Tracking agreement in debate simulations

Note

This recipe reproduces results from Chapter 6 in [Betz2013].

We want to know whether a pair of agents with randomly initialised belief systems can achieve consensus through argumentation.

Since we also want to understand how different argumentation strategy affect consensus conduciveness, we set up two sets of agents, one equipped with the “attack” and the other with the “convert” strategy.

from taupy import *
attack_population = [Position(debate=None, introduction_strategy=strategies.attack) for _ in range(2)]

convert_population = [Position(debate=None, introduction_strategy=strategies.convert) for _ in range(2)]

Next, conduct the experiment with fairly standard simulation settings:

Warning

The following Python code executes 12 simulations in multiprocessing twice. Unless otherwise specified, multiprocessing will use all CPUs available on your machine. On modern consumer hardware, a simulation with 20 sentences takes about 10–30 minutes, depending on the number of agents. If your machine can execute four jobs at the same time, the worst scenario is 180 minutes execution time for both experiments. A machine with 12 CPUs might do the same in less than 30 minutes.

Hint

The multiprocessing module can show the number of available CPUs on a machine

>>> import multiprocessing
>>> multiprocessing.cpu_count()

attack_experiment = experiment(
    n=12,
    simulations={"positions": attack_population, "sentencepool": "p:20"},
    runs={"max_density": 0.8}
    )

convert_experiment = experiment(
    n=12,
    simulations={"positions": convert_population, "sentencepool": "p:20"},
    runs={"max_density": 0.8}
    )

Let’s now analyse all our data. We are interested in the mean population-wide agreement at each debate stage, and the inferential density of that debate stage – this should tell us whether the continued exchange of arguments leads increases mean agreement between agents, and is thus consensus-conducive.

attack_eval = Evaluation(
    debate_stages=attack_experiment,
    list_of_positions=[e.positions for e in attack_experiment]
    )

attack_eval.densities()
attack_eval.agreement_means()

convert_eval = Evaluation(
    debate_stages=convert_experiment,
    list_of_positions=[e.positions for e in convert_experiment]
    )

convert_eval.densities()
convert_eval.agreement_means()

Our raw data is generated! Now let’s combine this data and plot it:

import pandas as pd
import seaborn

# Add a column to our data indicating which strategy was in use:
attack_eval.data["strategy"] = "attack"
convert_eval.data["strategy"] = "convert"

our_data = pd.concat([attack_eval.data, convert_eval.data])
# Convert data to types recognised by seaborn
our_data["agreement"] = our_data["agreement"].astype("float64")
our_data["density"] = our_data["density"].astype("float64").round(2)

# Plot
seaborn.lineplot(data=our_data, x="density", y="agreement", hue="strategy")