Analyzing agent strategies for insights into systems
How might we map the winding path of a highway, if we couldn’t see it?
One approach: By tracking the movement of cars traversing it.
What could we learn about a highway in general, if we only had GPS positions for a handful of cars? We could get a sense of congestion and construction zones and speed traps. Where the highway narrows from several lanes, or opens up again. We could even get a (very partial, approximate) sense of the positions of other cars, based on the passing and weaving patterns of the cars we’re tracking.
Sure, some cars off-road, but relatively rarely, and only in situations where either (1) road infrastructure is severely lacking en route to their destination, or (2) the driver’s perception or judgment fails. Why do drivers stay on the road, and obey speed limits? In part for practical reasons: convention is followed because convention is convenient. You drive on the right side of the road because it’s in your interest to follow convention. But in part because there is normative enforcement via honked horns and PD.
Any organism, any agent, carries mutual information with the environment he inhabits. He learns how that environment works through costly trial and error (often over millions of generations) and comes to embody (often very literally, cf good regulator theorem) that environment in his behavior and physical form.
This is a big part of why I’m interested in strategic interaction as a lens. The idea is that people have already done a lot of the hard work learning about how the social world works. This knowledge is mostly tacit and nonverbal. You can’t just ask people how the social world works. But you can look at their actions and reverse-engineer from there.
This is also why I’m interested in learning from children’s games specifically. There was a brief moment on Twitter a few months back where people were trying to square “People in the olden days weren’t stupid” against “The Romans literally used a Caesar cipher to encrypt messages.” The reasoning was, how could anyone be fooled by such a simple cipher? Well, the beginnings of anti-inductive treadmills always seem painfully naive to those further along. Precisely because, looking retrospectively, past strategies are inductive and factored in—are obvious because they are taken for granted in the present game-state among more sophisticated and experienced players. (Not generally sophisticated; sophisticated insofar as they are trained specifically in a suite of cultural practices relevant to excelling at the game.)
Similarly, my gambit is that kids play fundamentally similar games to adults, but are less specifically sophisticated players. They can’t cloak their motives in layers of obfuscation and rhetorical grandstanding. And they don’t really have to. They just have to be n+1 steps ahead of the other kids on the playground. (The same way we only have to be n+1 steps ahead of other adults.) Spoiler, kids’ n is a pretty small number. Can we “decrypt the Caesar cipher” of social gaming?