The big lesson: making an AI smart isnt one magic step - its lots of tries, learning from mistakes, and testing every idea to see if it really works. Even the failures taught us something!
At the start it flapped the paddle around randomly and lost every point.
After training it tracks the ball and beats the computer almost every time!
At first it just twitched and missed the ball straight away.
Now it digs a tunnel up the side and smashes bricks from behind the wall!
On day one it fired randomly and let the centipede crawl right into it.
Now it aims and dodges - and it kept getting better as it trained!
At the start it dropped blocks anywhere and made a messy pile full of holes - never clearing a line.
Our best Tetris AI yet! It stacks flat, clears over 1 line per game, and even works on a real phone!
At first the little frog just hops into cars and the river - splat!
It's learning to time the traffic and cross safely, one lane at a time.
On day one the tank spins and shoots at nothing.
It's learning to aim at enemy tanks and take cover.
At the start Q*bert jumps off the pyramid and falls to his doom.
It learns to hop across the cubes and change their colours for points.
First she wanders into the ghosts again and again.
She learns to gobble dots, dodge ghosts, and grab the power pellets!
At first it just sits there getting blasted by the aliens.
It learns to dodge the bombs, hide behind shields, and pick off the aliens row by row!