Something’s Rustling in the Grass: Signaling Games in Ambiguous Contexts
Presentation Date
7-2015
Abstract
We demonstrate via agent-based simulations that agents playing a signaling game with ambiguous observations develop conventions for transmitting information, despite the ambiguity. In a signaling game, one player observes a world state and sends a signal. A second player, who cannot observe a world state, acts according to the signal. If the action is correct, both players get a payoff; otherwise, neither gets a payoff. Signaling games are generally used to study conventions that arise from definitive observations, as when a vervet monkey observes a snake and sends an alarm call. Here we investigate four conventions that arise from ambiguous observations, as when a vervet monkey observes rustling grass (snake or leopard?). In one convention, agents treat ambiguous observations as definitive. In another, agents use multiple signals. In a third, agents use multiple actions. In the last, agents use multiple signals and actions. For each convention, our agent-based simulations show that agents accrue more payoffs than if they were merely guessing. These results apply to the philosophy of language and have broad applications to signaling systems of nature.
Something’s Rustling in the Grass: Signaling Games in Ambiguous Contexts
We demonstrate via agent-based simulations that agents playing a signaling game with ambiguous observations develop conventions for transmitting information, despite the ambiguity. In a signaling game, one player observes a world state and sends a signal. A second player, who cannot observe a world state, acts according to the signal. If the action is correct, both players get a payoff; otherwise, neither gets a payoff. Signaling games are generally used to study conventions that arise from definitive observations, as when a vervet monkey observes a snake and sends an alarm call. Here we investigate four conventions that arise from ambiguous observations, as when a vervet monkey observes rustling grass (snake or leopard?). In one convention, agents treat ambiguous observations as definitive. In another, agents use multiple signals. In a third, agents use multiple actions. In the last, agents use multiple signals and actions. For each convention, our agent-based simulations show that agents accrue more payoffs than if they were merely guessing. These results apply to the philosophy of language and have broad applications to signaling systems of nature.