OOTP AI Improvements

[luckymann]

I've been pecking away at this for a while now and with us in mid-cycle felt it was time for a revisit.

The consistent pushback I seem to get whenever I voice my opinion that the dodgy AI logic and behaviour in evidence across most facets of how the game's engine runs a baseball franchise is - along with the whole NeL shemozzle and its primitive reports / stat analysis capabilities - one of the major elements holding OOTP back from true greatness is that getting it right "isn't as easy as you might think".

I fully accept that. However, there is a massive difference as far as I see it between "getting it right" and "making it better".

I don't ever expect it to be perfect, however I really do believe concerted effort should be made for it to be made much better for v26 and beyond.

An overhaul is way overdue.

At least start with a few seemingly straightforward improvements. Some suggestions:

• cut out the code telling the game that any player turning 30 needs to immediately be gotten rid of and at any price
• add a line of code that tells the game no franchise needs 15 1B or 28 RP
• add a line of code that tells the game every franchise needs more than 1 C, SS and CF

Finally, one more specific behavioural trait of OOTP that, in isolation, may not amount to much. However, the fact that it happens multiple times throughout the timeline of a save adds up and leaves AI-run franchises financially constricted to varying but always material degrees.

This is a full minors historical universe currently in 1910. Financials are on but still in the pre-minors era when only Active and Reserve rosters are in effect.

Mainly default settings across the board but AI player evals are at 40-40-15-5. Active Roster is set to 25 limit; Reserve Roster currently has no limit in place. Personalities and morale are off.

This is from the Pirates' transactions log but I could give you a dozen more just like it:




So then, devs, explain to me why a franchise would release any player whose salary they are already on the hook for under all circumstances other than if they trade him, when there is no limit on the number of players they are allowed to have on their books and no extraneous factors such as him being a bad egg and disrupting the group?

Why, in fact, are franchises releasing any players at all under these circumstances when they are either already paid for or being free-carried on minors deals?

See what I am saying?

While I agree this area of programming the game is highly challenging, I feel you guys have caused most of the more obvious problems by incorporating a variety of bad decision-making into the way your are programming the game and that, if you took a short amount of time to address these, it would at least get the needle moving in the right direction in this regard and hopefully allow a process of incremental improvements to take hold from version to version.

Every journey needs a first step, so hopefully this can be made a highly-rated priority for future versions.

Thanks for your time.

G

[Hrycaj]

I have struggled with the ai myself at times so I feel your pain. I have been told in the past that dead-ball era saves are hard to run because the ai is coded for the modern game. Now whether that is true or not the sign and release thing that many have seen is very frustrating regardless of the era.

A couple things I would like to see get looked at would be the shop a player feature. I have stopped using it all together because it is singly the easiest game exploit out there. Maybe it has been worked on but you could take your worst player in the organization and get large upgrades that would never happen. I did a post one time somewhere where I edited a guy down to 1 in actual and potential in every category and was able to flip him for talent. Kind of comical.

The other thing I have been thinking about lately is an open ended question for the community. Does anyone feel that every single ai uses the same algorithm to run their franchise? The game has taken the leap and introduced chemistry and personalities into the game but all the GM’s seem to operate in the same manner. Maybe I’m the minority here but I would be much more interested in a system where some GM’s are just bad and these horrible decisions eventually lead to them getting fired. Or that some organizations have truly different philosophies. (Ex: drafting college bats early or high ceiling power pitchers) I feel like every team operates like Tampa Bay. Maybe it’s just me. I love the game but these are a few things I would love to see addressed.

[luckymann]

Originally Posted by Hrycaj

I have struggled with the ai myself at times so I feel your pain. I have been told in the past that dead-ball era saves are hard to run because the ai is coded for the modern game. Now whether that is true or not the sign and release thing that many have seen is very frustrating regardless of the era.

