Games and AI: Paths, Challenges, Critique

This editorial introduction to the 2019 issue of Eludamos, co-authored by Mathias Fuchs and Andreas Sudmann, examines the intersecting histories and futures of games and artificial intelligence. Its central argument is that the relationship between the two fields runs in both directions: games serve as a testbed for AI, and AI increasingly shapes game design, testing, and marketing. The authors frame the issue as a critical inquiry into the sociocultural, phantasmatic, ideological, and economic dimensions of AI research, rather than a purely technical account of what the technology can accomplish.

The text traces key milestones, including the sophisticated NPC behavior in the 2005 shooter F.E.A.R., DeepMind's reinforcement learning systems that mastered classic Atari games, and AlphaGo's 2016 victory over Go champion Lee Sedol (rendered in the document as Kim Sung Yong). It highlights a pivotal moment in 2012, when a University of Toronto team won the ImageNet competition using convolutional neural networks trained on GPU hardware developed for gaming, illustrating how the gaming industry indirectly advanced AI. The authors also note that Turing's imitation test has not yet been passed by any machine.

On economic grounds, the document cites estimates placing worldwide AI revenues in 2018 between 8 and 9.5 billion US dollars, far below the roughly 135 billion dollars attributed to the games industry, while cautioning that the boundary between the two sectors has become increasingly blurred. It points to game engine companies such as Unity and hardware manufacturers such as NVIDIA expanding into automotive and other non-game markets, raising ethical questions about the transfer of simulation-trained systems to real-world contexts where failures carry physical consequences. The authors conclude that games remain epistemically important to AI development, though they caution that skills learned in simulated environments do not yet transfer reliably to real roads.