Questionable AI decisions

[Dyzalot]

Originally Posted by tcblcommish

Once again you are stuck on "MLB managers do do that" and this becomes a statement everytime, you are convinced of a certain thing and nothing that is said can change your mind. I don't agree with a "6" coming in to replace a "13" when the hitter is a better hitter as well so that is why I ask for more information. Not that hard to follow right? Troubleshooting to get the answer?

However, like I said, you are convinced that this is wrong and should never answer so the only ones that can help you are Markus and whomever else actually worked on the game programming because they are the only ones that have the "know" on the programming.

Good luck with the game, I tried to help and even though I am the only one, I got sarcasm and ignorance. If you don't like the way the game produces the results, then don't play it. If the game is truly doing things that you see that should not happen, file a bug report with evidence so that it can be fixed and you can enjoy it.

I really do not get your argument. This is a sim right? So it should sim as closely as possible real life MLB yes? Even if modern MLB managers don't make the most optimal plays all the time, like sac bunting, it should still sim that right? Or am I mistaken as to the purpose of OOTP? You keep acting like if the AI is producing the most optimal play then it is fine but I disagree if a MLB manager wouldn't make that same "optimal" play in the normal course of a season. Because then it isn't a simulation anymore, it is just a video game.

So of course I'm going to be stuck on "MLB managers don't do that" until someone proves me wrong. I have been watching and following MLB baseball for forty years now though so I am pretty confident in my ability to know how the typical MLB manager does most things, especially when it comes to substitutions.

[tcblcommish]

I'm also going to point out that I have played thousands of games out as the manager and have not seen the screwy things that you are talking about which is why I ask questions like who is on the bench and game situations to get a better understanding of why things might be happening.

[Dyzalot]

This one is for Markus. How many times did Andrelton Simmons get pinch hit for last year in the 7th or later with his team leading? Or Alcides Escobar?

[tcblcommish]

NM, just not worth it

[Toast]

Originally Posted by tcblcommish

I'm also going to point out that I have played thousands of games out as the manager and have not seen the screwy things that you are talking about which is why I ask questions like who is on the bench and game situations to get a better understanding of why things might be happening.

I've seen what the OP is talking about but I think it comes down to how we set up the line-ups for the AI to manager. If we program many options - meaning possible changes to be implement - more times than not mistakes start to show up. Few changes programmed and the AI seems more consistent playing out the game in a realistic manner. There is ceiling to the AI.

Toast

[Dyzalot]

Originally Posted by Toast

I've seen what the OP is talking about but I think it comes down to how we set up the line-ups for the AI to manager. If we program many options - meaning possible changes to be implement - more times than not mistakes start to show up. Few changes programmed and the AI seems more consistent playing out the game in a realistic manner. There is ceiling to the AI.

Toast

I almost never personalize anything in the strategy stuff though. Rarely I set someone to "never be pinch hit for" because I think pinch hitting for them is unrealistic even if optimal sometimes. I want the manager to manage according to his strategy so that changing managers actually makes a difference.

[skoormit]

Originally Posted by Dyzalot

How many times did Andrelton Simmons get pinch hit for last year in the 7th or later with his team leading? Or Alcides Escobar?

Zero. And zero. But maybe their actual managers in MLB are less prone to pitch-hit late than is your manager. And maybe your roster is different than the Braves and Royals rosters last year.

What are your manager's tendencies? Is he a guy who likes to pinch hit an awful lot? Maybe that actual MLB manager actually would pinch hit for Andrelton or Alcides in these situations, if he had your roster.

What is the difference in batting ratings between your SS and the player that pinch-hit for him? What is the difference between the fielding ratings of your SS and the player who took his fielding spot?

The only way I could justify pinch-hitting for a great SS when I have a late lead is if a lot of the following are true:
a) my bullpen is weak and/or tired (and my SP if he is still in the game)
b) the opponent's bullpen is dominant
c) my PH is a great hitter
d) my replacement SS is decent with the glove
e) Andrelton is fatigued (or nearly so)
f) the pitchers I expect to use are very flyball prone
g) the park is very flyball prone
h) the opponent's lineup has a lot of LHB pull hitters with good power
i) I have 2 or 3 runners on base
j) I have 0 or 1 outs

[Dyzalot]

Originally Posted by skoormit

What are your manager's tendencies? Is he a guy who likes to pinch hit an awful lot? Maybe that actual MLB manager actually would pinch hit for Andrelton or Alcides in these situations, if he had your roster.