A couple things I would like to see get looked at would be the shop a player feature. I have stopped using it all together because it is singly the easiest game exploit out there. Maybe it has been worked on but you could take your worst player in the organization and get large upgrades that would never happen. I did a post one time somewhere where I edited a guy down to 1 in actual and potential in every category and was able to flip him for talent. Kind of comical.

The other thing I have been thinking about lately is an open ended question for the community. Does anyone feel that every single ai uses the same algorithm to run their franchise? The game has taken the leap and introduced chemistry and personalities into the game but all the GM’s seem to operate in the same manner. Maybe I’m the minority here but I would be much more interested in a system where some GM’s are just bad and these horrible decisions eventually lead to them getting fired. Or that some organizations have truly different philosophies. (Ex: drafting college bats early or high ceiling power pitchers) I feel like every team operates like Tampa Bay. Maybe it’s just me. I love the game but these are a few things I would love to see addressed.

Yeah, to the devs' credit the HARD TRADING feature is a literal game-changer in my book.

The entire AI is run on a pre-defined script from the looks of it. Have a look at the logs each year and you'll see clear patterns with regard to extensions and the like.

But that is likely necessary just to keep it working at all and something for down the track a bit.

The thing about doing it incrementally is with those simple suggestions I have made the immediate effect will be a noticeable improvement without really having changed much and therefore hopefully limiting the flow-on effects.

I understand that this AI programming is a fine balance and that it follows the Newtonian action-reaction principle. However, I also feel that if some of the basic rules are improved then those to follow should be easier to implement.

That just seems logical.

Now, as to whether that's how programming works...

G

[dodgerblue88]

One thing that bothers me quite a bit is that organizations will accumulate a stockpile of above average players at a given position, yet rather than trade these players to address glaring holes on their roster, they continue to stockpile and shuttle these guys back and forth from AAA, wasting valuable (cost controlled!) service years and an opportunity to build a true contender.

[dodgerblue88]

Another thing I see is teams failing to address roster improvements where there are glaring weakness. Example: a team in my fictional league finished with 85 wins, squeaking into the wild card only to be swept in the first round. They were 13th of 15th in bullpen ERA, and the whole core of the team was returning for the following season. What would you guess they did with their $20M in money available that offseason? Well.... they didn't spend a single $ on the bullpen I'll tell you that.

[uruguru]

I just want to chime in and say that writing a competent AI is literally the hardest part of developing any game. It's not just hard, it's really hard.

And when you have a game as complex as OOTP, it's not just really hard, it's really really hard.

I open-sourced my game about two years ago with what I thought was a decent AI (that I wrote) and I was extremely fortunate enough for it to attract the attention of the most accomplished AI developer in the genre. The game's feature set is completely static (it's released), and he is literally still working on the AI two years later.

There is no benefit to the developer in making a great AI. All it does is add development cost, delay releases, and make players angry -- because the one thing players hate worse than an AI that makes mistakes is one that's good enough to consistently beat them.

[Hrycaj]

Originally Posted by uruguru

I just want to chime in and say that writing a competent AI is literally the hardest part of developing any game. It's not just hard, it's really hard.

And when you have a game as complex as OOTP, it's not just really hard, it's really really hard.

I open-sourced my game about two years ago with what I thought was a decent AI (that I wrote) and I was extremely fortunate enough for it to attract the attention of the most accomplished AI developer in the genre. The game's feature set is completely static (it's released), and he is literally still working on the AI two years later.

There is no benefit to the developer in making a great AI. All it does is add development cost, delay releases, and make players angry -- because the one thing players hate worse than an AI that makes mistakes is one that's good enough to consistently beat them.

Very interesting take. I definitely believe there is a fine line gray area you walk in games. They are entertainment after all and players do want to have a feeling of success.

[Brad K]

Originally Posted by Hrycaj

Very interesting take. I definitely believe there is a fine line gray area you walk in games. They are entertainment after all and players do want to have a feeling of success.

Back in DOS days a switch for help human and help computer was common.

