How much does the AI evaluate his team...

[BTerran]

Before pulling the trigger on a trade? I just unloaded Vernon Wells (OVR: 80) at 11.15mil a year on the Yankees for 10mil in cash. The Yankees put Wells on the bench. I recognize that the Yankees have money (though they actually didn't have the financial room to comfortable do that trade), but still.

In an ideal world an AI team would look at a trade the way (I think) a human team does: How does this change my team? Is the 3B I'm acquiring better than the 3B I have? If so, how much. Is that difference worth the player's salary+the prospects I'm giving up?

I'm sure the AI already does that to a certain extent, but I honestly don't see why the Yankees would talk to me about Vernon Wells at all.

[Keith L]

Agreed. This is one of the things that has been discussed in the past regarding AI decisions... the computer doesn't have great awareness of its entire organization or recognition that a trade like that doesn't really add much (or any) value to it.

[Keith L]

The inverse it true too... I've been able (in BM08) to trade for the teams best front-line starters offering prospects and the computer AI jumps on the trade even though their in-organization replacements for the position are poor and there are no viable FAs available on the market either.

[ohms_law]

Yea

[free2131]

There's an easy way around this.

Don't make trades with the AI that you KNOW wouldn't happen in real life.

[HoustonGM]

There's an easy way around this.

Don't make trades with the AI that you KNOW wouldn't happen in real life.

Yeah, but it doesn't stop two AI teams from making boneheaded trades with each other, and it also applies to free agent signings.

[BTerran]

Yeah, but it doesn't stop two AI teams from making boneheaded trades with each other, and it also applies to free agent signings.

Not only that, but I don't believe in the concept that we should need to put additional rules on ourselves in order to make the game challenging. Setting the game on the highest difficulty setting should be enough.

The AI has holes and that's understandable. It's unrealistic to think that the AI will ever be flawless. However, we should be striving to correct errors where we find them, not making excuses for ignoring them.

[ohms_law]

However, we should be striving to correct them, not making excuses for ignoring them.

I agree but I just wanted to say... I don't think anyone (particularly Clay) is ignoring anything, especially about the AI. The reality, when it comes to programming and AI, is that making changes slowly usually ends up being more efficient.

I'm not about to tell anyone to quit making complaints or anything... heck, I'm not about to myself... the only point that I'm making here is that we all just need to be patient with this stuff. Even if the changes don't seem obvious, there are changes made to the AI in just about every patch.

[BTerran]

I agree but I just wanted to say... I don't think anyone (particularly Clay) is ignoring anything, especially about the AI. The reality, when it comes to programming and AI, is that making changes slowly usually ends up being more efficient.

I'm not about to tell anyone to quit making complaints or anything... heck, I'm not about to myself... the only point that I'm making here is that we all just need to be patient with this stuff. Even if the changes don't seem obvious, there are changes made to the AI in just about every patch.

Definitely, I'm not accusing Clay of ignoring the problem. My comment was directed at the person who said that the solution is not to do those sorts of trades. Which, in my limited experience, has been a common suggestion. I do avoid trades that I find to be especially unrealistic, but I don't see that as anything more than a stopgap measure.

I did the Vernon Wells trade because I wanted to see if the Yankees had any intention of actually using him. If they were looking to use him as a starter, than I could see the justification in them pulling the trigger. Even though it'd still be a poor move for them.

[ohms_law]

Well, yea. Suggesting that people simply not make those trades or do whatever activity isn't a solution to anything; it's simply a workaround. Most of us know that instinctively I think, and I don't think we as a community should be trying to squelch that sort of activity. That's my only real point.

Additionally though, you (probably unintentionally) make a good point with your second paragraph. Part of the problem with these topics is not changing the AI or any other part of the game so that it's so inflexible that people can't do what they want to do, even if it's a bit ridiculous.

[Keith L]

Not only that, but I don't believe in the concept that we should need to put additional rules on ourselves in order to make the game challenging. Setting the game on the highest difficulty setting should be enough.