Not sure. Pinch hitting tendencies aren't one of the things listed on his managing page. He does like to play "small ball" though.

[MKG1734]

I'm going to re-post this (because I feel it was lost in the shuffle of the back and forth between dyzalot and tblcommish), to make a point which I will elaborate on below:

Here is a similar issue I posted about last fall:

AI In-Game

--------------------------------------------------------------------------------

I have a long-running dynasty running (currently in year 2032 after starting in 2014).

I will only watch a game played (as GM) when I consider it an important game (usually an elimination-type game).

Recently, in a game #7 scenario I watched my team's manager make a move which was questionable at best....but seemed like the AI should easily NOT make this decision.

The game went into extra innings, and after several substitutions, my backup middle infielders were in the game.

I have a backup utility INF who is good at every infield position - but best at 2B, and a backup second baseman who is ONLY able to play 2B.

Sadly, my AI manager put my Utility INF at this best position of 2B but left my backup 2B at a totally unfamiliar posiiton of SS. This led to disaster as a ground ball hit to the SS ended the game on a 2 out error.

Why would the AI do this? Is there a measure to help this not happen that I am not aware of?

Thank you to anyone who has any help / comments to offer.

It is beyond clear that the AI should have put the backup utility INF at SS (since he could play 2B, SS, 3B equally well), and the backup 2B at 2B (since he could only play 2B).

The replies were (verbatim):

#1, The Wolf: Sounds like the AI programming needs more work
#2, JM Durron: I wish I could help more, but my solution to the in-game AI is to make all of the decisions myself during every game. That way, I only have myself to blame.
#3, SteveP: No. The game engine should optimize fielding any time there is a change, just as does when the AI sets a lineup/depth chart for a team. Markus contends that the engine does this, but it is certainly hard to see that. The game engine is designed so that any player can play any position (even catcher). I think this is a matter of expediency. That may help to explain what we see, since the probability of a player who is badly assigned making an error is significantly less that would occur in real life (or at least I think so -- it's difficult to make a statistically valid analysis). My solution is to interrupt the game and fix the fielding assignments as best I can, not only for my team but also for the other team. Not something that works for you considering how to play the game.

So, while dyzalot's deal with offensive substitutions at an inappropriate time that then affect defensive substitituions/positionings....mine deals with defensive positioning of players who have been substituted for.

And, yes, I can certainly stop the game and make the changes, myself, for my team (and the other team if I am in commission mode) ... but that is a "work around" which isn't suitable...bc I can't play-out every game between every team for the entire season, and the AI's teams are suffering from these types of decisions in games that they play against each other (as well as my own team suffering in its games that I'm not watching).

Above all, the issue I personally noted above seems like it should never, ever happen with the "checks" that the sim engine is said to run to ensure that this doesn't happen.

How many games are lost by an AI team playing an AI team because of this? How many games do I lose that I sim through? How many extra games can I win by using the "work around" of fixing my team and not the AI's? How skewed are the records of AI teams that lose games to other AI teams due to this? Or AI teams that win games due to the other AI team making this same mistake? That, honestly, borderline can be seen as game-breaking.

The solution is not a work-around. The solution is enhancing the sim engine / AI.

To Markus Heinsohn (in reply to your comment):

Please understand I am not trying to flame the thread or be obnoxious. My above post was posted with good and true intentions in order to display that, yes, many times the suggestions are to "work-around" this or that... but... that isn't what the answer should be.

Because, you have to keep in mind...that a work around helps you and hurts the AI. Or, if you use the work around on both teams in a game, it helps both.

