Interesting, you’ve turned the game into an optimal stopping puzzle. What I find interesting is that a player could determine their optimal strategy if they had an accurate probability distribution for how many moves it will take them to lose the game by opening a cell with a mine. Players who are better at the Minesweeper aspect of the game will, of course, be able to do better than others by finding more mines.
What I don’t get is this: what is “Bayesian” about it? It’s really more of a utility maximization/optimal stopping game.
It is, as you’ve already pointed out, a matter of perspective. A player (like me, I’m ashamed to admit) who considers minesweeper an instinct akin to breathing will be able to focus on the dilemma between taking a risk or stopping, if they reach a point where taking a risk is required to progress. Of course, a minesweeper addict capable of playing without shutting their mind off (wait, that’s why I’m playing it in the first place!) will recognise the mine and population densities as prior probabilities, and try to workout the conditional probabilities, and how they’re likely to impact the outcome.
Makes sense. The title caused me to assume that there was going to be some surface-level Bayesian updating worked into the core gameplay mechanic and I guess I was surprised when there wasn’t.
Interesting, you’ve turned the game into an optimal stopping puzzle. What I find interesting is that a player could determine their optimal strategy if they had an accurate probability distribution for how many moves it will take them to lose the game by opening a cell with a mine. Players who are better at the Minesweeper aspect of the game will, of course, be able to do better than others by finding more mines.
What I don’t get is this: what is “Bayesian” about it? It’s really more of a utility maximization/optimal stopping game.
It is, as you’ve already pointed out, a matter of perspective. A player (like me, I’m ashamed to admit) who considers minesweeper an instinct akin to breathing will be able to focus on the dilemma between taking a risk or stopping, if they reach a point where taking a risk is required to progress. Of course, a minesweeper addict capable of playing without shutting their mind off (wait, that’s why I’m playing it in the first place!) will recognise the mine and population densities as prior probabilities, and try to workout the conditional probabilities, and how they’re likely to impact the outcome.
Makes sense. The title caused me to assume that there was going to be some surface-level Bayesian updating worked into the core gameplay mechanic and I guess I was surprised when there wasn’t.
I think the “Now 20% more Bayesian!” slogan is just marketing to the LessWrong audience.