IGN member speak the truth about Football manager

[x42bn6]

Not everything in the game is written on top of old code. FM is an iterative process yes, but sections of the game will be totally re-written for a release if it's felt necessary. You make it sound like development at SI is a constant hack-fest. Give us a little more credit that.

What techniques do you use to write your AI? A*? Learning algorithms? Very curious.

[mantralux]

Your almost complete disdain for them and what they do, which you are continually showing in this thread, shows how little you respect them.

That's an interesting analysis. I have no disdain for SI at all. I'm just not blindly infatuated with them.

So what places have you worked that give you such insight into the games industry?

I've done a lot of work for FunCom (yes, coding), and I also work with coders across all ranges in my daily routine. Game coders, software coders, etc. Is the ******* contest done, or are we moving into namedropping territory?

[wally13]

I love the yearly experts who pop up on forums claiming what SI should't and should do and how easy it is or should be for SI to change this or change that, yet most have no clue or experience on how a game is put together !! especially one as complex as FM

[mantralux]

Not everything in the game is written on top of old code. FM is an iterative process yes, but sections of the game will be totally re-written for a release if it's felt necessary. You make it sound like development at SI is a constant hack-fest. Give us a little more credit that.

I know not everything is written on top of old code, and I have given you credit for many, many years. This seems to be yet another example of SI being incapable of receiving any type of criticism, but if you actually read what I've said in this thread, you wouldn't be as offended as you seem to be.

I've said that:

1. The code isn't magical

2. The majority of the game is good, but some basics are still problematic (specifically squad balance over time)

Nothing controversial there really, and I'm sure you'd actually agree with both those points.

[mantralux]

I love the yearly experts who pop up on forums claiming what SI should't and should do and how easy it is or should be for SI to change this or change that, yet most have no clue or experience on how a game is put together !! especially one as complex as FM

If you're referring to me, I actually do have experience on how a game is put together. And I'm not an expert. I've said that I wish SI and SEGA would be willing to skip a year so they could re-write the game from scratch. Wish. Not that this is the only correct thing to do.

[wally13]

If you're referring to me, I actually do have experience on how a game is put together. And I'm not an expert. I've said that I wish SI and SEGA would be willing to skip a year so they could re-write the game from scratch. Wish. Not that this is the only correct thing to do.

Nope was't preferring to you in particular i ment in genral as every year we seem to get these unearthed geniuses, just for the record though the code you have worked with could and probably is miles different to what SI use, im guessing there ?

[mantralux]

To get back on topic (as I do feel quite uncomfortable when moderators and official SI people gang up on forum members for daring to have an opinion):

I don't agree with everything in the review in the original post, and I don't agree with the tone of it. But I also think that the review wouldn't have been written if there wasn't some actual substance to it.

So my question is: does anyone from SI think anything in the review is valid? Or is it just dismissed as destructive criticism?

[Bertie BG]

What techniques do you use to write your AI? A*? Learning algorithms? Very curious.

I'm not a game developer so can't answer that I'm afraid.

I know not everything is written on top of old code, and I have given you credit for many, many years. This seems to be yet another example of SI being incapable of receiving any type of criticism, but if you actually read what I've said in this thread, you wouldn't be as offended as you seem to be.

I've said that:

1. The code isn't magical

2. The majority of the game is good, but some basics are still problematic (specifically squad balance over time)

Nothing controversial there really, and I'm sure you'd actually agree with both those points.

I don't see this an example of not being able to receive criticism, nor one of me being offended (I'm not). I was merely responding to your instigation that our development is held back by the limitation of always having to write over old code and that that is not the case. Just stating a fact

Oh, and for the record, no the code isn't magical and yes we always think the game can be improved. Hence why we keep working on it

[mantralux]

Nope was't preferring to you in particular i ment in genral as every year we seem to get these unearthed geniuses, just for the record though the code you have worked with could and probably is miles different to what SI use, im guessing there ?

I'm not claiming I know better, and again; code is code. My concern is that the game is in its nth iteration, and there are still some issues with very basic football mechanics. In my opinion, these should be prioritised ahead of adding fluff to the game. But I also fully understand that SI survives on yearly cycles.

[Bertie BG]

To get back on topic (as I do feel quite uncomfortable when moderators and official SI people gang up on forum members for daring to have an opinion):

I don't agree with everything in the review in the original post, and I don't agree with the tone of it. But I also think that the review wouldn't have been written if there wasn't some actual substance to it.

So my question is: does anyone from SI think anything in the review is valid? Or is it just dismissed as destructive criticism?

