I think the problem with this argument is that our observations are pre-selected before see them. This biases our predictions towards spectacular success/failure stories and exaggerated anecdotes. What would have happened if you were Facebook and thought, "we know social networking startups end in failure, and so we should try something else".