This project aims to create metrics for assessing the amount of skill involved in various games. Two possible methods are:
1. Use Elo rankings from Board Game Arena. Naively, one would expect that more skillful (less luck-based) games would result in a wider range of Elo scores. This rough heuristic will quickly give us a baseline across many games, but introduces potential issues: different number of players across games, confounded by length of the game, dependent on the population playing, etc. To remedy these issues, we can explore the same question using reinforcement learning agents.
2. Develop an AlphaZero-like agent to learn various games. Training a neural network to will afford us a gradient of AI "players" from random to close to optimal (depending on the game) that each have data on moves and states that they deem good. By looking at these NN moves, one can assess the skill of the game across various levels through the following heuristic (that can be sharpened into a well-