Advances in Measuring Inflation Within Virtual Economies Using Deep Reinforcement Learning

Conor Stephens, Chris Exton

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper assesses an improved framework for evaluating the performance of economies within online multiplayer games. Games in this medium traditionally require a great deal of testing and analysis to assess the outcomes and affects of player interactions. This process is normally imperfect and time consuming, leading a lot of games developers and publishers to maintain a lot of risk during the development and launch of multi-million dollar projects. The framework and example project presented within this paper uses deep reinforcement learning to simulate economic interactions between learning agents. This is possible by having agents learn from player demonstrations to interact with game systems such as Battling, Collecting Resources and Crafting Weapons which allows them to co-ordinate and cooperate to achieve long-term goals in the game. This paper shows recent breakthroughs in relation to training these agents to the variety of metrics learning agents can generate to help games designers test and polish multiplayer experience, which have proved hard to quantify and measure without extensive play testing. This paper is an extension paper to our publication in ICAART 2021 [21], within this paper includes the following additions: Further discussion on the economic properties of Eve OnlineAn evaluation of how player demonstrations can accelerate the training time for both agent’s policiesA measured example of how this tool can balance multiplayer game economies through parameter tuning.

Original languageEnglish
Title of host publicationAgents and Artificial Intelligence - 13th International Conference, ICAART 2021, Revised Selected Papers
EditorsAna Paula Rocha, Luc Steels, Jaap van den Herik
PublisherSpringer Science and Business Media Deutschland GmbH
Pages294-314
Number of pages21
ISBN (Print)9783031101601
DOIs
Publication statusPublished - 2022
Event13th International Conference on Agents and Artificial Intelligence, ICAART 2021 - Virtual, Online
Duration: 4 Feb 20216 Feb 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13251 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference13th International Conference on Agents and Artificial Intelligence, ICAART 2021
CityVirtual, Online
Period4/02/216/02/21

Keywords

  • Artificial intelligence
  • Game economies
  • Games design
  • Reinforcement learning

Fingerprint

Dive into the research topics of 'Advances in Measuring Inflation Within Virtual Economies Using Deep Reinforcement Learning'. Together they form a unique fingerprint.

Cite this