Machine Learning in Video Games

After Virtual Reality new hype in video games is: Machine Learning, which can be found in various areas of video game industry already although it’s just starting.

First of all let’s keep in mind that this what people think as “Artificial Intelligence in video games” since many years already actually was not true Artificial Intelligence or Machine Learning (which is subset of Artificial Intelligence) but hard-coded programming with hard-coded rules. On the other hand machine learning allows for dynamic adaptation to players.

Currently all the artificial intelligence buzzwords in video games that are hot are all machine learning related: machine learning (per se), neural networks and deep learning (subset of machine learning that uses neural networks).

Machine Learning in video industry can be found in various areas, including hardware, where companies like ARM, Intel, Nvidia and others are building chips (integrated circuits) that accelerate training of machine learning models in servers.

Another area are cloud services real aged to machine learning, offered by Google, Amazon (AWS) and Microsoft (Azure) where users can use pre-defined models or where users can train or upload their own models and then perform predictions or inferences.

Unity (Unity3D to be precise) is one of 2 most popular gaming engines nowadays and it recently introduced “Machine Learning Agents” where agents can be trained via imitation and training. Also Microsoft is releasing Windows ML soon as part of Windows 10.

Have you noticed sometimes when downloading/installing games to your smartphone that they can be very big – like several gigabytes big? That’s because game creators need to put textures of huge size. Machine Learning to the rescue: just imagine that instead of including texture (graphic pattern) for a game of size 150 megabytes, you could include machine learning model of size 15 megabytes! This is the trend: hundreds of megabytes if not gigabytes are used to train machine learning models and then all that gigabytes are thrown away (not included in games) and the game is using merely the models to generate beautiful textures.

Machine Learning is also used to optimize In-App-Purchase (IAP) pop-ups: for example: most mobile games nowadays are free with (IAP) and if specific gamer tends to spend just a little bit on IAP then pop-ups then machine learning offers to such gamer very low price IAP and users who spend more money are immediately offered bigger IAP – all adapted to specific person/gamers with use of Machine Learning. IAP is huge multi-billion business so even small optimizations bring huge money.

Machine Learning is also used by 3rd party libraries that are built-in into games and that discover what games user has installed in his smartphone and offer pop-ups with games that are similar, pop-ups that use Machine Learning base recommendation engine to offer such games that given gamer will be willing to install.

Machine Learning is used in gaming, similarly like in other areas, to analyze text and via text analytics to detect offensive game chats and forum posting.

Another application of machine learning is detection of cheating in games where server farms are analyzing movements of gamers to discover cheating patterns and eliminate such players.

Recently Google has released JavaScript for TensorFlow and one can see some games that use Machine Learning here: https://js.tensorflow.org : mostly these games use video camera of the device to control games via gestuers, to use objects in environment to control games or just to train games via camera.

Via unsupervised machine learning several people trained models that enabled to defeat video games much faster than humans can. Particularly old games from 8-bit era are very easy to train for machine learning.

Conclusion: machine learning in video games is everywhere and in gameplay it’s just starting – maybe soon we will see fantastic games that will impress us not just with prepared, pretty much “painted” graphics, but with generative based on machine learning textures landscapes and games that will adapt to gamer individually.

If I have missed above any application of Machine Learning in video games, please put it in comment below!

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s