However; it does NOT help the AI when the AI is playing itself during the sim. Last year, when I posted that issue, I was not aware of how to post my game logs, this, that...and, honestly, the answers I received about simply working around or, essentially, "I don't see this, ever" frustrated me and I stopped playing. I'm not crazy...I know what I watched happen. So I can understand where dyzalot is coming from.

Now...I do have the desire to continue playing and will be purchasing v16 and playing it (as it provides a great utility at a very modest price). However; if I see these issues replicated, I promise I will attempt to figure out a way to go through the process of notifying those who need to be notified.

But...the answer can't always be (and I'm speaking for the community vets who do answer like this): "real life managers do stupid things, too." ... Because...c'mon...thats not the way to attack this issue.

Of course real life managers make dumb decisions. But they don't make decisions like the one I posted above at the MLB level (which is what this game is attempting to replicate). This shows (and no, unfortunately and as I said, I was not savvy enough at the time to take the time to learn how to save and upload the game file to prove my point) that there are fundamental inadequacies that do exist within the AI. I have learned my lesson for moving forward - to learn how to upload the file / game situation to prove my point - and I suggest to dyzalot to do the same (because the only response you will get otherwise is what you have received....doubt and excuses).

That being said...the way to attack this is not to doubt the poster out of hand. I know, for a fact, that I am not exaggerating what happened in my game or making it up. Next time, though, I will prove it and I hope the answer is not, "you can adjust your settings on your manager" - which is a 'work around' and does not deal with AI vs AI (which then does not give a realistic result for AI vs AI games, which then skews season win/loss records, playoff win/loss records, etc.) nor should it be considered to be "working as intended," because what happened is clearly faulty sim engine / AI behavior.

[RchW]

I checked the top 4 starting SS in my league in August of the current season so around 120 games played. The top 3 had not been pinch hit for once ie never. The fourth was PH for several times but further checking showed that the AI manager was bunt happy. I wasn't in agreement with all the PH decisions but it was not enough to bother me. My conclusion. This issue may exist in only certain types of leagues. For fictional leagues like I play it has never been an observable problem and further checking seems to confirm that.

FWIW the team with the bunt happy AI manager is almost a cinch for the WC and has 13 more wins than me.

[Dyzalot]

I almost always play the modern day quick start and then sim out 10-30 years until I get bored with the team and then start over with a new team so I tend to hold the sim to the standards of coming close to what I see actual managers and players do in today's game.

[Dr.K]

Originally Posted by skoormit

Zero. And zero. But maybe their actual managers in MLB are less prone to pitch-hit late than is your manager. And maybe your roster is different than the Braves and Royals rosters last year.

What are your manager's tendencies? Is he a guy who likes to pinch hit an awful lot? Maybe that actual MLB manager actually would pinch hit for Andrelton or Alcides in these situations, if he had your roster.

What is the difference in batting ratings between your SS and the player that pinch-hit for him? What is the difference between the fielding ratings of your SS and the player who took his fielding spot?

The only way I could justify pinch-hitting for a great SS when I have a late lead is if a lot of the following are true:
a) my bullpen is weak and/or tired (and my SP if he is still in the game)
b) the opponent's bullpen is dominant
c) my PH is a great hitter
d) my replacement SS is decent with the glove
e) Andrelton is fatigued (or nearly so)
f) the pitchers I expect to use are very flyball prone
g) the park is very flyball prone
h) the opponent's lineup has a lot of LHB pull hitters with good power
i) I have 2 or 3 runners on base
j) I have 0 or 1 outs

Good post this is the logical thought process I go through. Every manager is different. Got me thinking....it seems the AI has been judged according to an interpretation of "the book".

I've read earlier in the thread, "that would never happen", well what facts do you have to back that up? What "book" are you reading?Really, here are some of baseball's gems that did happen that were not supposed to happen......but DID.

1) You fail to replace your gimpy fielder with a better defensive replacement when closing out a game to win the World Series.


In game six of the 1986 World Series with the Red Sox leading and with 2 outs, Bill Buckner stays in at 1st base despite a better defensive player Dave Stapleton, being on the bench. It wouldn't happen? Buckner lets the ball through his legs......

