Video game developers were interested in AI way before this technology hit the mainstream. Making computer-controlled opponents smart and neither easy nor too hard to play against is a never-ending puzzle that video game development attempts to solve.
Since the early days of AI development, researchers instantly recognized that classic games like chess are perfect for AI testing. It didn’t take long for AI to beat the world’s best chess players, solidifying its importance for video game development. Unlike in real life, video game conditions can be fully measured and observed. Most game events can be marked as ‘good’ or ‘bad’, making ML model training much more reliable and predictable.
The most widespread use of AI in video game development is to control non-player characters (NPCs). Conventionally, NPCs had nothing to do with AI. Designers used a pre-defined set of rules to order a character to perform actions based on a particular situation. This is known as the Finite State Machine (FSM) algorithm introduced in the 1990s. For example, if an NPC finds another defeated NPC, it will become more alert and actively look for a player. Quite obviously, the FSM method has one significant drawback — predictability.
To tackle this problem, developers came up with other algorithms like the Monte Carlo Search Tree that would randomly choose between several different actions instead of one. This is exactly what Deep Blue used to defeat Kasparov in a chess match in 1997. While this helped to make games more engaging and somewhat unpredictable, they still lacked this largely insubstantial ‘aliveness’ factor. Moreover, many of today’s games strive to be as realistic as possible, making pre-programmed NPCs look out of place.
The real AI
To make NPCs’ behavior life-like, it’s critical for them to react to the environment and evolve over time based on the players’ input. This is what the world’s biggest game studios are now trying to achieve.
Given the increasingly complex narratives that modern video games are currently pushing, the reliance on natural language processing (NLP) will most likely become a thing in the next generation of games. Instead of pre-written lines and pre-defined face expressions, NPCs will be able to adjust them on the fly based on players’ behavior.
For example, at this point, gamers are used to somewhat weird in-game moments when NPCs are heedlessly delivering story-important speeches while a player is pointing weapons towards them or lying on the floor. By augmenting characters with AI, they will be able to change their voice of tone and be visibly confused by the players’ actions, making the in-game experience more engaging. Current advancements in AI already allow for believable speech and animation synthesis. However, developers admit that a lot of work is needed to allow NPCs to have such freedom and depth of expression.
Designers are also looking to use AI solutions to adjust game systems according to the player’s unique playstyle. Especially when it comes to huge open-world games with an excessive number of things to do and places to explore, understanding what a particular player really likes can significantly enhance the in-game experience. For example, if a player spends more time in a specific place, the game can make the next quest happen in that area. Other games that have NPC co-op at their core will also be able to tweak NPC playstyle according to the player.
Many developers also enjoy a massive efficiency boost provided by AI’s ability to help with asset creation and animation blending. For example, conventionally, animators would have to program character driving movements for different vehicles separately. With AI in place, it takes animators to come up with driving movements once and allow algorithms to adapt animation depending on the vehicle.
Is it here yet?
Despite the aforementioned advancements of AI in video game development, its potential is not even remotely realized.
In many ways, AI remains too unpredictable for developers to use. Paradoxically, in the majority of commercial game development, the goal is to make NPCs act and look intelligent, rather than making them truly intelligent.
Letting AI learn and evolve on its own without any restrictions can produce both fascinating and game-breaking results. For example, the core technology behind procedural generation, one of the most promising uses of AI that is mostly used to create levels on the fly, hasn’t been changed much since the 1990s. Yes, it’s used at a much bigger scale and utilizes much more processing power, but the fundamental principles remain the same. Despite all the new ways developers can enhance player experience, the desired level of control over in-game systems hasn’t changed much.
Of course, there is the other side of video game development, where countless indie developers embrace AI unpredictability and hinge game design on the ever-learning nature of ML algorithms. Being removed from commercial pressures, these ambitious developers play an important role in discovering new applications of AI in game development.