AI & Machine LearningReinforcement Learning

Learning by Trial and Error

An AI taught itself to dig a tunnel in Breakout

Nobody told it the rules, or even what a ball was. So how did a program work out a trick in Breakout by itself?

You'll find out

  • What a program learns from points
  • How it found the tunnel trick
  • When points teach the wrong lesson

Why it matters

Points and penalties train game players and robots. Knowing how shows what they really learn, and what they miss.

About 3 min6 interactive slides · 5 questionsLesson 287 · Zone 32

Play this lesson

Free to play in your browser or on iPhone

More in this zone

  1. 280Game GraphicsWhy Old Games Looked Better on Old TVsPixel art was drawn for blurry screens
  2. 281Text AlgorithmsHow Autocorrect Knows What You MeantIt measures how many edits separate your typo from real words
  3. 282Fluid MechanicsHow Steel Ships FloatA ship floats when it pushes aside its own weight in water
  4. 283StatisticsThe Cover JinxAthletes often slump after appearing on a magazine cover
  5. 284Set TheoryBigger Than InfinitySome infinities are larger than others, and there's a proof you can follow
  6. 285Audio TechnologySound You Can SeeA record's groove is the shape of the sound wave
  7. 286Interface SymbolsThe Floppy You Never SawThe save icon is a disk most people never used
  8. 288RoboticsThe Uncanny ValleyAlmost-human robots creep us out more than cartoon ones