AI summary by TwinMind
Selects: A List Of Games You Would Surely Lose to a Computer | STUFF YOU SHOULD KNOW
Stuff You Should Know
56m 5s
Historical Context: The Mechanical Turk (1770s)
- Wolfgang Von Kempelen invented "The Turk" - wooden automaton chess player in Vienna
- Appeared to be mechanical figure seated at cabinet with chessboard
- Could play chess well, move pieces, detect cheating attempts
- When opponents cheated (like Napoleon), would move pieces back or sweep board clean
- Actually operated by small person hidden inside cabinet
- Sparked early ideas about man vs machine intelligence
- Influenced Charles Babbage to develop Difference Engine for automatic calculations
Evolution of AI Chess Programming (1950-2010)
- Claude Shannon's 1950 paper "Programming a Computer for Playing Chess" established foundation
- Two key concepts:
- Evaluation function: numerical scoring of board states
- Minimax algorithm: minimize maximum possible loss
- Programming approach: break chess into rules and shortcuts
- Deep Blue (1997) used this method to beat Garry Kasparov
- First computer to beat human chess grandmaster in regulation match
- Massive computer with intricate programming and chess knowledge
- Major cultural milestone, widely covered in news
Modern Machine Learning Revolution (2014-2017)
- Shift from programmed rules to self-teaching AI around 2014
- Sudden improvement in Siri, Alexa, navigation apps due to new machine learning
- Google DeepMind developed AlphaGo for ancient Chinese game Go
- Go has 10^170 possible board configurations
- Called "most complex game ever devised by humans"
- 1990s prediction: would take 100 years for computer to beat human at Go
AlphaGo Development Timeline
- 2014: Development began at DeepMind (Google AI division)
- End 2016: AlphaGo secretly deployed online, dominated players
- 2017: AlphaGo Zero replaced original, learned in 40 days what took predecessor 2-3 years
- May 2017: AlphaGo beat Ke Jie, world's highest-ranked Go player
- Self-teaching method: AI plays itself millions of times to learn patterns
Beyond Perfect Information Games
- Perfect information games: chess, Go (all information visible)
- Imperfect information games: poker (hidden cards, bluffing required)
- Carnegie Mellon's Libratus AI (January 2017):
- Played 120,000 hands Texas Hold'em vs 4 professional players
- Won by $1.7 million equivalent
- Poker player quote: "felt like playing against someone who could see my cards"
- University of Alberta's DeepStack also mastered poker
- Both systems developed intuition through self-play, mechanism not fully understood
AI Gaming Achievements
- Google DeepMind tested on 49 Atari 2600 games
- Miss Pac-Man most difficult, achieved record 99,900 points
- Hybrid reward architecture: senior manager agent + 150 worker agents
- Hierarchical decision-making system with confidence levels
Creative AI and Future Predictions
- Survey of 350+ AI researchers predicted:
- 10 years: AI drives better than humans
- 2049: AI writes bestselling novel
- 2053: AI performs surgery better than humans
- University of Falmouth's Angelina AI creates original game concepts
- Creativity becoming new frontier after chess/Go mastery
Solved Games vs Unsolved Challenges
- Solved games: perfect play always leads to known outcome
- Tic-tac-toe: always draws with perfect play
- Connect Four: first player always wins with perfect play
- Checkers: solved by Chinook computer in early 1990s
- Physical world challenges remain difficult:
- Tag, Jenga, dictionary games
- DARPA robotics challenge: robots struggled with basic tasks like opening doors
- Bipedal robots still fall frequently despite advances
Existential Considerations
- Stock trading algorithms execute thousands of trades per second
- Market crashes can last less than one second due to algorithmic trading
- European Union debated granting personhood to AI
- Elon Musk warns of rapidly closing window for AI safety measures
- Invisible online AI systems pose greater threat than physical robots
- Machine learning systems often operate beyond human understanding of their processes
