Randomness Is Needed to Escape Local Maximums
Randomness Is Needed to Escape Local Maximums
In order to escape local maximums, we need to introduce randomness. Drop yourself in random places and start hill climbing. After many attempts, choose the highest hill. This increases your chances of finding the biggest hill as you have a broader and clearer picture of the actual terrain. This is analogous to using chaos to escape complacent order, allowing you to grow and progress, and is another level of iteration.
Start with lots of randomness and reduce the amount of randomness over time. Hills with similar heights tend to be grouped close together (although you should start with more randomness as bigger hills can also be far from each other). Focus more randomness in these directions as you start to reduce randomness. This allows you to build on things conservatively as you narrow down on the global maximum, since experience gained from climbing these hills will be related.
For example, when choosing a career early on, we are inevitably put at the base of a hill that might have been chosen by our parents or ourselves. However, we don't know everything about the future, or even about ourselves. This might not be the best hill. We need randomness to survey the terrain and find the best hill that we might not even be aware of. Once we have a general sense of the direction of the best hill, we can start biasing our attempts in that direction as we reduce randomness and narrow down our field (i.e., focusing on science, and then narrowing down to a specific field and subfield).
Another example are random mutations that lead to new pathways for species to evolve down that are better suited for the environment.