At the opening of 2017

News from WBAI
2017.01.01

The Whole Brain Architecture Initiative (WBAI, the NPO) extends New Year’s greetings.

WBAI was established in August 2015 with its mission to promote the WBA approach ‘to create a human-like artificial general intelligence (AGI) by learning from the architecture of the entire brain’ and with the following six objectives:

  • Sustainability: retaining its objectives towards an expected advent of AGI (beyond 2030).
  • Publicness: publishing research results and developing software with an open license.
  • Cooperation with academia: promoting interactions among academic areas such as neuroscience, cognitive science and machine learning, with AI as the pivotal area.
  • Human resource development: fostering multi-domain knowledge required for WBAI research.
  • Infrastructure research: creating platforms for machine learners, evaluation methods for AGI, and research environments for simulation and data preparation.
  • Social activity: making advanced AI technologies transparent for the public to create the future with AI.

Last year, we carried out various activities in support of these objectives with the help of people including the supporting members, the WBAI Supporters, and those who helped us organize events such as the hackathon.

Looking back 2016

Trends around AI (domestic/overseas)

In 2015, organizations such as DeepMind, GoodAI, and OpenAI made their direction in developing AGI clear. In 2016, DeepMind, having declared the development of rat-level AGI, published a string of articles on their R&D, which seems to be focused on learning from the brain. DeepMind made presentations at SfN (Society for Neuroscience), the world‘s largest neuroscientific conference, and proposed a model of working memory with Dr. Hinton, based on fast synaptic plasticity found in neuroscience in recent years. While major AI-related companies have not made their AGI research public, it is quite possible that they are making their investments in areas where AGI’s feasibility augments.
In 2016, domestic activities were also intensified with the start of the five-year project: Comparison and Fusion of Artificial Intelligence and Brain Science and the MEXT Grant: Post K ‘The Whole Brain Simulation and Brain-like AI’ (cf. related article), Sony’s capital acquisition of Cogitai, and the inauguration of AIX Research Center at the University of Electro-Communications and Deep Insights, Inc. (the AI engine developer).

WBAI’s activities in 2016

In 2016, WBAI organized the following activities: the first WBA symposium focused on the vision of “the world in which AI exists in concord with human beings”; five WBA seminars; the second WBA hackathon and the “Communal making of Cognitive Architecture” event coordinated with the Japanese Society for AI; a session at ICONIP2016; sessions at JSAI2016; presentations at academic events such as the 19th workshop in Information-based Induction Sciences (IBIS2016); a meeting with investors and entrepreneurs; the startup and expansion of SIG-WBA; the development of LIS (Life in Silico), a learning environment simulator and BiCAmon (visualization software of WBA activities) in cooperation with Dwango AI Lab. Here, note that LIS, our AGI-fostering environment used in the hackathon and in other events, could contribute to the democratization of AI development.
In 2013, when the WBA movement started, we emphasized the importance of AI and made efforts in vitalizing the field. In 2016, as its importance had been generally established, we made more efforts to promote AGI and brain-like AI.

Towards future evolvement

If one considers the possibility of further acceleration of technological advances, it becomes quite important to think of the first AGI, toward the realization of the vision of “the world in which AI exists in concord with human beings.” We would, at the very least, like the first AGI to be congenial and transparent to human intelligence. So, as written below, it would be desirable that the first AGI be brain-like.

WBA as AGI in concord with human beings

In quantitative aspects of intelligence, AGI will overwhelm human beings. If AGI is heterogeneous to us, it may be difficult for us to understand, to embed ethics as a member of society, and to maintain control. Thus, we could say it is a relatively safe choice to build the first AGI in a form similar to us. Moreover, to make the development feasible, it would be effective to build it in a form similar to the brain.
As citizens deepen their understanding of the greater effects of AGI, they will want it to be the commons of humanity, to avoid the risk of its being monopolized by particular organizations. On the one hand, as investments in AI from profit-making enterprises increase, more people will be engaged in R&D in a closed form as organization personnel. On the other hand, the mechanism of AGI should be made public to guarantee the transparency of AGI. Meanwhile, neuroscience advances rapidly, and the increased understanding of the brain now has more value as a scaffold for building intelligence. Here, the transparency of AGI development will be heightened if WBAI promotes its democratic development with brain architecture as the field of open co-creation from the public standpoint as an NPO.

