Game Feature – The AI

First, like we said earlier this week on our Facebook page, the Artificial Intelligence (AI) is an essential part of the game. It has been a design and technology focus from the very beginning of this project, not only because it’s one of our specialties here at GolemLabs, but also because it was a priority for Nordic Games as well.

The AI we’ve developed for The Guild 3 is strongly based on some of our core beliefs and research in the field. Namely, how we alter and color the perceptions of the entities in the world, how we address solution-making, and how we keep everything dynamic in order to challenge the player more and more as time and the game progress. The Evolutive Human Emulator (EHE), the new AI system in TG3, is adaptable, relying on a 6 trait-system. With this said AI, behavioral patterns evolve because of its inherent ability to adapt to any given need and desire system.
First of all, you need to know that an AI entity is given to every group in the game. This means that every family has an AI, but other groups are defined as well, like cities, societies (e.g. the Freemasons or the Thieves’ Guild) and neighborhoods. These groups have different ranges of actions, defined by your character’s actions. These same groups have a series of values (6), divided in 3 axes, with a unique modifier at each end of the axe (so a single axe can have two different values).

The values are as follow:

– Construction: you like urban progress, and enjoy when things are ordered and linear.

– Destruction: you prefer natural habitats, organic development and chaotic patterns.

– Ego: you do things for you and the glory of your name only.

– Community: you belong with the group, and take pride in doing your part for the greater good.

– Power: you think that posterity can only be influenced by people who decide things, who rule, who have their face on a coin.

– Wealth: you think that money buys everything, even people of power, and you prefer to be the puppeteer.

As you can see, there are no “good” or “evil” values; it only depends on how you play them. Every action in the game, every little thing that you can do, build, or create, etc. has a set of values attached to it, an imprint of some sort. The score of each value results in a ”personality” which determines various aspects of the AI to create a more realistic gameplay. The player must understand that his decisions will have repercussions on the environment, economy and behavior of the NPC’s. The player can also analyze and discern the values of various surroundings and adapt his decisions based on that, to his benefit.

At the core of any adaptive Artificial Intelligence (AI), technology is the initial idea of learning. A system that doesn’t learn is pre-programmed. Sometimes, the ”correct” solution that programmers come up with is to integrate that pre-programmed system inside the game and the task of the system is to navigate through a series of conditions and determine which end of the pre-calculated decisions best fits its current condition. A ”fixed” system can have advantages like:

– The outcomes are “managed” and always under control.

– The programmers can debug and maintain the source code more easily.

– The design teams can help push any action in the desired direction to move the story along.

Because the people who are responsible for creating AI in games are programmers, and because programmers like to be able to predict what happens at any given moment, it’s no secret that they prefer systems they can maintain. But these previous advantages also have downsides: new situations, often introduced by human gamers, aren’t handled and decision patterns can be deducted by astute gamers. And once a player becomes more expert in a game, playing against the pre-programmed AI isn’t enough anymore and that’s when he looks for human opponents online instead.

But what if the system could reproduce the learning patterns of the human player? After all, playing a game is reproducing simple patterns in a more complex set of situations, time after time, and it’s something computers are made to do. To change all that and make the system learn how to play better, we need to look outside the field of computer software and go into psychology and biology fields. How come two different players can build two completely different styles of playing, while playing the exact same game?

To answer that question, we need to look at three different ways of learning. Take a brief look at them and try to see how computers and machines could use them to change your playing.

– The first kind of learning is through action: you lean forward a flower to smell it, a bee stings you and you associate flowers with bees for the time being. That’s a simple example of action/reaction. Looking at the consequence of the action, the effects are so negative and severe that the expected positive spur is exceeded.

– The second kind is built upon the first one: learning through observation. You see the flower, you see the bee and you deduct that there is a chance of injury for you. This way, you can predict the consequences of an action without even experiencing it. Basically, you know you could be stung by a bee because you saw someone experiencing it, or because someone got stung before. They learned through action, their effects were severe, and society learned from the mistakes of these people.

– The third kind (and the most interesting for gamers) is learning through planning. If putting you in front of a flower means possible injuries, then it’s possible to use that information on others. Like a nemesis, for instance. I don’t want to hurt myself, but I might want to hurt someone else. Again, I’ve never been hurt by a bee, but I am aware that it can happen and I am also aware of the possible side effects. So, I’m using that to project in time (a plan during an hypothetical fight).

These three types of learning can get more complex when you try to translate them for a computer software, but yet they represent simple and binary ways of thinking. Using all that information into uncomplicated elements allows computers to comprehend them and work with them.

We can conclude that all this creates a very different challenge for game designers and programmers because instead of scripting behaviors in a game, they need to teach the system about all the rules of the world. Then, the system needs to ”understand” how to use that knowledge in a player’s gameplay. Building, creating and training the gaming systems to go through all the different ways of learning are quite a challenge with the AI, but it is worth the effort. And in later articles, we will delve into more details on some finer points of it.