See the Phillies Danny Ozark not replacing Greg Luzinski with a better defensive replacement in Game 4 of the 1977 NLCS, The Phillies led 5-3 going into the 9th, with two out Manny Mota hit a drive which bounced off Luzinski setting up a Dodger winning rally.

2) You let your flame-throwing closer pitch to the opponents powerful slugger with 2 out, first base empty and a World Series place on the line.


With 2 out in the 9th, in 1985 NLCS The Dodger's led the Cardinals 5-4. They were one out from winning and advancing to the World Series. Runners on 2nd and 3rd, first base empty and the Cardinals best player Jack Clark at the plate. - what do you do? Walk Clark and pitch to Van Slyke? The Dodger didn't and Clark hits it out. Game over......It's a toss up as to the right strategy.

3) It's the 9th Inning, you are protecting a 2-1 lead with 2 outs, tying run on 2nd and your pitcher has a no hitter going.......what do you do?


Game 4 of the 1946 World Series, Yankee pitcher Bill Bevens had a no hitter with 2 out in the 9th inning, the score was 2-1..........however tying run was on 2nd base.....what do you do? Yankee Manager orders an intentional walk thus putting the winning run on base. The next batter doubled giving the Dodgers a 3-2 victory, no hitter gone, Game Over.

Bill James once wrote "the % calculations which drive strategic decisions in baseball are normally so close and complicated that it is categorically impossible to state with any assurance what the correct course of action would be"

So to say "it would never happen" ............................I'm not so sure about that, there is plenty of evidence that anything can happen with management strategies and not everyone will manage by the "book".

Whitey Herzog once said, "if a guy is good enough to be on his roster, then he is good enough to play". So if your GG short stop is replaced by a player who the AI Manager deems is going to help him, then hey - it does happen. I've only highlighted some famous second guessed moves. And in case isn't 2nd guessing part of baseball?

It seems like some are crying that "that wouldn't happen"... let me remind them, there is no crying in baseball !

[MKG1734]

Dr. K ... while your post is noted ... it is still an excuse in place of creating an improved AI.

Read my above post.

These are the types of things that the AI should not ever do bc there is no even remotely logical reason to do so. It happened to cost the team I managed an important game....however; that is not even the point. The point is....how many games is this costing an AI team that plays another AI team or a human team that is not controlled by a human player during a sim? Do teams lose several games because of this....and thereby, obstruct the integrity of the sim? It is very, very possible. And if it does occur, then a counter-argument of "all teams have it happen, so it balances out" is both weak and does not promote product progress. This type of attitude will lead to a substandard sim engine whereby a competing product may one day move to take market share (see, in real life: OOTP defeating Baseball Mogul).

There is nothing wrong with pointing out deficiencies with the goal of having other people also look for them and then to note them with the goal of upgrading and improving the AI / sim engine. Constantly striving for improvement should be the goal and will make OOTP a more impressive product than it already is.

[Dr.K]

Originally Posted by MKG1734

Dr. K ... while your post is noted ... it is still an excuse in place of creating an improved AI.

Read my above post.

These are the types of things that the AI should not ever do bc there is no even remotely logical reason to do so. It happened to cost the team I managed an important game....however; that is not even the point. The point is....how many games is this costing an AI team that plays another AI team or a human team that is not controlled by a human player during a sim? Do teams lose several games because of this....and thereby, obstruct the integrity of the sim? It is very, very possible. And if it does occur, then a counter-argument of "all teams have it happen, so it balances out" is both weak and does not promote product progress. This type of attitude will lead to a weakened sim engine.

There is nothing wrong with pointing out deficiencies with the goal of having other people also look for them and then to note them with the goal of upgrading and improving the AI / sim engine. Constantly striving for improvement should be the goal and will make OOTP a more impressive product.

I am all for the AI improving, we are on OOTP 16, I started with OOTP 2, trust me, we've evolved! I'm not sure we had sound, or logos of facegen back in the day. The managerial moves have evolved too so I'm sure Markus and the boys will move forward.

I think feedback and constructive criticism are what drives the product forward, in fact it drives anything forward. I saw FM evolve from what in 1993 being more or less a cult game to becoming the beast that it is today and also OOTP grow, so we'll get there one day!

