How do you fill a blog from the world of AI with exciting, interesting content? So many topics have been discussed at length by so many people in recent months. What new angles can we add? We asked ourselves these questions in the team today, and once again it turned out that spontaneous ideas are often the best. A new blog category, the “Splitblog”, is born. From now on, each month a team member may request a topic from the world of AI to be examined in this category.
We begin with our back-end developer Florian and the topic “AI as an opponent in video games”
A brief look at the video games of recent years shows that most improvements have largely concerned graphics. Ever more detail, ever bigger worlds, ever better resolution. In many games the graphics are now so refined that they are barely distinguishable from reality. But what about the behaviour of the various characters? When it comes to programming NPCs in particular, it seems reasonable to assume that artificial intelligence has long been in use. NPCs are the non-playable characters in a video game, such as passers-by, road users or opponents. In many games you can already interact and communicate with them, and their behaviour often appears unpredictable. But what looks from the outside like artificial intelligence is in reality the work of the game developers. Instead of artificial intelligence, video games often use what is known as “pathfinding”. That means the paths and actions of the NPCs were laid down while the game was being developed. The NPCs' options are limited in this case and may depend on various factors, such as the player's strength. The more possible options and factors were defined during development, the more realistic the NPCs' behaviour appears.
But why does game development not rely on AI?
It seems reasonable to assume that using artificial intelligence would improve the playing experience. NPC behaviour would be more situational, more individual and less predictable. Communication could be adapted to players far more closely as well.
What is often overlooked in this supposed improvement, though, is that a learning AI could quickly destroy any fun in the game. There is a high probability that it would be almost impossible for players to win against AI-controlled NPCs – not least because the NPCs could form teams.
And another factor would be hard to control: NPCs based on AI would be able to leave locations and simply “stop playing along”. Under those conditions, storytelling within a video game is not feasible.
Put simply, then, using artificial intelligence in video games would have a negative effect on the fun. Who enjoys playing a game they cannot win?
Even so, there are first games in which artificial intelligence is used. In these games the worlds and possibilities are still limited, but that will change. Development teams have already managed, for instance, to limit the superiority of AI-controlled NPCs. A good example is AlphaStar, an AI-based program trained to play StarCraft II. Here it has already been possible to throttle the AI's actions and abilities so that AlphaStar remains a fair opponent – at least for absolute professionals. It will be fascinating to see how AI can be integrated into video games in the years to come.