The AI has holes and that's understandable. However, we should be striving to correct them, not making excuses for ignoring them.

I agree 100%... no make that 120%. You'll find lots of people who disagree with that standpoint and I've personally been in VERY heated discussions (not as much on these forums, but for other similar sports sim games) where my "adversaries" INSIST that it's just not possible to make challenging AI for a game like this.

My "regular job" work involves creating a sort of "AI" to control real-world processes, so I know that premise is completely false. In my AI systems, the computers are as or more efficient & effective than human operators at doing WHAT WE ASK THEM TO DO. Just don't ASK them to do something they are not GOOD at doing.

Yes, it can be difficult to program solid AI... but it is far from impossible. Really the secret is to start with "less is more" attitude with respect to AI. I.e., restrict what you allow the AI to do to eliminate any "stupid" things it does. Then slowly add more and more AI functionality, ensuring each step of the way that any new algorthms don't cause old ones to "break". If done carefully, and to a master plan, eventually all the "little bits" of AI add up to one great big bit of very good, rock solid AI.

It's like's Ohms' sig, which I like a lot. If you code the computer PRECISELY with the right "rules" to follow... you'll end up with solid algorithms in the end.

That what shouldn't you ASK the AI to do? Don't try to code complex human-rationalization type "rules" because by definition, they ARE NOT going to be precise enough and you'll find unexpected negative results arising from all sorts of unanticipated scenarios. That's what we HUMANs excel at, not computers.

Instead, base AI decisions on the data available... let the computer number crunch and gain its competitiveness from "knowing" the exact "values" of it's own team worth... talent of players in the league... et cetera. Protect the computer from making stupid mistakes.

The one example I've given on these forums numerous times is when the computer trades away a front-line starter and has no suitable backup available in-house and also has no Free-agent alternatives. Let's face it... even if the "value" recevied in terms of prospects exceeds the value of that one player... this is NOT a good move for computer AI to undertake. Yes, you can argue that maybe the team is "rebuilding" and reducing team payroll and nebulous concepts like that.... but that's the problem... don't ask it to mimic a human GM. It in fact isn't going to have enough "awareness" to plan out and execute a strategy like that.

Those are HUMAN reasonings and THAT sort of AI programming is going to be difficult if not impossible to implement correctly.

So the simple restrictions of (1) don't trade away starters if it leaves a gap in the starting lineup and (2) don't acquire talent if there isn't already a gap in the starting lineup (your original post example), takes away a lot of the scenarios where the current AI makes "stupid" decisions.

Yes, I've gone on a rant... but this is my own personal saop-box favorite and it bears repeating this message once again. The road to better AI is really to allow it to do less and when it evaluates player "value" (which in a vacuum it seems to do pretty well). So no matter HOW good a particular player is based on his talent alone... the computer AI needs to recognize that that value diminishes to the computer's team even to the point of ZERO if that good player would be the 2nd, 3rd, 4th best player at that position within the teams organization.

OK, I feel better now. But in summary... yes I agree with you entirely. I play these games to be challenged and want to feel a sense of accomplishment when I'm actually able to beat the computer. I am not a historical simmer and I get no pleasure from playing "god mode" and just setting up things and watching them run whilst intervening to make up for any AI shortcomings.

Edit: I think to some extent Clay is progressing towards a philosophy similar to what I've described... simply because I SUSPECT the addition of depth charts might be to eventually help the AI figure out exactly what assets it really has and therefore whether any particular addition to the club really adds value or not... but that is just a theory. I have no insider knowledge whether that's the case or not.

[FRENCHREDSOX]

I agree 100%... no make that 120%. You'll find lots of people who disagree with that standpoint and I've personally been in VERY heated discussions (not as much on these forums, but for other similar sports sim games) where my "adversaries" INSIST that it's just not possible to make challenging AI for a game like this.

My "regular job" work involves creating a sort of "AI" to control real-world processes, so I know that premise is completely false. In my AI systems, the computers are as or more efficient & effective than human operators at doing WHAT WE ASK THEM TO DO. Just don't ASK them to do something they are not GOOD at doing.

