Opens in a new windowOpens an external websiteOpens an external website in a new window
This website utilizes technologies such as cookies to enable essential site functionality, as well as for analytics, personalization, and targeted advertising purposes. To learn more, view the following link: Cookie Policy
Brain-inspired AI as a way to desired general intelligenceドワンゴ 人工知能研究所
Brain-inspired AI as a way to desired general intelligence
Hiroshi Yamakawa* Naoya Arakawa* Koichi Takahashi*
*Whole Brain Architecture Initiative
Presented at "Gatsby-Kakenhi Joint Workshop on AI and Neuroscience ", May 12, 2017
Abstract:
It is critical to give artificial intelligence (AI) a general purpose problem-solving ability, as artificial general intelligence (AGI) with such functionality will bring about unprecedented intelligence development by its self-improvement capability.
With the advent of deep learning, Moravec’s paradox began to be eliminated and the current AI is progressing in a bottom-up approach following the ways of phylogeny and ontogeny. However, it is still struggling with intuitive physics that can be grasped by less-than-year-old infants. From this viewpoint, the realization of AGI seems far away.
Meanwhile, from the viewpoint of AI drawing on the brain, the realization of AGI does not seem so far away, as brain functions have been partially realized with artificial neural networks (ANNs). For example, general object recognition with a convolutional neural network (CNN) can be used to model the visual temporal lobe pathway and the delayed reward calculation of reinforcement learning for the basal ganglia. Similar cases can be made for other organs such as the auditory cortex and cerebellum. Those functional models will be gradually integrated to the whole brain architecture with mesoscopic connectomic information.
As the realization of brain-inspired AGI becomes more realistic, we must have a broad perspective in research and development, for its impact on society will be extensive.
Even if AGI goes beyond our intelligence, it will be relatively easy to understand it if it operates on the same architecture as our brain. AGI based on the brain architecture will more likely be a common property of mankind, because the architecture can be agreed upon by many and be used in a widely shared development platform. Thus, in order to build AGI in harmony with human beings, it would be desirable to strengthen the cooperation between the neuroscientific community and AI community and to promote the open development of brain-inspired AI.
INTERNATIONAL FORUM TOWARD AI NETWORK SOCIETY ドワンゴ 人工知能研究所
Panel Discussion: Management of Risks Brought about by AI Networking
In a desirable future, the happiness of all humans will be balanced against the survival of humankind under the support of super intelligence. In that future, society will be an ecosystem formed by augmented human beings and various public AIs, in what I dub an ecosystem of shared intelligent agents (EcSIA).
Although no human can completely understand EcSIA—it is too complex and vast—humans can control its basic directions. In implementing such control, the grace and wealth that EcSIA affords needs to be properly distributed to everyone.
This document discusses the history and goals of AI Lab, an AI research organization. It describes how AI Lab was founded in 1989 to conduct basic research on AI using neural networks. Its current goals are to develop human-level artificial general intelligence through techniques like deep learning and world models. It also aims to ensure AI systems are beneficial to humanity by designing them to be helpful, harmless, and honest.
12. CONFIDENTIAL MATERIAL / RESTRICTED ACCESSCONFIDENTIAL MATERIAL / RESTRICTED ACCESS
認知距離による問題解決#3 ー 環境モデルの導入ー
n 遷移可能状態の選択肢の生成: 環境モデル
n 行動選択: Selector
ai = si,ai[ ]= argmin
i
L si,ai[ ], sG,ag
!" #$( )
11
13. CONFIDENTIAL MATERIAL / RESTRICTED ACCESSCONFIDENTIAL MATERIAL / RESTRICTED ACCESS
認知距離による問題解決#4: 探索処理の統合
ゴール状態(SG)までの認知距離を得られるまで,環境モデ
ルを用いて反復深化深さ優先探索(IDDFS)を行う.
認知距離
FROM 状態
TO 状態
Forward
model
状態(t)
状態(t+1)
S
1
S
4
S
7
S
G
L(S , S )
7
G
Cognitive
distance
Di = argmin
i
L Si,SG( )
12