While the majority of decision theory focuses on long-term optimality, we take to heart John Keynes's idea that in the long run, we are all dead.
Doing so, we did a minor study on possible attacks on expected utility maximization. We performed numerical simulations to demonstrate that certain classes of scenarios with particular state space distributions make expected utility maximization run counter to what one may intuitively call 'rational' and indeed, often result in undesirable outcomes.
We also explored the exploration-exploitation tradeoff by comparing human agents playing a simple game to reinforcement agents playing it under variously parameterized epsilon greedy policies.
This is the reinforcement learning and data analysis part of the study.
The game can be found here.
This writeup is temporary.
- Include link to report
- Include link to game code