Problem-solving in the Wild

What’s difficult for humans is sometimes easy for computers, and what’s easy for humans is sometimes difficult for computers. Why is this?

Computers do very well with problems that have a well-defined set of rules. Take a game of tic-tac-toe, for example: from placing the first X, computers can easily simulate all possible future states of the game.

If you try to do the same—think about all possible movements that can be done on a single turn, and on each turn after that, until the end of the game—you will probably have a hard time without the help of pen and paper. This is because we can process a much more limited amount of this type of information at a time than a regular computer. However, through practice, you will learn to estimate what’s likely to happen next.

Computers have their limits, too: more complex games create more possible states, which takes a toll on computing power. A regular computer can be a worthy tic-tac-toe opponent just fine, but to train a neural network to the level of an expert player of, say, chess or go—you need a supercomputer (or a completely different approach to computing than our current solutions).

Robotics in natural environments poses an even greater challenge for computing, because for a robot, our environment is complex and unpredictable. To physically move a chess piece, it needs to calculate its distance to the piece, an optimal grip, the weight of the piece, the distance between the piece and the new location on the chess board, any chess pieces in the way, and many more parameters.