understanding of the brain will be increasing.

In neuroscience, a great deal of knowledge is continuing to accumulate.  For example, the Allen Institute has analyzed the hierarchy between regions in the mouse neocortex [3] and has revealed gene-type homology in human and mouse neocortex [4].  These are important pieces of information for building the brain architecture.

In order for AI to have general intelligence, finding a way to integrate machine learning technology is a major challenge.  In light of the recent technological situation, the approach that WBAI has advocated since its inception, ‘to create a human-like artificial general intelligence (AGI) by learning from the architecture of the entire brain’ will be increasingly effective.

Since its inception, WBAI has been engaged in activities to promote research and development of AGI by learning from the architecture of the entire brain.  In the meantime, technical environments for AGI development are being prepared, and the need for us to play much of it is decreasing. Thus, WBAI is turning to focus on the essential parts of the development of brain-inspired AGI; we decided to organize the knowledge in the neuroscientific community into the whole brain reference architecture (WBRA) to promote democratic joint development in the AI/ML community.

In 2019, we developed the brain information flow (BIF) format for describing WBRA and prototyped a database to store it.  With the advent of WBRA, AI/ML experts who do not necessarily have a deep understanding of the brain will be able to develop brain-inspired AGI using it as specifications.  We also released the GPS criteria to evaluate brain-inspired AGI; so we could be evaluating the biological plausibility of developed AI programs in terms of the criteria and WBRA.

Fig.1: Brain information flow (BIF) format under construction

The BIF format was proposed to hierarchically describe information processing specifications of the entire brain.  Its Circuits represent sub-networks in the mesoscopic level network of the brain, and Circuits corresponding to brain organs are called frameworks.  Connections between Circuits originate from what are called Uniform Circuits representing uniform neural populations. In order to support software design, BIF describes the processing functions of Circuits and the semantics of Connection signals as functionality.

In 2020, we plan to solidify the foundation of WBRA by R&D activities such as trial production of reference architectures in the BIF format for brain organs such as the basal ganglia, software development with the architecture, improvement of ontology in the BIF format, and data visualization in the database. 

In constructing reference data, we would like to ask for the cooperation of experts in neuroscience to advance the conversion of existing neuroscientific knowledge into data.  We will continue to build hypotheses on computational functions where knowledge lacks in the current neuroscience. Research has already been conducted on hypotheses of the path integration function of the hippocampal formation [5] and the reinforcement learning of attention in the basal ganglia and thalamus [6].

In 2020, we will continue to hold seminars and other events to foster human resources who can play active roles in this field, and to publish information internationally.

As for the movement of AGI development organizations, OpenAI, a non-profit organization, established OpenAI LLC as a separate entity and received $1 billion investment from Microsoft in 2019.  This demonstrates the difficulty of continuing to participate in the increasingly intense AGI development race as an NPO in the current economic situation. Conversely, one can see that WBAI has come into a more unique position as a non-profit AGI development organization in the world.

We would anticipate the emergence of a larger organization that will pursue development on the whole-brain architecture approach to embody a “world with artificial intelligence in harmony with humanity,” with revenue from the precursors of brain-inspired AGI to support the development system.  However, it will take more than five years to run up to such a technical situation. For the time being, WBAI will be working to improve the development environment for brain-inspired AGI to tug along the development to the point where it is more realistic, focusing on academic activities that will serve as a hub for transmitting the knowledge of the neuroscientific community to the AI/ML community.

In this way, we at WBAI intend to continue our activities from a long-term perspective this year as well around our research promotion and human resource development projects, and to continue our unremitting efforts.  We thank you for your continued support and patronage!

January 2020
Members of the Whole Brain Architecture Initiative 


  • [1] 三好康祐, 山川宏, 高橋恒一, ESNによる階層的予測誤差モデルを用いた音声Local-Global課題の再現シミュレーション, 信学技法, MBE2019-55 NC2019-46, 2019.
  • [2] Yamakawa, H. Peacekeeping Conditions for an Artificial Intelligence Society. Big Data Cogn. Comput. 2019, 3, 34.
  • [3] Julie A. Harris, Stefan Mihalas, […]Hongkui Zeng, Hierarchical organization of cortical and thalamic connectivity, Nature, 2019
  • [4]Rebecca D. Hodge, Trygve E. Bakken, […]Ed S. Lein, Conserved cell types with divergent features in human versus mouse cortex, Nature volume 573, pp.61–68, 2019.
  • [5]  Ayako Fukawa, Takahiro Aizawa, Hiroshi Yamakawa and Ikuko Eguchi Yairi, Identifying Core Regions for Path Integration on Medial Entorhinal Cortex of Hippocampal Formation, MDPI Brain Science, 2020. (to appear)
  • [6] Yamakawa, H. Attentional Reinforcement Learning in the Brain. New Gener. Comput. (2020) doi:10.1007/s00354-019-00081-z