Redesigning Matchmaking with Intelligent Artificial Players



An important reality in the vast universe of multiplayer games, is the necessity of good matchmaking. It is not really considered part of the game by most, but it’s a key aspect of the experience - as it has a disproportionately high contribution to how much fun a game can be.


Entertainment is what video game companies are after, and as far as matchmaking is concerned, we can get easily misled into thinking that accuracy is the way to go - but here you’ll find that there is a better formula for matchmaking than a straight line: the fun formula.


After investing so much into making a game great, challenging, and fun; video game companies work really hard - and invest even more - to let humans populate their leaderboards and hope desperately to achieve a critical mass that will take care of the constant availability of players needed to avoid the dreaded alternative: Bots. This is where Decisive AI’s Intelligent Artificial Players (IAPs) can really help, and typically, this initial ‘population’ is their main mission, among others. And yet not their only benefit by far...


Now, ask yourself: Once a critical mass of human players is achieved, have we solved matchmaking? With accurate rules in place, will matchmaking take care of itself if we have enough human players? Are we done?


No - is the answer: Once there is a sufficient mass of human players at any given time, any game quickly becomes a platform in which strong players alienate the not-so-strong. At the beginning, strong beats weak because there aren't enough players of each level. Then, once you have a critical mass of human players, the most one can aspire to is a natural 50-50 split between winning and losing. This might sound fair and mathematically Darwinian, but simply put, the resulting analytics of a mass of human players does not represent well the experience of the individual player.


If we were to focus on the individual experience, the ideal balance calls for a scratch-that-satisfies-the-itch ratio of about 70% wins (coming out 1st) to 30% losses (not 1st).


Why 70-30: because we all love to win but we don’t want it to be too easy. What can I say - human nature?


Your first thought should now be - that is impossible to achieve for all individuals, since such a distribution does not naturally occur when the level of expertise of the players increases and matchmaking takes care of keeping such levels progressing together. And you would be correct, and very eloquent too by the way, you should think of writing for us :)


Now enter Decisive AI’s Intelligent Artificial Players (IAPs) - allowing us to completely redefine the physics of matchmaking: Instead of averaging out players and putting them together as precisely as possible - benefitting only the lords of mathematics - we can now calibrate matchmaking to benefit the individual. The key for this is to understand that the average of proficiency of any given group of players, needs to be slightly lower than that of the individual, thus permitting the 70-30 ratio to surface naturally.


As an example, imagine a four-player game. Left to their own liner matchmaking devices, an all-human crowd would naturally group into very similar skill levels. At the beginning, if you have bots, the humans will quickly defeat them making the game easy/boring. If you don’t have bots, then the few humans will soon vary in proficiency so that the strong players will also have an easy/boring experience and the weak ones a hard/frustrating one. If, by a miracle (or heavy incentives), humans persist and further populate the game, the larger mass will allow for larger groups of homogeneous proficiency, thus resulting in an average 50-50 win/lose experience - which is plain boring for everyone.


Now imagine that same game, but populated with IAPs trained in such a way that there’s enough variety in their levels to match all possible human levels, and in large numbers and varied styles of play. In addition, the matchmaking is not designed to perfectly align players by equal level, but instead calibrated to match every single individual human player with three IAPs that average slightly lower in proficiency than the human: resulting in an average 70-30 win/lose experience - which is perfectly ideal for everyone.


How is this possible? Decisive AI’s IAPs are trained using Machine Learning, thus they truly learn to play, and progress in their proficiency as they gain more experience, and never get frustrated when they lose. Redesigning matchmaking using IAPs opens up a world of possibilities in which no human individual will be below average against others in any given episode, but will always be slightly above average. IAPs are the essential ingredient in a recalibrated matchmaking formula that can ensure a satisfying 70-30 ratio for all human players.