The Goodness of Memoryless Justice¶
Society plays a role for all of us in learning to become more virtuous humans. Each interaction shows us the highs of compassion, the complications of lying, or the emptiness of selfishness. A good society is a society that helps its members to become good. Society can do this by encouraging good behaviors and discouraging bad behaviors. However, not all encouragement is equally effacacious.
Have you ever had a friend or been in a relationship with someone who would constantly rehash old wounds in new fights? Maybe they thought you were careless or insensitive this morning, and when you show that you weren't, some black mark from your past is dredged up and paraded around in front of you. The discouragement you getting from that person would be relatively ineffective for several reasons:
- you have changed as a person -- it feels like getting punished for something someone else did
- you are in a different frame of mind; you can't actually weight the negatives against whatever positives you saw (SGD with lag)
- it is more complicated to learn causation over a larger window (sample complexity)
- the discouragement no longer feels like it's actually caused by your behavior -- it feels like your friend or partner is just looking for an excuse, and would be mad anyway
The more consequences can occur as soon as possible, the less a
Causal Determination and Causal Depth¶
(learning simulations: SGD with lag, sample complexity, false signal?)
In machine learning, ... [sample complexity].
sample complexity and NN depth?
The more complicated a model is, the longer it takes to learn. In the instance of Societal feedback, when more recent events are responsible for outcomes, citizens can learn more quickly. When less recent events are responsible for outcomes, the complexity of understanding exactly why it went wrong and how things could have happened differently