The e ects of irrationality on scienti c agents in the simulation of scienti c discovery

Keywords

Loading...
Thumbnail Image

Issue Date

2022-02-18

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

The Rich et al. paper of scienti c intractability [1] concludes that the solu- tion to scienti c inference is intractable given the current set of assumptions. This renders solutions to the problem of scienti c automation unsolvable in polynomial time. In an e ort to further explore the idea of scienti c automation derived from the notions of Rich et al. we attempt to use agent-based modelling to sim- ulate the e ects of irrational decision making on scienti c agents. In doing so we explore the e ects that irrationality has on agent's performance when solving scienti c problems. Agents representing scientists must ascertain the structures of a series of Finite State Transducers representing some cognitive system in reality that the scientists wish to understand. These scienti c agents build up their own FST, their own theory of how the cognitive system works, incrementally as to try and match the structure of the cognitive system. By assigning these agents some level of irrationality which makes agents's FST building increasingly random as it increases we test by means of various metrics how the agents's performance changes for various levels of irrationality. Ultimately, the ndings from this paper indicate that adding irrationality to the agents's behaviour provides no added bene t to the e ectiveness of the agents. An agent that chooses optimally every time will generally out- perform an agent that makes sub-optimal moves for any metric this paper tests for. The agents perform better or on chance level for low irrationality and on or below chance level as irrationality increases. In the end, these ndings remain in line with the Rich et al. paper of scienti c intractability when it comes to the di culty of automating scienti c solutions.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen