Getting the AI to play as shrewdly as a human?

[bababui]

Originally Posted by The Wolf

I play every game out and play with just stats, so don't tell me it's impossible. I know better.

There's a lot of ways to do these things. For the draft, you can use your scouts and pick the chioces they recommend. As an alternative, you can turn stars (only! no ratings) on just for the draft.

For your starter team the simplest way is to let the AI handle the initial setup. (The rationale for this is that the team the AI sets up is the team that you inherited.) As an alternative, turn stars (only! no ratings) on just long enough to set your initial roster. Or you can let the AI play a season or two out and then take over, with some accumulated stats you can use to figure out who's who and what's what. All of these methods work.

Playing the game this way is harder, more fun, and more immersive than vanilla OOTP. No, I didn't invent it, I just read about it here on the board and tried it for myself. You won't have a dynasty, you will be scrabbling to figure out who to play and who to promote and demote, you will have epic successes and epic failures, and you will be immersed in the game - because now every single piece of data you can get matters because you don't have any ratings to go by.

Oh, and it makes trades nail-biting, agonizing fun, too.

Want to really feel like a GM? Play it this way.

Sounds like its worth a try.

[bababui]

Is it fair to look at defensive ratings like range, catcher ability and catcher arm?

[The Wolf]

Sure, just turn off Player Actual Ratings and Player Potential Ratings. Leave Other Ratings on.

[The Wolf]

Originally Posted by bababui

Sounds like its worth a try.

I'm never going back. It may not suit you, but if it does, you won't either.

[OmahaBaseball]

Originally Posted by The Wolf

I play every game out and play with just stats, so don't tell me it's impossible. I know better.

There's a lot of ways to do these things. For the draft, you can use your scouts and pick the chioces they recommend. As an alternative, you can turn stars (only! no ratings) on just for the draft.

For your starter team the simplest way is to let the AI handle the initial setup. (The rationale for this is that the team the AI sets up is the team that you inherited.) As an alternative, turn stars (only! no ratings) on just long enough to set your initial roster. Or you can let the AI play a season or two out and then take over, with some accumulated stats you can use to figure out who's who and what's what. All of these methods work.

Playing the game this way is harder, more fun, and more immersive than vanilla OOTP. No, I didn't invent it, I just read about it here on the board and tried it for myself. You won't have a dynasty, you will be scrabbling to figure out who to play and who to promote and demote, you will have epic successes and epic failures, and you will be immersed in the game - because now every single piece of data you can get matters because you don't have any ratings to go by.

Oh, and it makes trades nail-biting, agonizing fun, too.

Want to really feel like a GM? Play it this way.

Great post - thanks for sharing this. I look forward to giving this a try.

[Skywalker76]

Originally Posted by The Wolf

I'm never going back. It may not suit you, but if it does, you won't either.

Same with me here. Being successful is really something special now and sometimes I already enjoy being above .500 (if with an small market team).

I am also never going back again.

[injury log]

Getting back to the original post, considering beta for v10 is almost upon us, it seems it would be useful to list exactly why it is easy to outsmart the AI. The OP listed two issues:

-it's too easy to trade guys of zero value for guys of some value;
-the AI doesn't compete for minor league free agents.

There seem to be other problems as well:

-it's too easy to stack a trade, particularly at the winter meetings or at the trade deadline, since you can just keep requesting more and more players until the AI changes its mind.

-the AI should claim a *lot* more guys on waivers. I can pass pretty good guys through waivers almost at will;

-it's too easy to stockpile minor leaguers. They all take minor league contract extensions, which is crazy if they're good and aren't being given an opportunity in the bigs. So you can stockpile depth, and it doesn't seem that the AI does the same.

-the trade AI *still* seems to give 'negative value' to some pitching prospects. The AI might not much like a trade, then I ask it to add in some AA pitcher who my scout thinks is a 4-star prospect, and the AI suddenly accepts the deal. Doesn't make sense, and it almost seems there's a stray negative sign somewhere in the code of the trade AI;

-the AI doesn't seem to appreciate how the OOTP dev engine works. It will give crazy contracts to guys who are clearly declining, or trade crazy packages to guys any human player would recognize is going to decline fast. This leaves the AI hamstrung with massive contract commitments to mediocre players, and lets the human fleece the AI in trade;

-I still don't think the AI assigns anywhere near enough value to prospects. I've traded one big league pitcher, for four pitching prospects each of the same Potential as the big league guy, according to my scout. Two years down the line, I have three big league starters who are all above average from the deal, and the declining pitcher I traded away is a borderline fifth starter type;

-hidden players don't seem to work properly. When the AI finds them, they go to the FA pool, giving the human a lot of opportunities to sign young free agent prospects. Since the human can find hidden players himself, and also sign the ones the AI finds, the human has a huge advantage;

-also, with hidden players, it seems that spending an enormous amount, which the AI doesn't do, yields an enormous number of hidden finds, many of which are very good prospects. There should be severely diminishing returns for increased spending, or else this can be exploited;

-AI teams (still) don't seem to properly value closers in the draft, an especially important issue now that relievers are capable of being starters. A shutdown closer prospect should probably be valued more highly than a guy who projects as a below average starting right fielder, but the AI almost always seems to wait until the second round of the draft to even consider relievers. It's easy to stock a bullpen cheaply by drafting relievers, who seem to pan out far more often than position players or starting pitchers.

Intended the above just to start discussion- if there are other issues, it would be great to get a comprehensive list together so that the next version of OOTP can be even better than it is now.

[Isura]

Nice list. I'll add:

-AI doesn't consider contracts/FA and dumping salaries for trades. They don't realize that trading upcoming FA for some value is better than letting him run for the money when his existing team is short on cash.