Yes, it can be difficult to program solid AI... but it is far from impossible. Really the secret is to start with "less is more" attitude with respect to AI. I.e., restrict what you allow the AI to do to eliminate any "stupid" things it does. Then slowly add more and more AI functionality, ensuring each step of the way that any new algorthms don't cause old ones to "break". If done carefully, and to a master plan, eventually all the "little bits" of AI add up to one great big bit of very good, rock solid AI.

It's like's Ohms' sig, which I like a lot. If you code the computer PRECISELY with the right "rules" to follow... you'll end up with solid algorithms in the end.

That what shouldn't you ASK the AI to do? Don't try to code complex human-rationalization type "rules" because by definition, they ARE NOT going to be precise enough and you'll find unexpected negative results arising from all sorts of unanticipated scenarios. That's what we HUMANs excel at, not computers.

Instead, base AI decisions on the data available... let the computer number crunch and gain its competitiveness from "knowing" the exact "values" of it's own team worth... talent of players in the league... et cetera. Protect the computer from making stupid mistakes.

The one example I've given on these forums numerous times is when the computer trades away a front-line starter and has no suitable backup available in-house and also has no Free-agent alternatives. Let's face it... even if the "value" recevied in terms of prospects exceeds the value of that one player... this is NOT a good move for computer AI to undertake. Yes, you can argue that maybe the team is "rebuilding" and reducing team payroll and nebulous concepts like that.... but that's the problem... don't ask it to mimic a human GM. It in fact isn't going to have enough "awareness" to plan out and execute a strategy like that.

Those are HUMAN reasonings and THAT sort of AI programming is going to be difficult if not impossible to implement correctly.

So the simple restrictions of (1) don't trade away starters if it leaves a gap in the starting lineup and (2) don't acquire talent if there isn't already a gap in the starting lineup (your original post example), takes away a lot of the scenarios where the current AI makes "stupid" decisions.

Yes, I've gone on a rant... but this is my own personal saop-box favorite and it bears repeating this message once again. The road to better AI is really to allow it to do less and when it evaluates player "value" (which in a vacuum it seems to do pretty well). So no matter HOW good a particular player is based on his talent alone... the computer AI needs to recognize that that value diminishes to the computer's team even to the point of ZERO if that good player would be the 2nd, 3rd, 4th best player at that position within the teams organization.

OK, I feel better now. But in summary... yes I agree with you entirely. I play these games to be challenged and want to feel a sense of accomplishment when I'm actually able to beat the computer. I am not a historical simmer and I get no pleasure from playing "god mode" and just setting up things and watching them run whilst intervening to make up for any AI shortcomings.

Edit: I think to some extent Clay is progressing towards a philosophy similar to what I've described... simply because I SUSPECT the addition of depth charts might be to eventually help the AI figure out exactly what assets it really has and therefore whether any particular addition to the club really adds value or not... but that is just a theory. I have no insider knowledge whether that's the case or not.

Agreed entirely BUT Clay has "insider" knowledge & should know WHY AI does this & WHERE the problem stems from - Catknight in another thread showed that starting a game in 2008 the results using the default rosters & NO external influence are pretty close to realistic till Year 4 -- the real Question is WHY year 4 ? What has happened from Year 1 to Year 4 ??


In other threads (on BM08) it has been assessed that either:

1) AI FA issues are a culprit; or/&
2) AI Arbitraition ; &/or
3) the expenses system (as they stand) creates a financial inequilibrium that enhances the differences between the rich & the poor*

or a combination of the above.



Clay has released BM09 (which is BM08 with a few tweaks such as the depth chart,splitting eras - which was necessary for historical simmers & very useful) with the intention to sort these problems in future patches & most PROBABLY before the CD version is released.Already 11.05 is planned to solve some minor issues of the download version.


Basically, you need BM09 to get an AI enhanced game when patch 11.50 or so it out & that the downloaded version (of which we received) is a paid beta test but with the conviction of it being a superior game end of April/early May.






