Fullscreen
Loading...
 

Welcome to Tutorial for Nobel-mimetic !

Table of contents


What is this weird stuff?


This model is based on Chavalarias 1998, and Roger de Gardelle 2005, which proposed a computational and multi-agent system approach of popperian epistemology.

Theories and researchers:

In this game, a community of researchers has to study a natural phenomenon, which can be described by many theories. A theory is said efficient, or good-quality, to represent the phenomenon if it can succeed in many experimental tests upon the phenomenon, i.e. if it can make good prediction for a given test. At each turn of the game, the theories are to be tested by researchers, and can be moved from the unpublished set, to the published set if submitted to the community, and then to the refuted set if a counterexample is found.

Game and strategies:

In this simplified view of scientific discovery, researchers work together to deal with a given set of theories. One researcher can first try to publish theories: he takes one theory from the unpublished set and makes several experimental tests upon the theory. If he reached a number of successive successful tests for the theory he publishes it. The “exploration-time-before-publication" value can be seen as the response of the researcher to the question: “After how many successive successful tests will I be confident enough and take the risk to publish this theory?". If he finds a test for which the theory is false he notices the counterexample. Once there are several theories in the published set, one can try to test (in an “exploration-time-before-refutation") a published theory, to replicate or to refute the publisher’s results. In fact researchers have to choose between publication and refutation processes.
Publications and refutations let researcher win or loose what we will call points (which can represent symbolic or material advantages...), and the researchers, who are connected between them in a network, imitate the strategy (i.e.: the values for the 3 parameters: “exploration-time-before-publication", “exploration-time-before-refutation", “preference-for-publication") of the “best-score agent" in their neighbourhood.

Netlogo Interface description


We propose you to handle this model by running a few simulations through a netlogo interface (see online simulation), and comparing the results with your own expectations.

So let us have a look on the different parameters of this model.


I Payoffs

The R/P parameter is crucial to determine the behaviour of your community: it represents the value of the “cost of a refutation" (R) divided by the “gain obtained with a publication" (P).
  •  you may set this cursor, and then click on the setup button to see the value of R/P for the simulation

II Theories distribution

You can specify the set of theories proposed to describe the phenomenon:
  • The number of theories to be tested (max is 5000).
  • Proportion (in percent) and qualities (in per-thousand) of under-sets of theories:
Usually theta1 are the good theories, theta2 is the set of middle-class theories, and theta3 is the set of bad theories.

III Size of the community

 

Determine the size of the community: the number of researchers in the game

IV Network connectivity

Determine the way they are connected together: 4 types of network are possible: “just-me", “local", “small-world", “global".If the network is local you may specify the number of neighbours of one researcher.
 
If the network is a small-world, you may also specify the size of the neighbourhood and the density of long-ties (for a description of what the small-worlds are, see …).

V There are other parameters to specify

  • “actualisation", which represents the fact that gains in a recent period are more important than older. At each turn, the score of each agent is multiplied by this value (in per-thousand).
  • “Imitation Threshold", which specify the level of competition in the community: at the end of a process, an agent decides to imitate his best neighbour if his own score divided by the mean of scores is lower than this value.
  • “Exit rate" is the probability that each agent leave the game, being replaced by a new random agent.
 

Run a simulation:

Click on 'Setup' and then click on 'Go'