Data, AI, and Games

Sony has patented an AI that adjusts gameplay difficulty while you’re playing! This is a topic that greatly interests me, as I myself play a wide variety of games that all have varying difficulty levels. Some of these games are too easy, and I wish they were harder, while others need the difficulty turned down a notch. Thankfully, these games have preset difficulty levels, so it’s easy to switch back and forth, but these difficult levels are static, they don’t alter with your gameplay. With this AI that Sony is developing, the video game would analyze your play-style, gather data on how you’re struggling, gather data on what mistakes you are making, and adjust the game’s difficulty to better suit you so you can experience the game the way it’s meant to be played.

It suggests using AI to alter a boss‘ moves “based on successful and/or unsuccessful methods or tactics” previously used by a player:

What’s crazy about this is that the data is instantly moving from the players’ mistakes, through the game, into the AI, and then to this example boss. I can’t imagine what the machine learning code would look like in this case, but I will one day! On the other hand, if a player is doing well against a boss, the game can detect that as well. The boss will better predict the players actions, notice trends and patterns that the player is using to be successful, and adapt to that. Players’ who are really devoted to finishing a difficult game would appreciate such an AI that increases the challenge presented. It gives a game replay-ability and overall makes the game more enjoyable when it matches to a player’s level.

“The reward may include, for example, a particular trophy, a particular medal, a particular amount of video game currency, a particular amount of video game points, and/or a particular amount of currency to spend via an electronic or e-commerce store.”

Adding a rewards system like the one mentioned above would also give players incentive to play better! They would put more hours into the game and improve their play from the difficulty matching AI, gaining those sweet rewards for progressing on a higher difficulty. This is something already in place in most games, but if the AI can make the difficulty a continuous adjusting variable, this could bring new innovations/reward systems as well. Awesome.

This AI, if successful, can be used in so many games, and could even branch out into other fields as well. I myself play Chess nonstop, so if I could play with a computer engine that would always match my chess strength, it would really allow me to improve. It would respond appropriately to my moves, or play even a little better. It would act as a mirror to myself and can really analyze what mistakes I’m making with a data gathering AI. So AI in games and game theory would be benificial to the player and the game itself.

If this type of AI could recognize mistakes in sports and offer improvements to the player it would prove even more useful. This is what coaches are for, but coaches do to not see every little detail of every player. If an AI had a complete view of the player and obtain some sort of data from movements/timing it would be an amazing tool to use. Imagine an AI having a complete view of a basketball game, gathering data on every players’ movement, position, aggression, defense, etc., and compiling it all into suggestions for each player. I personally believe players would improve (if the AI was perfect) and this would indeed results in more interesting games. Hopefully we would be able to see something like this in the future. We already have data analytics in sports, but an AI to this degree wouldn't be easy to create.

This is what data analytics and data science is all about, using the data that is available everywhere and interpreting it in a useful way. Games have become such a core hobby since the start of the pandemic, and making them even more accessible to larger populations would be wonderful to see. Even puzzle games that can change difficulty would be amazing to see. A game’s difficulty settings is not something everyone pays attention to either, so if nothing changes on the surface, the player would be none-the-wiser and experience a game with an AI targeted specifically for them. Sony is heading in the right direction in regards to this AI project.

I recently learned how to use grid search and pipelines in Python scikitlearn. This has proven very useful for my data analysis and predictive modeling skills with the current projects I’m working on. With these two new additions to my learning, I’m able to cut my code in more than half, and achieve better results as well. In the next weeks to come I hope to beat these processes in my brain so I can use them to their full affect! These predictive modeling techniques mixed with machine learning/AI could prove to be more interesting than I thought.