[uruguru]

Originally Posted by Hrycaj

Very interesting take. I definitely believe there is a fine line gray area you walk in games. They are entertainment after all and players do want to have a feeling of success.

The primary focus of AI development is to ensure that the AI behavior that the player can see doesn't make egregious mistakes. Because if the player doesn't see any AI errors, then their immersion is not broken. The parts of the AI that the player can't see can just be faked away if necessary.

For example, in my game, the player can see the AI expanding, the AI in combat, and the AI in diplomacy. Mistakes in those areas were what players would see and would break immersion so those got the most attention. Other aspects of the AI got a more streamlined treatment.

But not only is OOTP an incredibly complex game, almost everything the AI does is visible to the player. You can see how they make trades, what players they draft and sign, how they build their lineups and pitching rotations, who they protect (or don't) for Rule V and expansion drafts, what players they put on waivers, who they promote from or keep in the minors.

And if you manage games, you can see how they do substitutions, when they attempt to steal or pitch out.

In other words, literally everything at the AI does in OOTP is visible in some way to the player. And the worst part is that this isn't some fantasy game where the OOTP developers can make up the rules. It's a simulation of a real sport where all of the gamers already have a deep, in-depth knowledge of how it is supposed be played.

In gaming, I honestly cannot think of a more difficult challenge for AI development than OOTP. I wouldn't do it.

[luckymann]

This is exactly why I'm suggesting just a few straightforward tweaks to begin with that will materially improve the roster management performance of the AI without much needing to be done.

Then see how things sit and go from there.

I understand it might be difficult but how it is at the moment is simply unsatisfactory. You know it, they know it and I know it.

[David Watts]

Originally Posted by luckymann

This is exactly why I'm suggesting just a few straightforward tweaks to begin with that will materially improve the roster management performance of the AI without much needing to be done.

Then see how things sit and go from there.

I understand it might be difficult but how it is at the moment is simply unsatisfactory. You know it, they know it and I know it.

Did you think when you posted your suggestions a lecture on programming would be coming your way? Hopefully, you won't lose any sleep tonight worrying about how selfish you were suggesting them

Your thing about teams not carrying 2 catchers, 2 guys that can play SS and the same for CF, is one of my biggest AI pet peeves.

[uruguru]

Sorry guys, I guess I am just very sympathetic to what I see as an impossible task for the devs.

I really didn't mean for it to come across as a lecture. Please accept my apologies!

[David Watts]

Originally Posted by uruguru

Sorry guys, I guess I am just very sympathetic to what I see as an impossible task for the devs.

I really didn't mean for it to come across as a lecture. Please accept my apologies!

Bad attempt at humor on my part.

[fredbeene]

Originally Posted by luckymann

Yeah, to the devs' credit the HARD TRADING feature is a literal game-changer in my book.

The entire AI is run on a pre-defined script from the looks of it. Have a look at the logs each year and you'll see clear patterns with regard to extensions and the like.

But that is likely necessary just to keep it working at all and something for down the track a bit.

The thing about doing it incrementally is with those simple suggestions I have made the immediate effect will be a noticeable improvement without really having changed much and therefore hopefully limiting the flow-on effects.

I understand that this AI programming is a fine balance and that it follows the Newtonian action-reaction principle. However, I also feel that if some of the basic rules are improved then those to follow should be easier to implement.

That just seems logical.

Now, as to whether that's how programming works...

G

well it is not AI. it is an alogrithm and for some reason no random abilities.

Agree that there is so much that can be done to make the game better. But the course is GRAPHICs. likely because then people will watch other people play (i don't get it , but i am not the target)

It would be nice if you could play head to head. People don't even ask for that anymore?

[luckymann]

Originally Posted by uruguru

Sorry guys, I guess I am just very sympathetic to what I see as an impossible task for the devs.

I really didn't mean for it to come across as a lecture. Please accept my apologies!

Not at all, I absolutely agree with what you were saying. Thus why I'm only asking for tweaks and not the paradigm shift that is necessary but almost impossible to implement.