[Cornelius McGillicuddy]

I like to play out games and the main problem I see is poor lineup choice. A recent game had Henry Aaron batting seventh for example. Something basic like that should never happen. Setting both lineups before every game is a pain.

[Afino]

Originally Posted by injury log

My assumption is that this is how AI teams sign hidden guys - the game decides that an AI team finds a hidden player, then puts the hidden player in the FA pool. I don't know why the game doesn't just assign the player to the AI team's roster.

Yeah - I just noticed that a young, talented (65 current, 73 potential) RF was hidden (I'm assuming), was signed by a team, and repeatedly DFA'd by the team. The team had literally 10-12 players on their DFA list, none with ANY stats. I released the players into the FA pool - without picking any up myself.

[Pdubya64]

Great stuff guys!

About the only thing I would add is this:
Quite a few of the teams at ML level have mediocre head scouts. Given the discussions we have had with Markus over OOTP9, I am pretty sure he confirmed that if they have a so-so or lousy head scout then they are seeing everything through his "mud-colored" glasses.
So I would not be surprised that some teams seem oblivious to the fact that they have a 4-5 star minor leaguer in their farm system, hence you can talk them out of said player.

If anything seems to need an improvement, it's the AI's lack of improving their head scout and coaches. Just my own observation, and certainly not heavily researched.
It would almost be worth it to run a test where each team is given a perfect head scout to see how they would value their players then. If I had to guess, Markus has done this long ago when developing the AI.

Just remember that everything is filtered through the head scout's ability- so keep your salt handy.

[bigMatt]

Originally Posted by Pdubya64

Great stuff guys!

About the only thing I would add is this:
Quite a few of the teams at ML level have mediocre head scouts. Given the discussions we have had with Markus over OOTP9, I am pretty sure he confirmed that if they have a so-so or lousy head scout then they are seeing everything through his "mud-colored" glasses.
So I would not be surprised that some teams seem oblivious to the fact that they have a 4-5 star minor leaguer in their farm system, hence you can talk them out of said player.

If anything seems to need an improvement, it's the AI's lack of improving their head scout and coaches. Just my own observation, and certainly not heavily researched.
It would almost be worth it to run a test where each team is given a perfect head scout to see how they would value their players then. If I had to guess, Markus has done this long ago when developing the AI.

Just remember that everything is filtered through the head scout's ability- so keep your salt handy.

To that end, I guess one of the ways you could make the AI teams hardier would be to go in and edit their scout staff. By boosting their scouts to higher levels of competence (and extending their contracts so the scouts stay with those teams longer), you improve their ability to assess talent, thus making them tougher. Just a thought.

I play historic solo, but randomize the historic players to get somewhat atypical results. I like Wolf's idea of turning off ratings and might try this next go round.

[PotatoPeeler]

Originally Posted by The Wolf

You also need to dial Talent Change Randomness down to around 10 in Game Setup.

Where exactly is this setting? I'm having trouble finding it in the game.

Thanks.

[bigMatt]

Originally Posted by PotatoPeeler

Where exactly is this setting? I'm having trouble finding it in the game.

Thanks.

The "Player and Picture Options" tab of the Game Setup. 5th entry on the left hand side.

[PotatoPeeler]

Excellent. Thanks!

[Isura]

About player evaluation,
Has anyone tried 100% ratings ? Does that make the AI less erratic since ratings vary less than stats?

[The Wolf]

Trust me, go high stats.

[injury log]

Good points above. I'm going to keep adding things here as I find them, and pass this on to the beta teams dealing with transactions and trading. One more:

-It's too easy to dump bad contracts. While the AI won't accept a 'grossly overpaid veteran' in trade, it will happily draft one in the Rule 5. Even if a guy can refuse demotion, if you waive him and DFA him immediately prior to the Rule 5, and if he still has *some* ability left, the AI will pop him regardless of salary. So the Rule 5 AI should be a bit more sophisticated; the AI should not only look at stats+ratings, but also at contract value among other things.

[The Wolf]

Pass this on to the beta team as well: when drafting and asking for advice on who to pick from your scouts, make the game a little more friendly to those of us who play with stats off.

1. Add a "recommend best player at position (pull down position menu)" option. That way we can tell the scouts to find the best SP, 1B, or whatever we feel we need at that point, not the best overall pick.

2. Add some preference sliders for scouts. Allow them to weight values differently. Allow the to prefer (or not) stuff, control, endurance, etc. for pitchers, and contafct, power, eye, speed, positional defense, etc. for batters.

That way your scout could pick a team that suited you while you never actually saw the ratings.

I think it would make the game better.

[endgame]

Wolf hints at some rather keen observations in our quest for a more conceptual AI. In fact, the weighted value distinctions become an effective bridge to create what we, one day, hope will be a better variety of AI reactions, including the blunders as well as the shrewd. Personality and game philosophy should differ from one organization to the other. There should, as well, be times a proven strategy is mimicked by struggling organizations in an effort to succeed. Good points, Wolf.

Another fundamental problem associated with the AI that follows its protocol, regardless of 'human' alternatives is that of what transpires after the trade, i.e. what is done with the players acquired and how they get there. While I can't attest to coded methodology, I can say that my general impression is that the AI-acquired player, more often than not let's say, under many circumstances - contract, 40-man, etc... - puts a player on DFA and waivers, typically when intending to send him to the minors. Perhaps the AI is intent on 'optioning' him to the minors, not planning on keeping him on the 40-man. Regardless, the process seems much easier for the human player - in his transaction screen - to accomplish player placement than it is for the AI. In many cases, I can and have picked a player back up from the waiver wire that was recently involved in a trade, whether that trade involved my team or not.