Way before modern LLMs came along, video gamers used to talk about game AI. These were the systems behind games that told the computer what to do. In some games like chess, which had strict rules and relatively narrow domains, AIs were extremely powerful. Gamers could get the most challenging experiences that they wanted from these systems, and they would rarely make mistakes.
However, with less constrained games, especially those in 3D environments or involving real-time strategy, AIs were not as effective. Computer systems couldn’t figure out how to deal with a lot of human tactics, and they had no ability to adapt.
What’s more, many games were unpredictable. For example, team-based games like Dota 2 were difficult for AIs to get their heads around because of how many different things could happen. There was always something new that could occur, and the AI didn’t have a system in place to deal with them.
On the extreme end, MMORPG games and other strategy games or role-playing games had issues with NPCs. These non-playable characters would often repeat the same lines over and over again, leading to memes, even in top games like Skyrim.
All this is now leading to changes in the industry with the development of generative AI and more advanced AI systems in general. It’s now possible to imagine a world where the intelligence of the computer in games is no longer as disappointing. For example, NPCs and storytelling is likely to become much better. Because of LLMs, it doesn’t seem like much of a stretch to say that in the near future, it will be possible to have unscripted conversations with characters in artificial environments.
Already there are demos that show how NPCs can adapt their voices and conversation to what players say to them. This is going to lead to a far more interactive version of games where players are able to really probe these systems and learn as much as possible from them so that they can take the next course of action. Conversations will also feel less scripted and more organic, with NPCs responding specifically to the types of questions being asked. This means that gamers will have the opportunity to improve the way they interact with game worlds and perhaps even brush up on real-life negotiation and investigative skills. They may also be able to co-evolve with NPCs in the game, forming alliances with them, massively improving play.
Procedural worlds
At the same time, AI is making it possible to develop procedural worlds that are personalizable. Previously, game developers had to develop environments by hand, building features piece by piece. This was time-consuming and often challenging, leading to massive increases in game development costs. That all changed with No Man’s Sky nearly ten years ago, where AI systems generated the assets based on fundamental underlying rules.
However, these generative techniques have improved vastly in recent years, and now it’s possible to build entire game worlds from a single snippet or picture. Again, this is going to give players a more personalized experience. Gamers will be able to level up in environments of their choosing, and AIs will be able to create situations which are completely unique for the player. It’s possible that each player will have missions and campaigns that nobody else has experienced before.
Opponents and multiplayer
It’s also likely that opponents and multiplayer systems in these games will improve dramatically. AI agents are good at generalising narrow tasks, but there are already examples of autonomous characters in games like PUBG which can learn goals and human tactics.
How much more these develop is anyone’s guess. We can reasonably assume that the current NPC dialogue, which is dominated by scripted trees, will likely improve in the next 5 to 10 years. The model will move to an unscripted, memory-aware version where conversations occur in native languages. These will be dramatically more dynamic than previous conversations and are already proven based on existing technologies.
World generation will also change a lot. Current AI is able to do basic procedural, static environments. However, in the future, these environments are likely to become fully adapted and unique to the player. This means that players will get genuine infinite replayability, and games will appear differently to them every time.
What about enemy AIs? Currently, even the most sophisticated systems are based on rule-based patterns. These are predictable for players, allowing them to dominate in many games. But in the future, over the next five to 10 years, enemy AIs will become autonomous learners. This means that they will be able to learn human strategies independently and react to them. It’s possible that AIs will understand how human players are winning and then adapt their gameplay and style to counteract them.
This is all a far cry from games like solitaire. But that doesn’t mean that the value of old-style AI is completely gone. Many players prefer an on-rails experience that feels familiar, while AI systems will be able to provide players with a real challenge and turn gaming into something much more interesting. There’s also a role for fixed games with specific rules that don’t always adapt. There are numerous examples of how this has been done successfully throughout the history of gaming, and that’s unlikely to change in the future.
Limits to how much better things can get
Finally, it’s worth considering whether there are limits to how much better things can get. One of the biggest challenges is latency. Much of AI processing happens at third-party data centers that are a long way from gamers, which can introduce lag to the equation.
Another problem is the energy cost. AI systems are expensive, and including them in games is likely to cost developers a lot of money over the long term, especially if they retain players. In this world, it’s likely we’ll see many more subscriptions for games. Players may have to play on an ongoing basis in order to enjoy their favorite titles. AI may become an engine, but it’s not guaranteed.