For the fastest development

Even if WBA is desirable, unless it is completed as one of the earliest AGI, it would not contribute to the vision of “the world in which AI exists in concord with human beings.”
For that matter, we could expect WBA development to be the fastest path for AGI building, if the field of open co-creation has been established. This is because distributed collaborative development of architecture based on the ever-advancing understanding of the brain would promote itself, once it is established. It would also make cooperation of people easier for the democratic development of AGI to make it into the commons of humanity. Moreover, while the construction of large-scale systems often suffers from the problem of final stage integration, it would be mitigated if development is done along brain architecture from its earlier stages. Here, one could complete a WBA system with smaller computational cost (including training) by following the policy ‘to begin with a model the coarsest possible and make it detailed if necessary’ with regard to whole brain modeling.

Generality can be learned from the brain.

A large part of motivation for referencing the human brain in AI development is to obtain clues from the brain that could lead to computing functionalities yet to be realized. We already know how to construct functionalities by learning from data, which is difficult for humans to design. Areas that cannot be covered by learning may have functionalities acquired through biological evolution, the product of larger computational processes.
While narrow AI demonstrates good performances in particular task areas where data is abundant, the realization of generality is still difficult. Why is that? When human beings try to construct a large-scale intelligent system, they tend to decompose functionalities for specific tasks and combine parts to accomplish them. This approach cannot be taken in designing AGI, for it aims for problem solving with diverse tasks that are not foreseen. Meanwhile, as generality is realized in the brain, the design approach to learning from the brain appears to be promising.

WBAI’s activities in 2017

Beginning in 2017, WBAI will develop Whole Brain Connectomic Architecture (WBCA) based on the mesoscopic connectome as a whole brain reference architecture in collaboration with Dwango AI Lab. It will also construct a platform for distributed development to integrate machine learning (ML) modules toward the third hackathon in autumn. Eventually, we aim for realizing open co-creative WBA development with more than ten thousand developers.
WBAI also promotes catching up emerging ML technologies to be integrated into WBA. In catch-up, pre-existing technical information could help development using smaller computing resources by limiting meta-parameters. In this phase, technological enclosure is rare and open collaboration can be made more easily.
The feasibility of activities toward this end is based on the situation in which ML technologies tend to become commodities, and many high-skilled engineers interested in AI can be found domestically.
Adding to the above-mentioned activities, WBAI will communicate with the general public and policy makers, in cooperation with related academic entities, on new ideas, changes and breakthroughs as a public NPO with expertise. It will also expand and maintain the open field in which engineers and researchers interested in AGI can communicate and share information and promote development so that AGI inspired by the brain may be realized as soon as possible. It will carry on its educational business through seminars, SIG-WBA, and meetings with business people for the application of AGI, building on developments from last year.

Conclusion

In the situation where technologies exponentially advance, a minor factor will make a great difference at the time of AGI’s completion. Being in the field of constructing AGI, it is clear to us that the effects and risks resulting from technological advances, whether slower or faster than people’s expectations, should not be belittled.
If advances are slower, we run the risk that people will lose interest, R&D will stagnate, and preparation for the future will be neglected. In such a situation, WBAI will play a larger role in supporting the R&D of AGI on a long-term basis as we originally envisioned.
If technological advances are faster than expected, we will run the greater risk that society will be left behind, leading to a variety of problems. In such a situation, a larger role for WBAI will include guidance and promotion toward the direction in which the first AGI will be in concord with human beings.
While in the turmoil of rapid change, our standpoint is inevitably in flux. We would like first to create an open platform to be the field of co-creation for WBA, exploiting our public status as an NPO. WBAI will be pursuing its course with you through the above-mentioned activities for “the world in which AI exists in concord with human beings.” Thus, we believe WBAI can assume an important role in Japan and the world.
January 1st, 2017
From the members of the Whole Brain Architecture Initiative (the Specified Non-Profit Organization)