New year’s greetings from the Whole Brain Architecture Initiative (WBAI)!
It was imagined the arrival of Singularity to be in the 2040s in an article ”Envisioning the Decade After Singularity ” by the Chairperson of WBAI (Hiroshi Yamakawa) on the Journal of the Japanese Society for Artificial Intelligence in January 2013, two and a half years before the establishment of WBAI. Meanwhile, extrapolating from the surge of AI-related topics in 2023, 2024 might be called the first year of Singularity.
Looking back on last year, large-scale language models such as ChatGPT and image-generating AIs began to replace human capabilities, impacting and shocking many. This has spurred international AI governance debates as the potentially devastating impact of advanced AI on humanity came into focus. For instance, in May the Center for AI Safety issued a statement that “mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war,” and UN Secretary-General Guterres noted this in June. AI Safety Summit 2023 was held in the United Kingdom in November, and 28 countries, including Japan, the United States, Australia, and China, as well as the European Union, signed the Bletchley Declaration, committing to the safe, ethical, and responsible development of AI. Overall, these developments can be expected to have a positive impact on the future of humanity.
In recent years, the interdisciplinary field of “AI alignment,” to tune highly general AI, so called Frontier AI, to behave according to people’s intentions and values, has been rapidly developing. This field was positioned as AI control and monitoring around 2016. Currently, frontier AI systems are based on machine learning techniques, and important topics are, for example, learning systems appropriately and robustly, including ethical aspects, and the study of interpretability, which allows humans to understand the computations taking place inside the system.
With this regard, what contribution can be made by human-brain-morphic AGI, which WBAI is promoting to develop? We have stated that human-brain-morphic AGI has unique values for applications in interpersonal interaction, modeling of mental illness, and as a vessel for mind uploading.
In interpersonal interaction, it can bring the advantage of “brain-based interpretability” to AI alignment. Brain-based interpretability refers to the property of AGI’s computational processes (including conscious and unconscious aspects) to be understood at a cognitive level in terms of corresponding human neural activity. This is expected to make explicit the internal states (such as beliefs, intentions, pleasure, and suffering) that underlie AI decision-making and behavior (see Figure 1). If an AGI system with brain-based interpretability forms unfavorable intentions or deceptions, we would have a higher likelihood of detecting and addressing them, to help harmonize AI and humanity.
Figure 1 Brain-based Interpretability
To facilitate the creation of human-brain-morphic AGI, WBAI is promoting the construction of human-brain-morphic AGI based on the WBA approach or BRA-driven development. This involves reverse engineering of the neural circuits of the brain to create BRA (brain reference architecture) as blueprint for AGI, and building software based on it. BRA, playing a central role in the development, includes Brain Information Flow diagrams (BIF) constructed from mesoscopic anatomical knowledge, and Hypothetical Component Diagrams (HCD) that show the structure of functional components organized to be structurally consistent with BIF, and model human cognitive behavior.
BRA-driven development can be accelerated by creating the blueprint (BRA) in natural language and utilizing state-of-the-art AI techniques such as LLM. In 2023, we implemented a tool (PDF2BIF) that extracts connections between neural regions and excerpt sentences describing them from the literature.
Human-brain-morphic AGI with brain-based interpretability should be realized as early as possible, even if other forms of AGI are realized ahead. We propose a technology roadmap (Figure 2) to complete the whole brain reference architecture (WBRA) by 2027 by accelerating BIF and HCD construction through the use of LLMs and other technologies.
Some of the results have been presented at Computational Neuroscience 2023, INCF Neuroinformatics Assembly, National Conference of the Neural Circuit Society of Japan, National Conference of the Japanese Society for Artificial Intelligence (JSAI), and SIG-AGI of JSAI.
Figure 2 WBA Technology Roadmap
Last year, we conducted an education business to develop human resources that bridge the brain and AI. We held three WBA seminars (in Japanese) with the themes of “Approaching the Mystery of the Circuitry of the Cerebral Cortex and Its Roles,” “Applications of ChatGPT and other AI techniques in Neuroscience,” and “Considering the ‘Intellect’ of the Brain, Society, and AI from the Self-Constructive Nature of Memory.” The 8th WBA Symposium was held (in Japanese) under the theme of “Development and Significance of Whole Brain Architecture Approach in the Generative AI Era,” where TAGUCHI Shigeru and YAMADA Kyūri gave talks, and in the panel on “the Role of Human Brain AGI for a Good Superintelligence Era,” the participants discussed the significance of human-brain-morphic AGI having an affinity with people built through the WBA approach, in light of the coming age of superintelligence, rapidly approaching with the development of generative AI.
In the Symposium, awards were presented for contributions to activities around WBAI. The WBAI Incentive Award was given to Sūn Yùwěi, who proposed a new algorithm involving the architectural level of distributed neural networks in his research on “knowledge reuse,” a central capability of AGI. The Awards for Meritorious Service for the NPO were given to Sūn Xiǎobái, who contributed to the supervision of the WBA seminar executive committee, and KAMBAYASHI Atsushi, who was the auditor of the NPO and gave advices from business viewpoints.
We shall continue our education business and research and development business to promote our human resource development and R&D support in 2024. In the R&D business, we plan to update the format of BRA data that includes ‘requirement realization graphs,’ which will not only enable flexible computational semantics for a wide range of brain regions in the reverse engineering, but also facilitate its automatic evaluation. We intend to use the new BRA data to promote the accumulation of BRA data for brain regions. Requirement realization graphs are composed of top-down decomposition of individuals’ abilities based on their anatomical structure and the interpretation of the cognitive functions of brain organs.
Last but not least, the various activities promoted by WBAI have been continued with the support and cooperation of many people, including the supporting members, WBA seminar executive committee members, SIG-WBA, advisors, and affiliated organizations. We would like to express our deepest gratitude to all those involved. To note, for these two years, some of our members have been participating in the team “Analysis of Behavior Change using Brain Reference Architecture” of the Kakenhi project “Deciphering and Manipulating Brain Dynamics for Emergence of Behaviour Change in Multidimensional Biology.”
As AI and AGI technologies develop further, we hope to add options to contribute to a future ‘to create a world in which AI exists in harmony with humanity (*) ’,by actively incorporating these technologies into the automation of BRA data construction and promoting the development of human-brain-morphic AGI with superior interpretability. We hope for your understanding of our purposes, and look forward to your continued support and patronage. We also sincerely hope that 2024 will be a year that brings happiness to people, while it could be a year that makes us feel the advent of the Singularity.
Members of the Whole Brain Architecture Initiative
(*) The concept of ‘to create a world in which AI exists in harmony with humanity’ was introduced as a basic idea by WBAI in 2017.”