Chess (1996)
Kasparov
Go (2016)
Lee Sedol
Computers vs Humans
Gaming AI
- An abstract board game for two players
- Played on a 19×19 board
- Playing pieces are white and black stones
- Stones placed on vacant intersections of board
- Player which surrounds more territory wins
- One of the most complicated games
'Go' game
What is it and why?
'Go' game
What is it and why?
1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000
possible positions
'Go' game
What is it and why?
- Impossible to calculate every possible move on board
- Bruteforce method used by most AIs clearly fails
- Go requires more intuition and experience than just logic
- Becomes necessary to mimic human mind
Alpha Go
What is it?
- A 'General Purpose' AI program that can play Go game
- Developed by Google Deep Mind
- Not a pre-programmed algorithm, it learns from itself and from others
AlphaGo
Reinforcement learning
AlphaGo
What is next?
AlphaGo Master, AlphaGo Zero and AlphaZero
Updates
Games
Starcraft II, Dota II
Future
Climate modelling, disease analysis
AlphaGo
Personal thoughts
Learning from humans does not mean acting like one
AlphaGo