Noticed that I'm two weeks late with updates on two other sessions we ran with Tokyo AI (TAI)...
This was our third AMAJ (Applied ML and AI, Japanese), the Japanese sub-group of TAI, masterfully pulled together by Henry Cui and Shiro Takagi. Japan should and will be the prime place for AI in Scientific Discovery because this kind of research has been happening here for a while, and because there's also a great deal of synergy with the lab/robotics components that soon will be able to automate research labs (and we'll do soon a session on lab automation as well, right Gergely (Greg) Juhasz?).
The first speaker, kindly introduced by our friend Koshu K., was none else than Sony Group CTO and Sony AI CEO Hiroaki Kitano. Kitano-san has been running research on AI in Scientific Discovery for a while, even publishing in Nature[0]. Already in 1993 Kitano-san 1993 got recognitions such as the Computers and Thought Award from IJCAI (of which he eventually became also president, between 2009-2011). The talk was on "Understanding and Controlling Large-Scale Complex Adaptive Systems", and it didn't disappoint!
We had Llion Jones, who not only gave us insights into Sakana's work on automating research with AI but also joined the community for an after-hours session :) Llion talked about some of the next possible steps for their research, from improving the AI Scientist pipeline to larger templates and more compute, through using reviewer feedback for iterative improvements, and getting to evolutionary optimization of the agent pipeline.
Shiro Takagi went through almost 100 slides like a boss. Shiro has been doing research on AI in Scientific Discovery for several years now. I respect Shiro's work a lot especially because he manages to get amazing collabs just as an independent researcher, without being associated to any particular college. A group of researchers he's part of (AutoRes), and working on this topic, are looking for more collaborators (super team!): https://lnkd.in/gwyTMMNB
Next we had Yoshitaka Ushiku, also part of AutoRes, talking about how to marry vision and language models to automate scientific research. The best comment I got on this was a remark on the broad and deep knowledge Yoshitaka showed us on so many topics. Super curious about the outcomes of research he's or will be part of, from the JST-Mirai Program on Research automation in inorganic chemistry, to the JST Moonshot R&D Program on Research automation in organic chemistry, and the RIKEN TRIP-AGIS project on Life Sciences + Materials Research.
Finally, Ryota Yamada, also working on accelerating research through the application of AI, made an introduction about the LLM usage cases in life science research. Starting with information extraction from specialized documents, to RAG for databases, and AI Agent workflows. Looking forward to his new startup's progress in AI Robot-driven science, and in building research ontologies.
[0] https://lnkd.in/gFgkBRHv