By Monte Carlo experiments. One very crude example is:
Construct a circle in a square that just touches all sides of the square.
Put the paper on the floor.
Drop a pencil on it repeatedly and as randomly as possible.
Count the number of times, a, the pencil lands in the square.
Count the number of times, b, the pencil lands also in the circle.
Assume distribution of the landing position is uniformly distributed throughout the square.
The true probability of any drop of the pencil landing within the square also lands in the circle is
(area of circle)/(area of square) and can be estimated by b/a where the larger b is should provide a better estimate.
Thus, (pi x r^2)/(2r)^2 ~ b/a
pi ~ 4b/a
Practically it is hard to implement because you have to 'know' somehow where is the square, so not random and the distribution is not likely to be uniform.
You can try simulating the above using pseudo numbers in computer.