AI & Games: A meaningful relationship.

This week I’ll be looking at things from a different perspective. Instead of how other technologies can enhance game development I’ll be writing about how games can be a boon in others. There are many examples of this ranging from usage in film making to raising awareness for causes but as this is primarily an AI focused blog that’s what I’ll be writing about.

Traditional AI and games have almost always had a very close connection. Early AI development was most famously measured by playing chess against the equivalent human experts. But today games are playing an even more important role in the development of AI. They provide simplified versions of reality, usually with clear rules and reward systems meaning that not only can AI tackle these problems but can easily be told when it is doing well simply by making it pay attention to the game’s ‘score’. In the case of multiplayer games, they also come with huge amounts of gathered training data from the thousands or tens of thousands of players.

Until recently this was done as a purely theoretical exercise, so researchers could measure and study their creations similar to the chess tournaments of long ago. However, recent advances have enabled the AI to learn several games at the same time which is one step closer to ‘transfer learning’ (Economist, 2017) which is when an AI can use pre-existing knowledge of another task to help in solving a new one. This could be another game or a real-world application.

A prominent example is Google’s DeepMind being trained upon the game ‘StarCraft II’ (Blizzard Entertainment, 2010) where an AI has managed to defeat a player on a 1 versus 1 match amongst many other approaches of AI learning using this game. To play StarCraft successfully a player must not only know how to react to what they see in front of them but must also be aware of what might be happening off their screen. Due to the games ‘fog of war’ feature a player can only see a limited range around them. This means that a successful avenue for an AI would require it developing a rudimentary memory in order to work out where players are and what they are doing. The potential for such a memory on non-game applications cannot be overstated.

With algorithms being trained on games already having real world achievements (Evans, 2016) it seems that the relationship of AI and games will continue to be a long and fruitful one.


Vinyals, Oriol; Gaffney, Stephen; Ewalds, Timo (2017). DeepMind and Blizzard open StarCraft II as an AI research environment. Available from:

Evans, Richard; Gao, Jim (2016). DeepMind AI Reduces Google Data Centre Cooling Bill by 40%. Available from:

Economist, The. (2017). Why AI researchers like video games. Available from:

Starcraft II: Wings of Liberty (2010). (Computer game).

Microsoft Windows, OS X. Blizzard Entertainment

Recent Posts

See All

Self-Organising Maps & Clustering Algorithms

During last week’s blog post, we had a look at artificial neural networks. These were trained using data where a developer would know the input and correct output. Unlike them the algorithms we are lo