Wednesday, December 5, 2018

AGI in Japan, 2018

Here is my report on AGI in Japan for 2018.  (See my reports for 20142015, 2016, and 2017.)

AGI-related sessions such as one on autonomy, emergence, and general AI architecture and another on a new eco-system woven out of humans and AI (both in Japanese) were held at the annual convention of the Japanese Society for AI (JSAI) in June.

The SIG-AGI@JSAI held three workshops (see the event page in Japanese).  Through the workshops, I learned recent advances in Japan including:

  • Susumu Katayama's work on AIXI
    While AIXI has been thought to be incomputable and without implementation, there were actually a computable variant and a Haskell implementation which won an award at General AI Challenge in 2017!
  • Symbol Emergence in Robotics
    Well, I knew their advances from before.  But I learned that one of their results had been pretty well established: Tomoaki Nakamura et al. showed that a robot could learn a (subset of) human language in a multi-modal environment.  This is good news for the realization of human-like AGI.  Somehow I haven't been able to locate introductory articles on it in English in their publications.
  • RGoal Architecture
    The architecture by Yuuji Ichisugi et al. can plan its action (i.e., on the zero-shot basis) based on reinforcement learning.  Again, English articles on it will be listed sometime in their publications.

The Whole Brain Architecture Inititative (NPO), of which I am an insider, had activities such as:
As for brain-inspired AI, an academic project Correspondence and Fusion of Artificial Intelligence and Brain Science has been actively working on research in biological intelligence.