AI to AI trade logic is awful

[Lukas Berger]

Originally Posted by sutnopia

Tone that down for the next patch means the patch available as public beta? (24.2.44)?

Thanks in advance!

No, next after that.

[jfb8300]

Originally Posted by Lukas Berger

I do suspect the AI may not take injury proneness and injury history into account enough when calculating trades.

This!



On a side note, is injury proneness/history a consideration for the AI when making free agent offers?

And could both possibly be made available to the player as a setting? Would like to have it set at Very High for my sims.

[sutnopia]

Originally Posted by Lukas Berger

No, next after that.

Thanks Lukas!

[wallewalls]

Originally Posted by jfb8300

This!


On a side note, is injury proneness/history a consideration for the AI when making free agent offers?

And could both possibly be made available to the player as a setting? Would like to have it set at Very High for my sims.

It certainly hasn't seemed that way in my experience throughout the years. So many times do wrecked players get signed to big-dollar, long-term deals.

[Charlie Hough]

Originally Posted by Lukas Berger

The idea that the AI should somehow be far, far better than real, live human GM's, with all the countless terrible trades they've made, and never do anything questionable is just kind of hard for me to get a grip on.

You certainly have a tough job in trying to balance all of this out, but ideally there need to be two options. One is to have the AI be "realistic" in the sense of simulating real-life GMs and their dumb moves. The other is for the AI to be "realistic" in the context of a baseball sim, where it's not competing against real-life human GMs but against a human user and other AI who have access to player ratings, data and game settings that real-life GM's don't have.

A human baseball sim user is extremely unlikely to make the same mistakes as a real-life GM. If the AI is not programmed to analyze and make decisions according to in-game baseball sim factors with a similar level of reasoning and reliability as the human user, then it is at huge disadvantage. It also tends to annoy users who want a more competitive AI where the human user and AI can compete on more of a level playing field.

I've been playing baseball sims for nearly 40 years, and this has always been an issue. An AI that can truly compete with the human user has been elusive for decades. OOTP has done a great job of creating the most competitive baseball sim AI to date, but the next step is to close the remaining gap by creating an option where the AI doesn't randomly do dumb things based on external reality. That should be there for those who want it, but we need an option for those who want a level playing field with the human user.

[Dave Stieb II]

Originally Posted by jcard

While the underlying argument here does not even merit a rebuttal, even the Categorical Bad Trade of baseball history it cites is misrepresented. Everyone knows that there were extra-baseball considerations behind that trade. The consideration received for Ruth was not a bunch of Quad-A players, but $100-125k and a 300,000 owner-to-owner loan.

What is this?
You are using historical facts to counter a stereotypical, oft-repeated passive aggessive response to a concern several customers have about a component of the game that was worked on and adjusted over the off-season.

[md40022]

It's not like there is a "make good trades" option that the programmers just mistakenly set to "off."

The nature of a simulation is that outcomes are going to be unpredictable. If you and I trade the quarters in our pocket and then flip the traded quarters 100 times, one of us may get heads 60 times and the other might only get heads 45 times.... and these are the exact same identical "players", each with a .500 batting average for getting heads, that we just traded. Yet their simulations played out differently though. Did we make a "bad" trade or did the simulation just play out different for two very similar players (identical players in this case)

The data that a few people posted earlier with the amount of trades that were happening was definitely a little out of whack it seemed, and if the formulas and algorithms that the AI uses in evaluating trades needs some fine tuning, so be it. That very well might be the case. I'm not deep enough into the game yet to have a real opinion on it. It seems like some tweaking from the programmers might be needed.... That said though, the idea that the AI should only make "good" trades at all times is kind of unpractical though just based on the entire idea behind what a sim actually is.

[Lukas Berger]

Originally Posted by Dave Stieb II

What is this?
You are using historical facts to counter a stereotypical, oft-repeated passive aggessive response to a concern several customers have about a component of the game that was worked on and adjusted over the off-season.

Facts which don't move the needle of the discussion at all, given that it was still one of the worst trades in baseball history and still would have been even with multiple millions of dollars in 1919 currency thrown in.

[Lukas Berger]

Anyway, obviously the intention is not to have the AI make bad trades, but when you take into account all the countless variables involved, as md40022 alludes to above, realistically it's impossible not to have this happen occasionally both for real-life human GM's and for the AI.

But for sure, it is possible not to have this happen too often with the AI, and not to have trades in general happen too often, which as we've already stated repeatedly, is something we'll be looking at closely.

[jkeller]

Not trying to pile on or pick apart the Trade AI more than what's already been done but just thought to provide another example of a fairly outrageous trade, albeit in a contract year for Ohtani but at such an early stage of the season, retaining a majority of the contract, and for such an underwhelming return felt worth noting.

