There’s no AI in Overwatch.
There’s a dev update from Jeff where he says that AI isn’t in the blizzard match making software but he thinks it’ll be a great addition when their systems are more modernized.
stats compared to your teammates don’t matter (the medals)
stats compared to other people playing the same hero at your SR do
2 Likes
That’s not quite what’s going on. The whole “50%” thing simply refers to the fact that as much as possible they try to form matches out of two teams with equal SR.
The only calculation beyond finding (as close as possible) equally rated players is treating larger groups as being stronger than smaller groups or individuals, in the cases where they cannot match group sizes on the two teams.
A players stats at the end of that match are used to determine PBSR awards, so in the sense that stats in each game affect the change in rank from that game, you can say a player’s stats matter… but they aren’t part of any direct calculation during matchmaking.
Oh, well, sure. That’s true, it’s just a different issue from representing those results in per-10s.
AI stands for artificial intelligence. Now is there anything in the game which seems you artificially do anything which would require an intelligence, like maybe matchmaking which is a process normally done by humans now done by the AI, or to like maybe calculating the value for pbsr which is now been handed over to the AI attempt.
It’s artificial intelligence.
1 Like
No. Those are examples of calculations and algorithms.
In computer science, “Artificial Intelligence” specifically refers to systems which are capable of “learning”.
You program an algorithm (write out all the steps), but you teach an artificial intelligence (by inputting large amounts of example data).
1 Like
Well I guess engineers and…. Everybody else 
must have two completely separate definitions of artificial intelligence.
2 Likes
That’s only one part. It’s more like ‘computer doing things we’d usually need a human to do’. Most of that includes learning (or self-refinement), although, I’m not sure that ‘learning’ is necessary. HolloH got it right.
Automation doesn’t fall under machine learning and artificial intelligence. Automating match making through data science/analytics is far from implementing a neural network.
And as enjoyable as semantic debates on the definition of software applications is, Jeff Kaplan explicitly states no machine learning processes had been implemented (nor were they planning to) for Overwatch 1.
Nice try though…. 
2 Likes
“competently mimic how human beings make decisions”
Uh oh. More semantics incoming.
In all seriousness though, I’d be very surprised if they didn’t use machine-learning to define the relationship between MMR and winrate, at very least initially. That would have been an incredibly bad choice. Plus, Jeffy’s quote could pertain to active machine-learning processes.
Why would you waste artificial intelligence on match making? It’s literally an automation process using a multi-input function.
Could you do match making with AI? Absolutely!! And it would probably be bomber!! Except then people would REALLY complain because after 5 or 6 matches, everyone in Overwatch would experience 50% win rate because the system would be significantly more competent at determining ranks correctly as it could
Then again 
It would be able to learn and observe play making/game changing decisions rather than giving you a couple extra SR here and there.
I wouldn’t…. Machine Learning is still in its infancy. It’s getting bigger faster (thank tech giants!! 
) people specializing in machine learning and artificial intelligence are generally not your average BS CSCI kids.
2 Likes
To establish a relationship between MMR and chance-to-win.
Yes. And changes in player skill will be quickly reflected in MMR. That’s the goal. This is the system working perfectly. SR isn’t a reward.
No? Machine-leaning is really old. Like 1950s-old. It’s just an advanced data-model-building method. Algorhythmic data-analysis comes under machine-learning.
Machine Learning as a concept is many decades old. In implementation and complexity it is still very young.
But legitimately everyone claims how bad the system is at detecting MMR.
So which is it? Is the system a machine learning marvel and incredibly proficient at correctly matching people?
Or is it broken and your friend is a super star player who deserves to be higher?
Because it clearly can’t be both.
Then why are any of us grinding competitive?
You wouldn’t need Machine Learning for this.
Here’s a match maker coded by a Forum regular that, to my knowledge, uses no machine learning whatsoever. And from reviewing all of his notes, it’s not horrible for a make shift program written by 1 dude.
2 Likes
And what would you define as learning? The process of learning in any AI is algorithms and recording results. The question is asked what is the player’s MMR, the AI uses the knowledge of its disposal to essentially learn what your MMR is through the study of potentially your actions, statistics or whatever it uses. When you look at the patents for the matchmaking that Jeff Kaplan has, even that patent says that the Matchmaker will learn your habits of what you like to play and will match you accordingly. It maintains a continued profile on you as a player and continually updates your profile based on the new statistics that has received to use again in matchmaking. Seems rather indicative of learning to me.
The definitions I found are for AI however describe it has more or less the same thing
Artificial intelligence, the ability of a computer or computer-controlled robot to perform tasks commonly associated with intelligent beings.
“At its simplest form, artificial intelligence is a field, which combines computer science and robust datasets, to enable problem-solving.” - IBM
”a branch of computer science dealing with the simulation of intelligent behavior in computers” - Webster Dictionary
“Artificial intelligence algorithms are designed to make decisions, often using real-time data.”