But if (as my friend Lans likes to say) the perfect is allowed to be the enemy of the good then we might as well all just give up and go home, right?

[Pelican]

As to the Shop Player part, I found that moving the slider for Trade Difficulty up to (or close to) the Hard setting ended all of those generous offers for marginal players. I might get one worthwhile guy, but he probably has a bloated contract. Yet shopping a player can be an opening for further trade negotiations, adding guys on both sides.

I also play with the slider having the AI favor veterans, so that I can sometimes shop them and get prospects. I do that in recognition of the sad fact that I am invariably taking a bad team and trying to improve, over time, so that prospects make sense.

Hard Trading has been a dud for me. If hard means no trading, that would be my experience. Either no offers or only insulting ones. I don't get why the delay in response was implemented. It makes the back and forth interminable, which, among other things, is unrealistic. Pretty sure IRL GM's deal in real time, and not over several weeks. Why the leisurely pace? [Of course, this from the guy who also cannot stand delayed diagnosis of player injuries. No patience.]

[kq76]

Originally Posted by uruguru

Sorry guys, I guess I am just very sympathetic to what I see as an impossible task for the devs.

I really didn't mean for it to come across as a lecture. Please accept my apologies!

Don't worry about it. I for one was happy to read your post. And I'd like to read more from the perspective of developers. I especially liked your distinction between AI the user sees and the AI they don't. As much as I'm with the others in wanting to push the devs to make the game as good as it can be, I like also being able to read what maybe harder than other things to implement. That doesn't mean we all still shouldn't constantly strive to make the game better, just that it's good to get a reality check from time to time. After all, there's AI stuff the devs could probably work on without too much trouble and there's stuff that a good way to improve it probably just hasn't come to them yet. Let's work on the former and give the latter some time.

I don't really agree with luckymann when he says "an overhaul is way overdue" because I think any devs reading that would likely think, "yeah, no, an overhaul is too much for not enough benefit", but I do agree with him when he calls from incremental improvements like the ones he listed. You'll see different people post these "straightforward improvements" from time to time, and I hope the devs keep a list of the ones they see and come to mind and try to bang them out as they can. Oftentimes it's the little things that make a big difference, for good or bad.

[Pelican]

Originally Posted by Hrycaj

The other thing I have been thinking about lately is an open ended question for the community. Does anyone feel that every single ai uses the same algorithm to run their franchise? The game has taken the leap and introduced chemistry and personalities into the game but all the GM’s seem to operate in the same manner. Maybe I’m the minority here but I would be much more interested in a system where some GM’s are just bad and these horrible decisions eventually lead to them getting fired. Or that some organizations have truly different philosophies. (Ex: drafting college bats early or high ceiling power pitchers) I feel like every team operates like Tampa Bay. Maybe it’s just me. I love the game but these are a few things I would love to see addressed.

This is a good question. I have puzzled over it myself. I'm curious what others think.

The one difference I see in AI GM behavior is determined by finances. Teams with funds will sign free agents, extend star players, absorb huge contracts. Lowly franchises without dough will shed large contracts, lose good players, avoid pricey free agents, and even worry over bonus money for draftees. As IRL. I have ground-truthed this by changing the finances of modern "poor" teams like Tampa Bay or Pittsburgh. Those teams then can retain players and spend money (though not always wisely). So, you can indirectly influence GM behavior by fiddling with the finances. Or, you could impose a hard salary cap, that would impact wealthy teams like the Yanks and Dodgers and Braves.

A curiosity I have is the setting for a team's "philosophy", whether they are rebuilding or going for it all. If that is changed, does it influence the AI GM behavior? I have not tested this.

Other than the above, I have not seen any differentiation between/among AI GM's. They all behave the same, and cannot be directly customized, so far as I know. Which causes me to intervene, as Commissioner, in the interest of parity (and common sense), to take temporary control of woebegone teams to make changes. What I wish, and the post suggests, is that I could simply adjust those parameters for a given GM, and then let the AI take over and follow my lead.