[Cactusguy21]

Originally Posted by jkeller

Not trying to pile on or pick apart the Trade AI more than what's already been done but just thought to provide another example of a fairly outrageous trade, albeit in a contract year for Ohtani but at such an early stage of the season, retaining a majority of the contract, and for such an underwhelming return felt worth noting.

That one is still justifiable bad. I'm fine with bad deals being made because they get made all the time. It's just the unrealistically bad ones that are rough.

[Lukas Berger]

Originally Posted by jkeller

Not trying to pile on or pick apart the Trade AI more than what's already been done but just thought to provide another example of a fairly outrageous trade, albeit in a contract year for Ohtani but at such an early stage of the season for such an underwhelming return felt worth noting.

I would be surprised if he fetches much more than this if he's traded in-season this year without an extension agreed to in advance.

I mean, maybe depending on the team it's more like #20, #60 and a borderline top 100 guy. But unlikely to be much more imo.

In the current climate, teams are not going to give up truly elite prospects for anyone, no matter how elite, that's in their walk year without a contract extension agreed to. Maybe 10-15 years ago someone would have.

Of course, I could be wrong.

[PSUColonel]

Originally Posted by Lukas Berger

I would be surprised if he fetches much more than this if he's traded in-season this year without an extension agreed to in advance.

I mean, maybe depending on the team it's more like #20, #60 and a borderline top 100 guy. But unlikely to be much more imo.

In the current climate, teams are not going to give up truly elite prospects for anyone, no matter how elite, that's in their walk year without a contract extension agreed to. Maybe 10-15 years ago someone would have.

Of course, I could be wrong.

No..you are correct…it makes fiscal sense too!

[Matt Arnold]

FYI Freeman would also likely be in the top-100 prospects if he qualified, I think he just passed his rookie eligibility last year.

I mean, that's the sort of deal that wouldn't happen, because Ohtani's true value especially when you count market and merchandising and reach would massively surpass anything we could reasonably represent in-game, IRL they wouldn't trade him until the last moment, and even then, you almost feel they would hang on and force someone to outbid them to steal him.

In terms of the deals listed here, other than those factors which we really will never be able to fix unless if we simply make it impossible to ever trade for someone like him, I find that deal a lot less objectionable than a lot of the others mentioned in this thread.

[Dave Stieb II]

Originally Posted by Matt Arnold

FYI Freeman would also likely be in the top-100 prospects if he qualified, I think he just passed his rookie eligibility last year.

I mean, that's the sort of deal that wouldn't happen, because Ohtani's true value especially when you count market and merchandising and reach would massively surpass anything we could reasonably represent in-game, IRL they wouldn't trade him until the last moment, and even then, you almost feel they would hang on and force someone to outbid them to steal him.

In terms of the deals listed here, other than those factors which we really will never be able to fix unless if we simply make it impossible to ever trade for someone like him, I find that deal a lot less objectionable than a lot of the others mentioned in this thread.

I agree.It would be very difficult to replicate everything the unicorn truly brings to the table so this package of genuine prospects doesn't look all that bad to me.
I have a bigger problem with the trade just above this one - I know the Jays are in constant need of BP upgrades but would they really trade Bargar and another prospect to recycle Ryan Tepera? That trade seems to fit the pattern some others have described earlier. Paying way too much for aging relievers.

[Lukas Berger]

Originally Posted by Charlie Hough

I've been playing baseball sims for nearly 40 years, and this has always been an issue. An AI that can truly compete with the human user has been elusive for decades. OOTP has done a great job of creating the most competitive baseball sim AI to date, but the next step is to close the remaining gap by creating an option where the AI doesn't randomly do dumb things based on external reality. That should be there for those who want it, but we need an option for those who want a level playing field with the human user.

The reasons for this are pretty unavoidable.

Even leaving aside the question of whether we would want to have the AI never make bad trades with itself, which I do think is very much a debatable thing, the problem is that programming a truly realistic and challenging AI in a game with as many variables as OOTP is hard. Really hard.

I think I mentioned 100 high level Google devs working a year to make what people might consider a genuinely "good" trade AI at some point in this thread, that would be a real challenge to humans and that's probably not really too much hyperbole.

It would for sure take a large team whose price range is way beyond anything we could afford to really make a massive dent in this.

Games like OOTP, even as much as we've grown, are way too much niche games to be able to afford to hire the amount of staff that it would take to do something as challenging as this. Never mind those other baseball sims you mention.

Even a game like FM gets an enormous amount of complaints about their AI, when they likely have as many folks working on AI specifically as we do on the entire game.