- Brookings.edu
This happens when someone is underperforming at their current SR.
It’s performance base SR which is a thing for ranks below diamond.
4 Likes
Thread will become necro’d in a few days with no definitive conclusion.
Conclusion: As defined and outlined by 8 other responses (myself not included) it is agreed on that your friend is likely not performing as well as you and your other friends believe him to be.
In your initial discussion you mentioned he is a Bronze player. He likely has much individual work to do in order to overcome his shortcomings and receive PBSR gains.
2 Likes
When I first got the game in 2018, I made the mistake of checking out comp before I knew how to play the game well. I placed into silver, then quickly dropped down to bronze (this was Season 13 Open Que). I then stopped playing comp until basically season 25. When I que’d with my friends, I would see them gain 30-100 SR per game, while I would gain 5 SR on the same matches. Then when we lost games, I would lose 30SR+. My stats were high compared to Overbuff averages. I tried making an alt account at this time and placed into plat, and climbed to diamond. Whether or not I played with friends, I would gain SR if I maintained a 55% win rate.
If you have been hardstuck in bronze for years, the game seems to have a function that detects that you are que’d with a “booster”, and that is the only possible explanation for your wins. This is clearly seen when smurfs group with randoms during unranked to gm. As they get closer to 1000 SR difference, the person they are grouped with starts getting sub 10 SR per win. This is what makes climbing an actual hard stuck account so impossible. Because even if you are carrying games, if you group up with anyone, your SR will be penalized. It’s hard to play well enough to solo que and carry games, plus your chances of running into throwers and leavers go up substantially. I mean that’s guaranteed not to happen if you are in 6-stack of reliable friends.
I had to only solo que season 25 to climb from bronze to gold, and the games were extremely difficult. It wasn’t until this season that I climbed from gold to plat, and I still have the issue, when I que with friends, it assumes I am being boosted, and I will lose SR unless we have a 70% win rate. I can only climb when I solo que on this account. However, this also seems to be an issue with grouping with people who are 1000 SR higher. It assumes those players won the game no matter what your stats are. I can group up with random people I just played with in solo que, since they have very similar MMR to me, and it doesn’t trigger the ‘boosted penalty’.
The only way around it is for him to figure out how to climb while solo queing or to make an alt account and start over. Don’t touch comp unless you are in prime physical and mental shape. In fact, you want to play your absolute best on that alt so you get a high mmr, because that will dramatically influence your first competitive match.
Or… the higher you climb, the harder it is for a boosted account to contribute adequately and so they get negative PBSR. Doubly difficult when queued with a smurf because the smurf will be doing lots of damage and kills, making it easy to coast and hard to pad stats.
The MMR for QuickPlay, Competitive, and Arcade are all separate (and I’m pretty sure that Arcade has multiple MMRs, i.e. your Mystery Heroes MMR is not the same as your Lucioball MMR … I don’t think we’ve had official word on this, but it seems likely).
Everyone starts Competitive at the same MMR, and only your actions during your initial placement matches affect your initial placement.
2 Likes
No, I’m not playing with smurfs. I am playing with players about my skill level or slightly less, proven by alt accounts, and 1v1 match ups while in que. And the fact that my PBSR is higher while on an alt account with the same exact performance. The game definitely penalizes accounts that were placed into bronze and left there for years.
They alluded that initial MMR from quickplay can impact initial MMR in comp on the old forum when the devs were less secretive about how it all works. Also the stats from quickplay might play a role. To play it safe, if you are going to invest into creating an alt account to get a higher rank, you should play your best in quickplay, just in case it does have a baring. The whole point of your alt account is that it’s for your best play, only, no matter what. It’s not just the $15 investment or whatever, it’s the hours into getting level 25.
Oh and it also penalizes players who play with anyone that have 600+ SR more than them. It assumes you were carried, no matter what the stats say.
The same skill level should yield much higher absolute numbers at a lower rank. There’s also a tendency in low rank for players to chip damage each other constantly with a combination of bad positioning and bad aim, enormously inflating ‘average’ damage and healing. It’s also possible and has been claimed that smurf performance distorts the PBSR’s understanding of average performance in very low ranks (Bronze/Silver).
It is true that mature accounts will tend to gain or lose smaller amounts of SR … there is a ‘confidence’ number based on how much you’ve played recently and how stable your ranking has been. That may also account for some differences that you see between accounts.
They never alluded to any such thing, they simply failed to specify either way. Consequently, players have verified experimentally that you start in the same place and can rank up (or down) equally easily regardless of QP performance in the level up phase. Heck, you could level up in Arcade and Custom Games and never have a QP MMR at all if you really wanted.
2 Likes