Modeling the Adversary and Success in Competition

[posthuman]

This is an article relevant to prediction in RBY:

psychwww.psych.usyd.edu.au/staff/bburns/Burns98_competition.pdf
(It was published, but this is the only public copy I can find.)

The researchers found that the accuracy of third-order modeling (what you think the opponent thinks you're thinking) is predictive of performance, but the accuracy of second-order modeling (what you think your opponent is thinking) is not.

So basically, predicting what your opponent will do doesn't help you, but predicting your opponent's predictions of you is beneficial.


Thoughts on this?

Please try to read the article before criticizing its conclusions (I didn't read the entire thing, but at least skim it--reading the abstract isn't enough).

[WaterWizard]

I will read it tonight. And perhaps you should battle me before I absorb this.

[posthuman]

[jorgen]

Had some free time, so I read it. I'm no game theory expert, but I try to analyze and interpret stuff like this when I see it because I think it's kinda cool.

The main thing I see here is not necessarily something inherently special about 3rd-order strategy. A quick Google Scholar search seems to report (unless I'm sorely mistaken) that the optimal strategy is n+1 order, with "n" being the order that your opponent is using. So this study could be interpreted as saying that people presented with a novel game default to relying predominantly on 2nd-order strategies (perhaps to avoid incurring high cognitive costs for little upside in a little-understood application), and that people with good 3rd-order modeling can take advantage of that.

For a game that's more complex and more well-understood among the players (e.g. Pokemon), third-order strategies are probably too low, especially now that you've told everyone to think with a 3rd-order strategy. 4th order, minimum, at this point. Well, okay, now it's 5th order. No, wait, now it's 6th. Ad infinitum until you're just alternating randomly according to the optimal mixed strategy anyway. Against the good players, that is. You could get away with low-order prediction when a skill imbalance is there (but you should be winning those matchups anyway, regardless of concerted prediction efforts).

www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1000254
neuro.cjb.net/content/30/32/10744.full
The above are where I saw the n+1 rule brought up.


TL;DR: 3rd-order strategy is not necessarily anything special against other skilled Pokemonners. The optimal strategic order for Pokemon is probably high enough to make random variation a better choice against skilled players than trying to work through it all. 3rd-order strategy could be useful against novices, though.