In-Game Economies, Deep Neural Networks

Drawing on expertise in complex systems and emergent financial phenomena, my primary point of interest would be exploring how availability of in-game resources and trading behavior impacted macroscopic indicators in in-game economies. This work would be directly applicable to the development and regulation of in-game markets, with the overarching goal of creating fair and enjoyable economic circumstances for players across genres. Comprehensive exploration of behaviors would require agent based simulation at a large and parallel-executed scale, with empirical calibration from real data key to applying findings to a specific game. Frameworks and other useful tools should be developed to complement in-depth analysis following realistic-sized simulations, which could be shared to promote inter-disciplinary collaboration. This focus invites a number of peripheral research opportunities, with applicability of findings to popular commercial titles the core objective.

Many financial crimes with significant real-world repercussions can be employed in in-game economies with relatively limited consequence. This includes coordinated trading for market manipulation, and automatic trading/gathering to 'farm' currencies often with great cost to the developers. Work investigating in-game market mechanics could incorporate anti-botting investigations through the unconventional channel of real-time player trading analysis. This work would be functionally similar to anti-fraud measures in real-world financial systems, and would yield techniques applicable across huge economic systems such as those found in MMORPG titles. This work could incorporate an agent based approach using avaricious agents in a regular population, or real data if available. After some iterations the gamified system could be analyzed using traditional statistical approaches to identify suspect trading patterns, or more nature inspired methods including deep neural networks for pattern recognition. Varying neural network architectures would be explored and compared for this specific application domain. The key application of this research is improving player experience through identifying malicious parties. Botting and other undefined behaviors can drastically impact both the popularity and longevity of a multiplayer title, working to improve in-game economies would have widespread and valuable results.

Publications following the above planned research would contribute to the fields of game development, game analytics, and potentially financial systems security upon completion. Findings would be applicable to developers at all levels, across multiple genres, and would utilize leading techniques in financial analysis, simulation, and deep-learning to rigorously identify interesting phenomena and non-standard behaviour. Target genres would include those where players have the capacity to store wealth in any form and trade between, making findings especially meaningful given the current popularity of MMO, FPS, and BR style games where aesthetic items can hold significant real world value. This work would complement much of the design, AI, and cognitive research at IGGI and could inform design decisions incorporated in such projects.