For 2024, I experimented with taking over the Royals, Rays (rebranded as the Montreal Mounties), and Rockies, as well as the dreadful A's, using this approach. With mixed results. Note that the Rockies, who somehow draw well and have plenty of cash, were the easiest to handle, spending lavishly on pitching. The A's with cash gave a whole new meaning to "moneyball". It is hard to turn around a team in a year. (The Rays, already pretty good, were easier.) It would be fun to "set and forget" team GM's like that, and then play with my chosen team.

Pretty sure that the White Sox will be my reclamation project for 2025. LOL [meaning "lots of luck" as well as laughing].

Thoughts?

[Hrycaj]

Originally Posted by Pelican

This is a good question. I have puzzled over it myself. I'm curious what others think.

The one difference I see in AI GM behavior is determined by finances. Teams with funds will sign free agents, extend star players, absorb huge contracts. Lowly franchises without dough will shed large contracts, lose good players, avoid pricey free agents, and even worry over bonus money for draftees. As IRL. I have ground-truthed this by changing the finances of modern "poor" teams like Tampa Bay or Pittsburgh. Those teams then can retain players and spend money (though not always wisely). So, you can indirectly influence GM behavior by fiddling with the finances. Or, you could impose a hard salary cap, that would impact wealthy teams like the Yanks and Dodgers and Braves.

A curiosity I have is the setting for a team's "philosophy", whether they are rebuilding or going for it all. If that is changed, does it influence the AI GM behavior? I have not tested this.

Other than the above, I have not seen any differentiation between/among AI GM's. They all behave the same, and cannot be directly customized, so far as I know. Which causes me to intervene, as Commissioner, in the interest of parity (and common sense), to take temporary control of woebegone teams to make changes. What I wish, and the post suggests, is that I could simply adjust those parameters for a given GM, and then let the AI take over and follow my lead.

For 2024, I experimented with taking over the Royals, Rays (rebranded as the Montreal Mounties), and Rockies, as well as the dreadful A's, using this approach. With mixed results. Note that the Rockies, who somehow draw well and have plenty of cash, were the easiest to handle, spending lavishly on pitching. The A's with cash gave a whole new meaning to "moneyball". It is hard to turn around a team in a year. (The Rays, already pretty good, were easier.) It would be fun to "set and forget" team GM's like that, and then play with my chosen team.

Pretty sure that the White Sox will be my reclamation project for 2025. LOL [meaning "lots of luck" as well as laughing].

Thoughts?

I think your point about finances is a good one. You have gone further than I have in the testing. I have created leagues where I have gone in as commish and slashed the budget at the beginning of the off-season and the beginning or the pre-season, as those are the two times a year money gets added to the budget, and was very successful in destroying the roster talent of a big market high payroll clubs as a test. It was though, like you said something that has to be done manually. I haven’t looked how the rich teams acted in comparison. It is good to hear that they retain players and sign big FA. This is one of my concerns when I look at solo leagues. I feel like I always get whatever free agent I seek. I don’t even need to outbid teams, there is no competition for a guy that would be an upgrade for half of
the other clubs in the league and I’m getting them for fair market value.
My wonder then would be if OOTP accurately creates a real financial disparity like you see in the current landscape? Something to look at I guess.

I have had zero luck with setting hard caps. I would love to know your settings. Every time I try an institute a hard cap teams slowly go over it and within 5 seasons everyone is operating over budget and ignoring the setting. I heard it works, but I’m clearly doing something wrong.

[Pelican]

Yeah, one problem with a hard cap - with contemporary sims - would be the many long-term, bloated contracts. Suddenly springing a hard cap on the AI in 2024 is tough. Some teams will immediately exceed the cap. They may overreact, and shed contracts. Realistically, this is a feature that should be phased in, over time, so that teams can adapt. Probably what would work best is implementing a firm salary cap in conjunction with free agency starting, back in the 1970's. And then allowing that cap to slowly rise, year by year.