I always thought of any human progress like gradient descent algo. Hindsight is clear, and if current tools/services/methods/processes don't work, look at how i

I always thought of any human progress like gradient descent algo. Hindsight is clear, and if current tools/services/methods/processes don't work, look at how it used to be done. Was it better objectively? May be we took a step in the wrong direction, go back and learn; try to take in a different direction, may be there is a higher ground ahead. It is also important to allow (and be tolerant) to experimentation in different directions because without it; we risk getting stuck in a local optima forever. Opposite of that is Chesterton's fence, there is a reason why we are on this hill, if you go back, it's a steep cliff.