Computer program that learns to play classic NES games
suckerpinchYouTube ↗
Further reading 3
What if the best-scoring move is to pause the game? →Optimizing a score can reveal what its designer forgot to ask for.Did the game agent see too little or optimize the wrong thing? →A limited input can hide distinctions needed for a decision. A misleading objective can reward a bad move even when the state is visible.A better network or a longer look ahead? →Evolving a network changes its structure. Looking farther ahead changes how an agent evaluates future moves. Each addresses a different limitation.
Caption notes
Learn a progress measure from NES memory traces, then search for inputs that increase it. Looking ahead helps avoid locally attractive but unproductive moves.
Original passages · 5
It's just looking at RAMs and it's trying to find sequences of inputskeep track of a bunch of random future inputs that we might executeplaying Tetris well requires some thinking ahead, and this algorithm does not think very far ahead.this is a really simple, mathematically elegant and stupid techniquethis algorithm does not think very far ahead