However in communication, what we say and how we say it heavily determine the outcome of the communication. Too often I get the flavour of "the game is broke", "it's broken" which as a loyal patient supporter of many years get's me a little irritable and less inclined to understand the point the OP in this case is making.

We know feedback is listened to, like I say, travel back to 1999 and see the progress. The feedback should be welcomed if not liked, from buyer as well as product maker in any industry, it is as I say a question of how it comes across and backed up with some facts, rather than say some sweeping statements like "it wouldn't happen"

Thank you for your input though, I'm sure it will go some way to the continued evolvement of the games AI.

[Harry_BallSack]

The point of AI is to think like humanity and be indistinguishable from same (our fears lie in the outthinking) - the only true way would be to feed in every historical MLB and Minor league managerial decision in since the dawn of ball, weighing each for all possible combination of ingame and seasonal (as well as known personality in Billy Martins case) factors. Now from a big data perspective is this doable? Well sure, go ping Andrew Ng at Stanford. But from a practical POV for this game at this time with these moans and groans, what Markus and team have put on the market is pretty damn incredible - the bellyaching over marginally arguable points only speaks to the passion that exists in the user base to eat up this product (which is devoid of real competition).

[Dyzalot]

I was never convinced that Stapleton was better defensively than Buckner. And I would charge that none of those are as bad as pinch hitting for your two best defensive players late in the game with a lead.

[Dyzalot]

Originally Posted by Harry_BallSack

The point of AI is to think like humanity and be indistinguishable from same (our fears lie in the outthinking) - the only true way would be to feed in every historical MLB and Minor league managerial decision in since the dawn of ball, weighing each for all possible combination of ingame and seasonal (as well as known personality in Billy Martins case) factors. Now from a big data perspective is this doable? Well sure, go ping Andrew Ng at Stanford. But from a practical POV for this game at this time with these moans and groans, what Markus and team have put on the market is pretty damn incredible - the bellyaching over marginally arguable points only speaks to the passion that exists in the user base to eat up this product (which is devoid of real competition).

No not this. The game has evolved over time. Maybe if you said all MLB managerial decisions for the last ten years I'd agree. And when the AI has a marginal decesion then he should default to what is standard in the current MLB. Again, I play this game to simulate modern MLB not to play a video game and min/max my optimal decisions.

[Harry_BallSack]

Originally Posted by Dyzalot

No not this. The game has evolved over time. Maybe if you said all MLB managerial decisions for the last ten years I'd agree. And when the AI has a marginal decesion then he should default to what is standard in the current MLB. Again, I play this game to simulate modern MLB not to play a video game and min/max my optimal decisions.

Simulate is not recreate really there is an element of algorithmic variability which creates playability if you want perfection re-read the box scores. The game has to be playable and replayable - in almost every conceivable historical context which is what it has achieved and continues to perfect on. Tweak by tweak it gets better, patch by patch release by release. It is not a simple math equation to plug into Excel and get a 3-2 score on May 17th 1967 when history told us it was 3-2. It might be rained out. That wasn't historical either, hell we can argue about weather patterns and El Nino in 1967 then, that's not the point - its a monument to baseball sim in much the same vein as any other wonder of the world.

[Dr.K]

Originally Posted by Dyzalot

I was never convinced that Stapleton was better defensively than Buckner. And I would charge that none of those are as bad as pinch hitting for your two best defensive players late in the game with a lead.

Dyzalot - I can see where you are coming from, I really can, but I don't have the context of your game to truly understand how you feel. I think your manager just went against the book. Or your book, or whatever the book is. We can argue about Stapleton all day long but the consensus at the time and over the years was that he at that point was the better defensive player - I can give you more examples if you like.

so I go with Bill James below,

"the % calculations which drive strategic decisions in baseball are normally so close and complicated that it is categorically impossible to state with any assurance what the correct course of action would be"

[Lou Gehrig]

Originally Posted by Harry_BallSack

hell we can argue about weather patterns and El Nino in 1967 then, that's not the point

I think we've already had a thread arguing about that.