AlphaGo Beats Lee Sedol, 2016

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-history-alphago-2016-builders

1. In what city was the AlphaGo versus Lee Sedol match held in March 2016?

   1. Seoul
   2. Tokyo
   3. Beijing
   4. Singapore

2. What was the final score of the five-game match between AlphaGo and Lee Sedol?

   1. 4 to 1 in favor of AlphaGo
   2. 5 to 0 in favor of AlphaGo
   3. 3 to 2 in favor of AlphaGo
   4. 4 to 1 in favor of Lee Sedol

3. How many people watched the match online?

   1. Over 200 million
   2. About 10 million
   3. Around 1 billion
   4. Fewer than 1 million

4. Which game did Lee Sedol win to avoid a shutout?

   1. Game four
   2. Game one
   3. Game three
   4. Game five

5. What was special about move 37 in game two?

   1. It was played on the fifth line and initially seemed like a mistake
   2. It was the first move of the game
   3. It was a move that only beginners would play
   4. It was a move that ended the game immediately

6. What did commentators initially think about move 37?

   1. It was a mistake
   2. It was brilliant
   3. It was against the rules
   4. It was illegal

7. What does the policy network in AlphaGo do?

   1. Suggests which moves are most promising to play
   2. Calculates the exact probability of winning
   3. Searches through all possible future moves
   4. Plays the game without any training data

8. What does the value network in AlphaGo do?

   1. Estimates the probability of winning from a given position
   2. Suggests which moves to play next
   3. Plays games against itself to improve
   4. Searches through possible move sequences

9. What is Monte Carlo tree search used for in AlphaGo?

   1. Looking ahead thousands of possible move sequences
   2. Generating new training games
   3. Evaluating board positions at the end of games
   4. Choosing which network to use

10. What is self-play reinforcement learning in AlphaGo used for?

    1. Improving beyond what human games can teach
    2. Learning the basic rules of Go
    3. Replacing the neural networks
    4. Playing against human opponents

11. How did AlphaGo Zero differ from the original AlphaGo?

    1. It started from random play and used only self-play
    2. It was trained on more human games than the original
    3. It could not beat the original AlphaGo
    4. It required human teachers to learn

12. How long did it take AlphaGo Zero to surpass the Lee Sedol version?

    1. A matter of days
    2. Several months
    3. About a year
    4. Several weeks

13. What did AlphaGo Zero prove about AI systems?

    1. A system can reach superhuman play without imitating human games
    2. Human games are the only way to train AI
    3. Games are too complex for self-play to work
    4. AI can only learn from other AI, not from rules

14. What is the branching factor of Go compared to chess?

    1. About ten times larger
    2. About the same
    3. About ten times smaller
    4. Exactly twice as large

15. What did experts predict before AlphaGo's victory?

    1. It would take at least another decade for a computer to beat a top human
    2. Computers would never beat humans at Go
    3. It would happen within a year
    4. Go was already solved by computers