Source-linked AI summary
Human strategy updating in evolutionary games
Arne Traulsen, Dirk Semmann, Ralf D. Sommerfeld, Hans-Juergen Krambeck, Manfred Milinski
TL;DR
The paper asks which strategy-update rules describe human imitation dynamics in evolutionary games. Using a behavioral experiment with players arranged on a spatial lattice, it measures imitation and spontaneous switching, finding that random strategy changes are more frequent than typically assumed and that the measured rule reproduces observed cooperation dynamics.
Problem
The paper examines which strategy-update rules can describe human imitation dynamics, because update-rule details affect evolutionary-game outcomes such as cooperation.
Method
The authors measure strategy updating in a behavioral experiment where individuals are virtually arranged on a spatial lattice, comparing decisions across heterogeneous and homogeneous neighborhoods.
Results
Spontaneous strategy changes were frequent, imitation probability increased with payoff difference, and simulations using the measured update rule showed no significant difference from experiments after Bonferroni correction.
Takeaways & Limitations
The experimental approach provides measured properties of strategy-update mechanisms for mathematical models of cultural evolution.
Abstract
from arXiv · showhide
Evolutionary game dynamics describes not only frequency dependent genetical evolution, but also cultural evolution in humans. In this context, successful strategies spread by imitation. It has been shown that the details of strategy update rules can have a crucial impact on evolutionary dynamics in theoretical models and e.g. significantly alter the level of cooperation in social dilemmas. But what kind of strategy update rules can describe imitation dynamics in humans? Here, we present a way to measure such strategy update rules in a behavioral experiment. We use a setting in which individuals are virtually arranged on a spatial lattice. This produces a large number of different strategic situations from which we can assess strategy updating. Most importantly, spontaneous strategy changes corresponding to mutations or exploration behavior are more frequent than assumed in many models. Our experimental approach to measure properties of the update mechanisms used in theoretical models will be useful for mathematical models of cultural evolution.
I. RESULTS
Human strategy updating combines imitation with frequent spontaneous or exploratory changes, and the resulting rule reproduces the observed cooperation dynamics. Players also respond probabilistically to payoff differences, with strategy adoption differing between cooperators and defectors.
- Strategy updating: 62% of 5760 decisions initially followed the imitate-the-best rule, while 38% were unexplained by pure imitation and declined roughly 4% per round.The unexplained fraction decreased approximately exponentially, with ν0 = 0.380 ± 0.013 and Γ = 0.962 ± 0.003.
- Cooperation dynamics: 70.0% cooperation with fixed neighbors and 70.6% with random neighbors at the experiment’s start were not significantly different during the experiment.The fixed-neighbor treatment used 15 repeats and the random-neighbor treatment used 10 repeats.
- Cooperation dynamics: A model combining imitation with time-dependent random strategy choice captured the behavioral cooperation levels without a significant simulation–experiment difference after Bonferroni correction.The model used fitted parameters ν0 = 0.38 and Γ = 0.96; fixed- and random-neighbor simulations were nearly indistinguishable at this random-choice probability.
- Imitation and exploration: Cooperators switched spontaneously to defection with probability µC = 0.28 ± 0.07, while defectors switched to cooperation with probability µD = 0.25 ± 0.01 in homogeneous neighborhoods.These changes were interpreted as spontaneous mutation or strategy exploration.
- Imitation and exploration: The probability of imitating a better-performing different-strategy neighbor increased with payoff difference according to p = (1 + exp[−β∆π])−1, with β = 1.20 ± 0.25.Defectors were more resilient to change than cooperators; separate fits yielded βC = 0.67 ± 0.28 and βD = 0.99 ± 0.23, with αC = −0.11 ± 0.23 and αD = 0.79 ± 0.14.
- Strategy updating: The probability of cooperation remained below 50% even when all neighbors cooperated, indicating decisions more complex than simply imitating the most common strategy.This analysis did not take payoffs into account.
II. DISCUSSION
The experiment finds that human strategy updating combines payoff-sensitive imitation with spontaneous random changes, whose probability is higher than typically assumed in theoretical models. Random strategy choice captures the general dynamic trend in this small-group system, while more complex social structures remain for future behavioral experiments.
- II. DISCUSSION: Players imitate others with probability increasing with payoff difference, but also switch spontaneously to new strategies.The paper interprets payoff-sensitive imitation as selection and spontaneous switching as mutation or exploration.
- II. DISCUSSION: The probability of random strategy changes is much higher than typically assumed in theoretical models.
- II. DISCUSSION: Modeling random strategy choice can produce different theoretical dynamics and captures the experiment’s general trend.
- II. DISCUSSION: The experiment studies a small system, although human interactions commonly occur within small social groups.The authors note that behavioral experiments cannot feasibly examine the very large populations often used in theory.
- II. DISCUSSION: The study analyzes the simplest spatial game, leaving heterogeneous, dynamical, and set-structured populations for future behavioral work.
III. METHODS
The experiment recruited 400 students and organized them into groups playing repeated Prisoner’s Dilemma rounds. In the spatial treatment, participants occupied a torus-shaped 4 × 4 grid with four fixed neighbors and made synchronous cooperation or defection decisions.
- III. METHODS: 400 students were divided into 25 groups of 16 players each.Participants were recruited from first-semester biology courses at the Universities of Kiel, Cologne, and Bonn.
- III. METHODS: In the spatial treatment, 16 subjects occupied a periodic spatial grid with four fixed direct neighbors.The torus geometry removed edges, and players were identified by letters to preserve anonymity.
- III. METHODS: Participants used private silent YES, NO, and OK buttons and received written game instructions while anonymity was maintained.
- III. METHODS: Each round required one cooperation-or-defection decision against all four neighbors simultaneously.This setting implements synchronous strategy updating.