Machine Culture Empirica

This project was generated with create-empirica-app.

Getting started

Setup

To build this app a custom verison of the empirica-core is required. This is included as a submodule in this repository

git submodule update --init --recursive

Install the corresponding version of meteor.

curl "https://install.meteor.com/?release=1.9.3" | sh

Install the meteor project.

meteor npm install

Insert environments

Enter folder

cd python

Create environment, enter environment, install packages.

python3.9 -m venv .venv
. .venv/bin/activate
python % pip install -r requirements.txt

Insert environments into database

python upload_environments.py ../data/select_environments.json

Run locally

To run this project locally:

NODE_TLS_REJECT_UNAUTHORIZED=0 meteor --settings local.json

Introduction

See also: https://docs.google.com/presentation/d/1WeNTlvzua9MRpHEa1imxPuRFjh-6oa3Aq1bhDpM9vlQ/edit?usp=sharing

The objective of this application is to study cultural evolution, expecially when machines and humans co-develop. We study the play of different games. Within the games different actions do have different rewards. The objective of the player is to maximise his total reward. Each game can have different settings, which in general are randomly generated. A specific setting is called environment.

To mimic cultural evolution, we are running series of game rounds, where in each round a player see the solution of a previous player. She is then invited to give her own solution and potential improve on it. A Series of game rounds, in which information passed on, is called a chain. Within a chain the games is played with the same environment.

Each chain has the same number of game rounds (lengthOfChain). At an specific position in the chain a machine can be inserted (positionOfMachineSolution), if the chain is supposed to have a machine solution (chainsHaveMachineSolution). In this case the environment, solution of the previous player, as well as, the machineSolutionModelName is send to a seperate endpoint, which will return its (machine) solution.

Similarly for the first round within a chain, the same endpoint is requested for a starting solution to be presented as the previous solution (startingSolutionModelName).

Mapping on empirica concepts

• empirica batch is a collection of games. Upon creation a defined number of chains is generated (see below). These chains are fetched during gameplay on demand.
• empirica game Each player has his own empirica game. For this reason, the number of games should be as large as the maximum number of player expected.
• empirica round The game has numberOfRounds rounds, plus 2 practice rounds. On each each round, a new chain is selected for the player to play. It is ensured, that never two player can play the same chain simultaniously and that never the same player is playing the same environment twice
• empirica stage Each game has three stages. In the planing stage the player is time to plan his moves (planningStageDurationInSeconds). In the response stage she has time to enter a solution (responseStageDurationInSeconds). Finally there is a 5 second phase to review the reward recieved.

Empirica Factors

The factors described previously and a few more can be controlled by empirica. There are two type of factors. Global factors are the same for all chains. Then there are chain factors. These factors can be different for different chains.

If the batch has only one treatment, then for each environment numberOfChainsPerEnvironment chains are created.

If the same batch has multiple treatments, then for each environment and each treatment numberOfChainsPerEnvironment chains are created. E.g. if there are three treatments, 80 environments and numberOfChainsPerEnvironment: 2, then 3 x 80 x 2 chains are created.

Note: Each player can only play once each environment.

Global

• playerCount (int) - should be 1
• numberOfRounds (int) - maximum number of rounds each player is playing
• experimentName (string) - used for selecting the right environments
• numberOfChainsPerEnvironment (int) - number of chains for each environment and treatment to be created
• lengthOfChain (int) - number of player (human or machine) sequentially playing in a chain (not including the starting solution)
• debug (bool) - shorter intro for debugging
• planningStageDurationInSeconds (int)
• responseStageDurationInSeconds (int)

Per Chain

• positionOfMachineSolution (int) - position in chain in which the machine solution is entered
• startingSolutionModelName (string) - model name used for the starting solution
• machineSolutionModelName (string) - model name used for the machine solution
• chainsHaveMachineSolution (boolean) - flag to indicate if the chain should have a machine solution

Architecture

Three main components:

• Empirica
  • Empirica Core
  • Empirica Reward Networks (this repository)
• Mongo Database
• Machine Backend

Empirica

Empirica consist of the empirica core (https://github.com/empiricaly/meteor-empirica-core). We use a slightly modified version (https://github.com/LBrinkmann/meteor-empirica-core). Most of the custom code of empirica is in the empirica folder within this repro. Empirica contains both, the frontend and the backend of the empirica app.

Mongo Database

All data from empirica, as well as our custom extension are stored in a mongo database. All environments for an experiment have to be uploaded to the database before they can be used.

Connection

Setup

Copy machine-culture-2.pem into ~/.ssh.

Set permissions for the .ssh key.

chmod 600 ~/.ssh/machine-culture-2.pem

Add to ~/.ssh/config the following lines:

Host mpi-ec2-emp1
  ForwardAgent yes
  Hostname 3.127.208.75
  IdentityFile /absolute/path/to/machine-culture-2.pem
  User ubuntu

The /absolute/path/to/machine-culture-2.pem needs to be addopted.

Creating Tunnel

To be able to connect to the database a ssh tunnel needs to be created. It is as simple as:

ssh -L 3002:0.0.0.0:27017 mpi-ec2-emp1

Using Mongo Compass

With the tunnel active the mongo database can be reached with:

mongodb://0.0.0.0:3002/?readPreference=primary&appname=MongoDB%20Compass%20Community&ssl=false

Machine Backend

The machine backend is a seperate endpoint (a different server). Whenever the empirica backend needs a machine solution a request is send to the machine backend. This request contains the environment, the previous solution and the modelName.

To test the machine backend: https://reqbin.com/pd5d7svg

To check the machine configuration: https://reqbin.com/aweszjm6

To update the machine configuration: https://reqbin.com/pthff5f5

Protokoll for running experiments

Step 1: Define Factors in Logbook

Step 2: Check if all factors are avaible in Empirica

Step 3: Create Treatments

Step 4: Calculate number of participants

Without machine:

rounds = (number of environments) x (number of chains for each environment) x (number for length) + With machine: rounds = (number of environments) x (number of chains for each environment) x (number for length)

number length without machine = 3 number length with machine = 2

Minimum number of participants: (total number of rounds) / (number Of Rounds per participant)

Step 5: Create Batch

Game Count: 2x (Minimum number of participants)

Step 6: Check if machine backend is correctly configured

Step 7: Send a test request to machine backend

Step 8: Start the experiment