Boards›Gameplay

Is it hard to make an AI that can look at properties in a game for ranked?

Subject 5··2 votes·565 views

For example, looking at a game and giving a player a result based on what they've done / what they've accomplished.

If it isn't, then why not make a system that doesn't punish people for having a 0/25 bottom lane:

5LP For a win -5LP For a loss

3LP For creep score (make the AI ignore this factor when playing a support, read below for the replacement value)

REPLACEMENT HERE: 3LP For pinging strategic locations (An AI can trace whether or not something is being pinged for whatever or not.)

3LP Based on kill participation (this includes assists)

3LP Based on Objective Participation.

5LP Based on Contributed Deaths (AI should be able to tell whether or not something is cancelled. For example, trading 1 for 1 isn't bad - but trading 0 / 1 is awful while trading 1 / 2 is very good.)

1 Additional LP based on honor

Remove the placement tests and instead make a static number like in Rainbow Six Siege / Overwatch.

This will enable people to enter elos that they're suitable with REGARDLESS if they're winning or not, as winning is a lot different than outperforming those around you. Let's say:

You're a good player and you:

Have 100 CS at 10 minutes: +3 You lost the match: -5 Have 90% kill participation +3 You got ALL of the objectives yourself: +3 Contributed deaths: +5 People honored you because bottom lane fed 25 kills and top fed 7 kills to the enemy Riven while the jungle Eve keeps thinking that split pushing is a good idea: +1

Regardless if you lost, you just gained 10 LP because you as a player have performed very well; regardless of how shitty your team is playing.

Now . . . you're a bad player and you:

Have 40 CS at 10 minutes: +1 You lost the match: -5 Have 0% kill participation, because you're playing pacifist tryndamere: 0 You got NO objectives yourself or with your team: 0 Contributed deaths: -5 People didn't honor you because you're playing AP Zac and ended up feeding 15 deaths: 0

You just lost 9 LP, you're not very good at the game so you should be losing LP to find a spot or elo that is more fit for you.

9 Comments

KirbyCake3 votes

If your measurement of skill is how often you can win games, then measuring that property is already an unbiased estimator, which the elo system already is.

Your suggestions would be as absurd as rating chess players based on how often they are material up in a game. Sure people with more material are more often winning but there are certain playstyles that sacrifice material that still win games.

Have you not looked at the alpha zero games? What they are saying is that humans should not decide the weights on variables for evaluating something. The only variable that matters is winning/losing games which is the most accurate. In this case, the simplest is the most accurate, considering alpha zero only looks at .1% of the total positions compared to stockfish who has humans putting weights on its evaluations for trying to determine what wins games like material count/pawn positions/squares controlled.

There is no point making an AI system for estimating skill, it has already been mathematically proven that the perfect AI would implement the exact same system.

Subject 51 votes

{quoted}

If your measurement of skill is how often you can win games, then measuring that property is already an unbiased estimator, which the elo system already is.

Your suggestions would be as absurd as rating chess players based on how often they are material up in a game. Sure people with more material are more often winning but there are certain playstyles that sacrifice material that still win games.

Have you not looked at the alpha zero games? What they are saying is that humans should not decide the weights on variables for evaluating something. The only variable that matters is winning/losing games which is the most accurate. In this case, the simplest is the most accurate, considering alpha zero only looks at .1% of the total positions compared to stockfish who has humans putting weights on its evaluations for trying to determine what wins games like material count/pawn positions/squares controlled.

There is no point making an AI system for estimating skill, it has already been mathematically proven that the perfect AI would implement the exact same system.

Holy this comment is ear-rip-inducing, read my post again. I pretty much specifically specified with specifications about how winning elo shouldn't be specifically gained by specifically winning, because in specific winning doesn't mean much when you get carried and losing doesn't mean much when you have a 0/25 bottom lane.

You know, just to be specific.

KirbyCake1 votes

so basically your definition of skill has nothing to do with winning or losing, then thats fine with you, I already mentioned that's as absurd as rating chess players based on how much material they are up during the game and ignoring the end result.

Subject 51 votes

Forgot to mention, the AI should be able to detect if the support has these items: item 3301 item 3303 item 3097

The AI should also be able to detect if somebody bought these items, and are clearly not supporting: item 1055 item 1054 item 2033 item 1056

FooDang1 votes

Nope, bad idea. This would make players play the system rather than play the game to win.

Rand0mH3r0X1 votes

An artificial intelligence is required where?

Do you even know how to program? I mean like, bare minimum, remake Minecraft programming.

All you did was propose a simple counting system, and cite any logic operators as AI.

Subject 51 votes

{quoted}

An artificial intelligence is required where?

Do you even know how to program? I mean like, bare minimum, remake Minecraft programming.

All you did was propose a simple counting system, and cite any logic operators as AI.

https://puu.sh/yFhul/ad94069f7a.png

The post is requesting information.

ZephyrDrake1 votes

there are so many variables in the game that you cannot simply say "X amount of CS by Y time means you win/lose Z LP". There are so many different variables just in the amount of CS someone has like

  1. were you counterpicked?
  2. were you camped by the enemy jungler?
  3. is your champion good at CSing?
  4. did you roam a lot?
  5. did you get zoned due to a bad trade early?
  6. is your champion a late game or early game champion?
  7. what were your masteries?
  8. what were your opponents masteries?
  9. did your jungle gank a lot?
  10. did you jungler took CS every single time it came to lane?
  11. did your X lane came to yours after winning theirs and you had to share CS with them?

did you look at how many i just named from the top of my head? look i haven't even touched builds, or if you were in a specific lane, or summoner spells from both you and your opponent or either of the junglers or the masteries of the junglers or who even are the junglers or people who just afk farm the entire time!

And this is just from the CS perspective. You also have to take into account how much does certain amount of CS even amounts when compared to kills and objectives. There are so many dumb little variables that needs to be taken into account for every single little thing if you want a system that actually does what you want it to do and actually prove that is actually superior to one that is just "you either win or lose". The only thing you need to ever be worrying about it winning and not losing. I shouldn't have to worry about my CS being low or my Kill Participation being low because you know what? different champions have very different playstyles. Not every champion wants to group into a 5v5 some just want to splitpush you to death and do not want to be anywhere near a 5v5 fight.

The more things you add the more people will want to "game the system" just so they can get the most out of it instead of worrying about just finishing the game and winning.

More from Gameplay

3 Refunds per season, upvote if you want 3 refunds per season.
Gameplay764 comments
7,167
votes
3 Refunds per season, upvote if you want 3 refunds per season.
Gameplay156 comments
3,510
votes
UPVOTE FOR URF
Gameplay606 comments
2,440
votes
Yes plz.
Gameplay456 comments
1,721
votes
3 Refunds per season, up vote if you want 3 refunds per season.
Gameplay114 comments
1,716
votes
When is ANYONE at riot going to address the awful new player experience?
Gameplay634 comments
1,383
votes

More by Subject 5

The meta is not fun
Gameplay116 comments
101
votes
We are running out of unique ways to play champions
Gameplay188 comments
99
votes
Did you know that some people actually believe that AP items are superior to AD items?
Gameplay185 comments
82
votes
I would rather these ADC's be strong than these ADC's (It actually works!)
Gameplay97 comments
49
votes
Took like a month break from the game and it still feels like a piece of shit
Gameplay22 comments
45
votes
Why bother with feedback
Gameplay65 comments
40
votes