The Art of Game Matchmaking ft. AI: The Usage of AI in Online PvP Matchmaking
Valorant, one of the most popular PvP (Player Vs Player) online multiplayer games, uses sophisticated techniques for its matchmaking (teaming up and pitting players against each other, the reverse of the matrimonial process, to be honest). With a new Act in play in the game, it's essential we discuss the crucial features of this game and how AI improves it. Riot Games Inc., the developer of this game, is a famous company with other top-of-the-line games like League of Legends. They have developed the process of matchmaking so well that they claim it's close to perfect at the moment, and chances are it's just going to improve. Let's find out what makes it so, or at least scratch the surface of it.
 |
| VALORANT, on playvalorant.com |
FIRST THINGS FIRST:
The biggest misconception is that the matchmaking system aims to give the player equal losses and equal wins or to make the player addicted to the game, or to create a true to its word kind of fair match. What the matchmaker seeks to do really is get the players to enjoy the game, getting the most player retention. The company claims to make the gaming community welcoming and healthy, keeping everyone playing happy.
HOW DO THEY DO IT?
How do they do it? Valorant uses AI and machine learning heavily for this. It uses advanced neural networks (essentially the learning part of the machine learning system. The network of neurons learns everything about the player) to make the system learn everything about the player patterns, how they play against certain players and how well they team up. How they use their guns, how many bullets they waste, the number of kills they have, the times they die, and the many times they leave their team hanging only to just go to the kitchen to grab something to eat. They analyse everything they can to ensure you wind up with the best team for you and against an appropriate team.
WHAT THE SYSTEM USES TO DECIDE MATCHES:
Getting down to the critical factor that plays a significant role in matchmaking, it is the skill level that makes the most difference. Now, for obvious reasons, Riot would not disclose its proprietary methods. But assuming some features already in use in games like CS: GO or Apex Legends doesn't hurt, especially when these techniques are proven and efficient. One of these techniques is the ELO rating.
- Elo rating is a number assigned to each player based on how the player plays the game. The Elo rating is a score assigned to each player, and it's up to the developer to use it in their own way to make the matchmaking in their desired way. The score is used for creating a fair game with a win/loss outcome and a reduced chance of a draw.
- Now Elo rating is a simple method to perform skill analysis of a player because it has few factors, such as match outcome. For better matchmaking, Microsoft Game Studios developed the TrueSkill algorithm in 2006 for more advanced player analysis by including factors such as player uncertainty and play variability, making the matchmaking even more robust than before. They even made the algorithm open-source! Companies that make sophisticated tech open-source (free to use for everyone) deserve a lot of respect. The algorithms predict the outcome of team-based matches and can also dynamically update player ratings over time. It can adapt to new data, account for player progression or regression, and adjust ratings based on recent performance.
- Bayesian Networks are used as well, but not only for matchmaking. Bayesian networks or simply known as belief networks, are neural networks that calculate or result in predictions using a probabilistic approach. They consider factors such as win/loss history, individual performance metrics, and team composition. These networks can help predict match outcomes and adjust player ratings accordingly, giving a match the player would enjoy.
- The last is reinforcement learning. So, to understand what reinforcement learning is, we need to see what the other types of learning are in machine learning:
- Supervised learning: Supervised learning is basically telling the computer what everything is and then asking it questions based on it. For example, you may show a kid apples and cars and tell them they are apples and cars. The kid now knows what an apple and a car are and can distinguish between them if asked.
- Unsupervised learning: This is quite the opposite of supervised learning. In this type of learning, we show the computer objects, and it's up to the computer to discover patterns and features, and it will give them its own names. Taking the same example from above, suppose we don't tell the kid the names of the objects. The kids will come up with their own names for the objects in their own language, but they will still be able to distinguish between them. So instead of car, they might name them as Ikran or something else, totally up to them. But for computers, it's usually in the form of integers.
- Reinforcement learning: This type of learning focuses on improving on previous performance. The model uses a reward system, aiming to try and maximise the rewards. So, it learns from its own mistakes.
More about reinforcement learning
Now that we understand reinforced learning, can you relate to how developers might use this technique to get a good match? The neural network will learn from its mistakes and try to get a better game. Now defining a good match is up to Riot Games, but whatever their definition is, the neural network model will try and maximise the rewards, which here is the rating of the match on a scale.
AVOIDING DRAWS:
That irritation of a draw extending multiple times gets on your nerves. First off, getting a draw is irritating; extending the draw is even more traumatising. Valorant uses its state-of-the-art neural network to pit your team against another team such that the chances of getting a draw are smaller—a brilliant method for player retention, isn't it? There is a skill gap between the opposing teams, but it's the right amount to ensure that either team can win without resulting in a draw.
THE TAKE-AWAYS:
So, to summarise, we have discussed a total of 5 methods which popular games use to create a good match. Using great neural network models, Elo ratings, TrueSkill, reinforced learning and Belief/Bayesian networks.
Next time you play a match, try to make sense of what you've just read and see how advanced these systems are!
Also, if you play Valorant, be sure to comment your favourite agents! Mine are Phoenix and Kay/O :-D
.png)
A well written article, this happens to be.
ReplyDeleteFavorite characters of mine are Reyna & Brimstone & Omen.
Thank you, dear reader. :)
Delete