Nobel Turing Challenge - OECD.org · 1997 Chess 2014 Shogi 2016 Go Complete Information Problem Massive Data Massive Computing Advanced Machine Learning . Process of Scientific Discovery

Nobel Turing Challenge

Hiroaki Kitano The Systems Biology Institute

Sony Computer Science Laboratories, Inc.

Can we build machines that can win the

Nobel Prize?

Photo by Hiroaki Kitano, 2002

By 2050, develop AI systems that can make

major scientific discoveries that worth Nobel

Prize in Physiology and Medicine

Intelligence

Au

ton

om

y

Nobel Turing Challenge

AI Assistant for Scientific Discovery

Current Status

Alfred Nobel Alan Turing

The Turing Test at the Nobel-

quality scientific activities

Grand Challenges in AI

1997 Chess 2014 Shogi 2016 Go

Complete Information Problem

Massive Data

Massive Computing

Advanced Machine Learning

Process of Scientific Discovery

Serendipity

By Accident

Scientific Intuition

6 Photo by Hiroaki Kitano, 2010

Scientific discovery is at pre-

industry revolution level

8

Oda and Kitano, Molecular Systems Biology, 2006

50

We are terribly bad in understanding behaviors of systems

that are high-dimensional and non-linear

Limits of human cognition

12

Moonshot

Redefining Scientific Discovery

Massive search and verification of

hypotheses space

13

How efficiently can we execute?

What are science specific constraints?

What is computational definition of “Serendipity”?

Capability to recognize

patterns that human cannot

recognize

14

15

大規模データと機械学習、知識発見を前提とした医療へ

Deep Clinical Phenotyping

Unsupervised learning for patient stratification

Unsupervised learning for patient stratification

outliers

In transition?

anything special?

RA Patient Data from Univ.

Tokyo Hospital

Capability to perform high-

precision experiments that

human cannot carry out

18

Two Arms Robot for Lab

19

Crowd Robotics Consortium

Machine Creativity?

Capability to explore

hypotheses that human

cannot imagine

Human played games

Possible moves based on past human played games

All possible moves on GO

Human played games

Possible moves based on past human played games

All possible moves on GO

Tabula rasa based generation of games

AlphaGo Zero

AlphaGo

25

The Beatles The Beatles Style

Music

28

Discovered knowledge Human discoverable knowledge

Scientific truth

Discovered knowledge Human discoverable knowledge

Scientific truth

Knowledge that human may to be able to

generate and understand

Evolution of Human civilization is

evolution of tools

Industrial Revolution Borns and Stones

Information Revolution AI: Machines that think

Machines that Creates

Knowledge