What is the theory in which observed data will become more accurate with more samples?

For example a survey will smooth out and be a more accurate indicator if more participants are used.