It’s been a fixture of artificial intelligence in science fiction for a long time now for the AI in question to learn from its experiences with humans, and video game AI is no exception. A recently-filed Sony AI training model patent, however, is apparently now going whole hog on the concept. According to the abstract:
“…wherein the training causes the machine to imitate the interactive gameplay of the user.”
In the patent, the intent is apparently to help come up with a system that uses the data of players who had previously played the game, to teach the AI how to play more efficiently. This process makes use of a variety of methods across many different genres of games, ranging from turn-based strategy to first-person shooters. For instance, as explained in the patent:
“…In some implementations, the video and user input are recorded by the computing device and then uploaded to a cloud system, which receives the video and user input and applies them to train the machine learning model.”
Sony’s AI training model patent could help with many games in the future to train them to be more difficult for players, especially if building off of existing systems. For instance, many racing games often have a mode where players can race against the “ghosts” of other players who did timed runs and see if they can beat that time. The patent also says, however, that the training can be used to determine how a player responds to something:
“…other kinds of data including certain types of game state data can be used, for purposes of training the machine learning model, to define the user’s actions in response to what they experience during gameplay.”
If you’d like to know more about Sony’s AI training model patent, you can find the full document here. Otherwise, if Sony really does ever make use of this patent, we can only wait and see just how it will affect video game AI overall.