* Personally I believe it is this option which is the primary cause,Ohms & some others believe that it is more likely 1 & 2 combined.....

[BTerran]

I think the financial inequilibrium is a big part of the problem, but the arbitration system is what's amplifying it. My limited experience with BM2009 suggests that the arbitration system is a lot like it's been for a while. That is to say, 1st/2nd year arbitration players seem to be much too expensive. That's devastating for small market teams who depend on all 6 years they control a player before Free Agency to compete. As long as players are successfully demanding too much in arbitration, you will have teams like Tampa Bay find themselves in a never ending losing spiral by the start of year four.

Beyond that, I suspect that small market teams don't really know how to be small market teams. I think the AI needs separate coding depending on their financial situation. Maybe they already have that I don't know, but it's not a secret that Tampa Bay/Oakland have to employ a very different philosophy than New York/Boston. If the AI is essentially the same for every team, than the team with the most money will win almost all the time.

[FRENCHREDSOX]

I think the financial inequilibrium is a big part of the problem, but the arbitration system is what's amplifying it. My limited experience with BM2009 suggests that the arbitration system is a lot like it's been for a while. That is to say, 1st/2nd year arbitration players seem to be much too expensive. That's devastating for small market teams who depend on all 6 years they control a player before Free Agency to compete. As long as players are successfully demanding too much in arbitration, you will have teams like Tampa Bay find themselves in a never ending losing spiral by the start of year four.

Beyond that, I suspect that small market teams don't really know how to be small market teams. I think the AI needs separate coding depending on their financial situation. Maybe they already have that I don't know, but it's not a secret that Tampa Bay/Oakland have to employ a very different philosophy than New York/Boston. If the AI is essentially the same for every team, than the team with the most money will win almost all the time.

Agreed again (wow this imust be a record for me,lol ).Small markets in Baseball do use (like small businesses IRL) differing strategies of building & competing than their larger cousins.I suggested a simple method here which not only would allow the real influence of their spending to be felt BUT simultaneously CUT into the Large markets profits EVEN if they wanted to compete in this area - to me the simplicity plus the actual results would outweigh the "unrealistic" aspect of going away from the actual $-$ model.

It is clear that the FA model has to be tweaked in some respect,IRL Large markets do compete v one another for the Andruw Jones' & fill their holes through this market as their primary source of talent influx.The mid market teams also,but for Second tier players (upto 6-10 mill per annum) eg Carlos Silva & the small markets for periodical fill ins eg Gonzalez/Floyd this off season.

However,small markets (& this has been discussed ad infinitum I feel) use minor development as their primary source (coupled with trades).The model as it is now doesnt emphasize their investment nor the return on that investment - an example is the development of teams like Arizona or Colorado last year (& the way Tampa is coming along*) & over time these small market teams still pick high BUT (& it is a big one) their players' never hit peak as their farm spending tends to D+ thus turning a 72/95 into a 85....




* Tampa rated 1 by Baseball America as having the best minors system in 2008.

[Red Sox Fan]

I find the arbritration system to be better than it has been in previous game versions. My eligable players are asking for reasonable money not the ludicrous amounts as if they were free agents.

That being said, I agree that before pulling a trade for a computer's up and coming star player you have to think 'if the this would occur in real life'. Obviously a crazy trade wouldn't happen or would it? Where is the line in 'boderline'?

There are GMs in MLB that have pulled off wild trades. I thought Cabrerra and Willis was a steal in Detroits favour...however 3 months later Detroit is paying the price on Cabrerra ($163M for 8 years).

I never thought that San Diego would trade Alomar and Carter for MCGriff and Fernandeaz...and what about Babe Ruth?

Lets face it, we all exploit the computer in one way or another because its too hard not to. I think you have be real but remember this is a game.

I want competition & realism as much as everyone else. I'm just not sure how you get it from this game - I don't think deciding that 'the other GM would never make that deal' is a good way to play. Heck, 50% of my trades might never be made.