Exploring a More Honest 4X AI
Recent advancements in AI have provided the video gaming industry with a lot to think about, but one of the things I’m personally the most fascinated by is the ability for computer controlled players in games to be more thought driven. This concept of thought-driven computer controlled players is applicable across almost every type of game in the industry, but we’re going to focus on 4X style games today. 4X is shorthand for explore, expand, exploit, and exterminate, and this is the kind of game I can get lost in for hours on end, simply because the style of game forces you to think carefully when making your decisions. There are many well-known titles in this style of games, and they can fall under both turn-based and real-time strategy, but in both of those scenarios, it is not uncommon for a single player to take on an entire lobby of (or at least a few stand-in) computer controlled players.
Historically, the 4X style of games has given the player a handful of difficulty settings to choose from for these computer controlled players, and each of those difficulty settings enables a different decision tree for the computer to follow in its gameplay execution. The complexity and effectiveness of these decision trees can vary greatly, depending on the chosen difficulty setting. However, one thing that is incredibly common is for the more advanced difficulty settings to allow a certain threshold of cheating (or at least fudging) where the computer controlled player is concerned. This allows the game to present the player with a greater challenge, but it can also detract from the enjoyment and replayability of the game. Once you realize that the computer controlled player almost never changes its strategy (it simply executes it more swiftly and effectively), it becomes easy to predict what it will do.
I believe that the recent advancements in AI have paved the way for us to create a computer controlled player in 4X style games that is not as reliant upon these cheats. We can compile data the computer controlled player would have access to, including things like map resources, army positions, current alliances, and known factions. Then, using generative AI, we can consult that data and compile a real-time strategy that could lead to that computer controlled player’s victory without needing to allow any cheats, such as increased resource gains or shorter production cooldowns. Additionally, as the data set changes, the computer controlled player could even pivot to a more effective strategy, making it less predictable overall.
We can also adjust the difficulty of these new computer controlled players by determining how much information they are provided about the overall game mechanics at the outset of the game session. A more beginner friendly computer controlled player might only understand how to play their own faction, while a more advanced setting might enable the understanding of all factions and how to interact with them. Aside from the baseline difficulty settings, we could also collect data about how the player reacts to certain scenarios over time and factor that into the computer controlled player’s strategies. This allows us to create a more tailored experience for the player, providing them with an adequate challenge that doesn’t feel unnecessarily unfair.
Even with recent advancements in AI technology, it’s not the same as playing against another human, but it’s definitely a step in the right direction, especially for the solo gamer that enjoys a good 4X title. Who knows? Maybe computer controlled players in future titles might even teach you some new tricks.