We do give this significant time and attention each year though, to the best of our abilities. While the year to year progress can sometimes be harder to discern, especially early in the cycle, I think if you compare OOTP at five or ten year waypoints, you'll see that we have made a discernable amount of progress in building a better AI. Even so, it will likely always be a work in progress to some degree.

[jcard]

Originally Posted by Dave Stieb II

What is this?
You are using historical facts to counter a stereotypical, oft-repeated passive aggessive response to a concern several customers have about a component of the game that was worked on and adjusted over the off-season.

Hmm. I was actually supporting the concerns. The proposition that a trade from over 100 years ago (when baseball was different in both economics and the level of administrative expertise) that had (conveniently omitted) financial considerations somehow mitigates the dissatisfaction with the examples of bad trades reported in this thread is absurd.

[wallewalls]

Originally Posted by jkeller

Not trying to pile on or pick apart the Trade AI more than what's already been done but just thought to provide another example of a fairly outrageous trade, albeit in a contract year for Ohtani but at such an early stage of the season, retaining a majority of the contract, and for such an underwhelming return felt worth noting.

I don't think that one is bad actually. Williams and Rocchio are very well-rated in the game and are more than fair for Ohtani without an extension.

If we are bringing up Ohtani though, I would love if it was possible for his contract demands after the first season to be a bit higher. He signed for less than his current IRL salary in my game (and with the Reds to boot).

At this point, I'm just debating going 100% fictional.

[shakturi101]

Originally Posted by Charlie Hough

A human baseball sim user is extremely unlikely to make the same mistakes as a real-life GM. If the AI is not programmed to analyze and make decisions according to in-game baseball sim factors with a similar level of reasoning and reliability as the human user, then it is at huge disadvantage. It also tends to annoy users who want a more competitive AI where the human user and AI are on more of a level playing field.

I've been playing baseball sims for nearly 40 years, and this has always been an issue. An AI that can truly compete with the human user has been elusive for decades. OOTP has done a great job of creating the most competitive baseball sim AI to date, but the next step is to close the remaining gap by creating an option where the AI doesn't randomly do dumb things based on external reality. That should be there for those who want it, but we need an option for those who want a level playing field with the human user.

Can you name a game in the strategy/management market that has been able to do this well? I've played paradox games (EU4/Stellaris), civ games, Football Manager, and the most popular complaint across all communities in the game is the inability to create an AI that can adequately challenge a human. I'm not an expert on how to create good AI and don't know what is going on under the hood, but if all these companies, including some with lots of resources, are struggling to create a challenging AI, it would seem it's not really that easy to just "fix."

The games like EU4/Stellaris/Civ just have the AI cheat in higher difficulties by giving them bonuses, which does help a lot, but once you get past the initial phase where the human is weaker, the human is able to snowball and become insanely powerful. Also, playing with huge advantages to the AI sometimes requires specific actions that feel exploity because the actions only make sense with the knowledge of the difficulty settings you are using and, therefore, break immersion.

Another way games can mitigate this issue is by having natural challenge (that don't necessarily break immersion) built into certain starting points for a human. For example, in EU4, you can start as a really small nation with little resources, which can make it really hard to ultimately compete with the big countries like France and the Ottomans. In Football Manager, you can start in a really low tier of Football (like the sixth tier in England) with really low finances and try to work your way up all the way to the top of the league from there.

OOTP has this as well, but most people, rightly so, want to play in MLB. But even playing the MLB team with the lowest resources (and they even get revenue sharing lol) pales in comparison to the natural resource gap you can create for yourself in a game like EU4 or Football Manager. As much as baseball fans complain about the resource disparity in MLB, it pales in comparison to the European soccer leagues, especially for teams in the lower leagues.

Ultimately, in all these games, if you are a competent human player who understands the game's mechanics, you will consistently make better decisions than the AI in every area, which will lead to the human outperforming the AI based on their starting point. But playing with the A's with an endboss of the Yankees/Dodgers isn't quite the same in difficulty as a non-league English club with an endboss of Manchester City.

[jcard]

Originally Posted by Lukas Berger

Facts which don't move the needle of the discussion at all, given that it was still one of the worst trades in baseball history and still would have been even with multiple millions of dollars in 1919 currency thrown in.

1) At least I had the facts correct for YOUR reference trade.
2) If you think that such a fallacious justification warrants anything beyond dismissal, that is rather discouraging.

To be clear- I think that trying to get AI realistic and flexible / believably fallible must be very difficult to achieve and require quite an investment of resources. That said, that the default settings could be so inferior to what a user says he has found is possible by testing alternative settings makes me wonder if the same testing is performed in-house. Regardless, my central issue was not with the quality of the AI, but the quality of the response to the concerns communicated by others in this thread.