by lsusr1 min read10th Nov 20195 comments



Some things can be described only via experience.

  • Direct sensory experience (such as the color red)
  • Foreign untranslatable words and phrases
  • Rasas
  • Certain meditative states (such as kenshō and satori)

Other things cannot be precisely described at all.

  • Any particular noncomputable number

Indescribable things cannot be described in a finite number of words. That's because each one contains an infinite quantity of information. I don't mean they convey this information all at once (except for noncomputable numbers). Rather, they open up a new channel of information.

Opening up a new channel of information is mathematically equivalent to adding an input node to a neural network. This is a totally different process from training a machine learning system. When you train a neural network you adjust the weights of the connections between neurons to solve a problem. Adding hidden nodes amounts to basically the same thing. Adding an input node to a neural network unfolds a new dimension of the problem space thereby adding information for the network to work with. In other words, adding a new input node doesn't improve your solution to a problem at all. It makes the problem easier instead.

For example, the traditional Chinese method of teaching strategy involves memorizing ancient Taoist texts. This pedagogical technique is off the radar of modern MBA programs for reasons independent of its effectiveness. The "memorized passages of concise time-tested wisdom" input node is just missing.

You can't tell someone an indescribable thing but you can tell her where to look. You can tell someone to identify the color of blood, listen to a piece of music or stare silently at a blank wall until you see the essence of reality.

The arts of war…cannot be ignored.

―first line of Sunzi's The Art of War

Where else should I look?