I agree that there are aspects of the game that can be improved, including some of the ones that were mentioned. As Neil mentioned earlier in the thread, everyone here at SI are fans of the game (many of us came to work here through being part of this very community!), and we're just as critical of some aspects of the game as any of you are. However, we can't fix everything that anyone has an issue with as a) there's a lot of disagreement about what actually are issues and b) we have to prioritise what we look at each iteration. You may disagree with those priorities, and you're entitled to do that, we just do what we think is best.

The bottom line for us is that we just try to put out the best game we possibly can on a yearly basis. That's it

[mantralux]

I don't see this an example of not being able to receive criticism, nor one of me being offended (I'm not). I was merely responding to your instigation that our development is held back by the limitation of always having to write over old code and that that is not the case. Just stating a fact

That's not what I've said though. I've said that the legacy code makes it complex, and that legacy code holds development back. This is true for all development, so I'm not sure why it wouldn't be true for SI.

Oh, and for the record, no the code isn't magical and yes we always think the game can be improved. Hence why we keep working on it

We're on the same page then.

[mantralux]

The bottom line for us is that we just try to put out the best game we possibly can on a yearly basis. That's it

I don't think anyone doubts that. I think the earlier point in the thread was that lack of competition makes SI (consciously or subconsciously) more content with certain things, as the is nothing to measure against. Again, in all other industries competition promotes development, so it must be true in SI's case as well.

And I think the competition point was how it got sidetracked into namedropping development studios. =P

[Bertie BG]

All I can say is that it's certainly not consciously if that is the case. We're our own biggest critics.

[mantralux]

All I can say is that it's certainly not consciously if that is the case. We're our own biggest critics.

No I think the guy who wrote the review in the OP is your biggest critic.

[Bertie BG]

I don't know, this guy on Twitter runs him close... https://twitter.com/jsfoao/status/226023256149331968

[mantralux]

I don't know, this guy on Twitter runs him close... https://twitter.com/jsfoao/status/226023256149331968

You win.

[thomit]

AI ... just wondering; what sort of computer does it take to be able to simulate intelligence like we want it? Given what sort of machinery we, the players of the game, has to get by with, I don't think that the AI in FM is bad at all, really.

Other games has good AI? Yes, but AI needed in a game like FM is in another league entirely, from most other games.

We would need hardware to play on that are at least on par with the very best modern chess computers (you know, those that can take on and sometimes beat human chess grand masters), to be able to have anything like "good" AI in FM.

Solution? Well, spend more money into developing online playing. Set up servers, organise leagues, create a online "world" that is easier and more user friendly, and invite people to join. When you play against other humans, there's no need for AI. Playing against other humans are infinately more gratifying than playing against AI.

The job for SI and Sega is to set up a online "world" and league system that's easy to join, where you find other players and leagues to join easily, and of course, to streamline FM for online gaming. As it is now, it is up to the customers to organise things - SI and Sega should take a much more active role in this, I feel.

[nickdc]

Hello SI, I love the game but PLEASE let the manager be able to choose the reserve team line up even if we don't want to take control of said reserves. PLEASE!

See, I've used capitals on both occasions so I must REALLY mean it.

[CaptainPlanet]

I do believe that at least one strong competitive franchies would see the quality rise significantly, as in most businesses.

But regarding the general quality of the game as it stands now, I vote with my wallet. If I think the game will provide enough entertainment per £ spent, then it gets my money, if not then I'll put my disposable income somewhere else.

I still play FM09 because I enjoy it, I haven't bought any of the versions since because from playing the demos I haven't enjoyed the changes that have been made and the direction the development of the game has taken enough for me to think that the money spent will provide a better level of entertainment than the version I'm currently playing.

[edgar555]

AI ... just wondering; what sort of computer does it take to be able to simulate intelligence like we want it? Given what sort of machinery we, the players of the game, has to get by with, I don't think that the AI in FM is bad at all, really.

Other games has good AI? Yes, but AI needed in a game like FM is in another league entirely, from most other games.

We would need hardware to play on that are at least on par with the very best modern chess computers (you know, those that can take on and sometimes beat human chess grand masters), to be able to have anything like "good" AI in FM.

Solution? Well, spend more money into developing online playing. Set up servers, organise leagues, create a online "world" that is easier and more user friendly, and invite people to join. When you play against other humans, there's no need for AI. Playing against other humans are infinately more gratifying than playing against AI.

The job for SI and Sega is to set up a online "world" and league system that's easy to join, where you find other players and leagues to join easily, and of course, to streamline FM for online gaming. As it is now, it is up to the customers to organise things - SI and Sega should take a much more active role in this, I feel.

Personally I have zero interest in playing FM online.

I'd like them to just make the AI better.

[x42bn6]

AI ... just wondering; what sort of computer does it take to be able to simulate intelligence like we want it? Given what sort of machinery we, the players of the game, has to get by with, I don't think that the AI in FM is bad at all, really.

Other games has good AI? Yes, but AI needed in a game like FM is in another league entirely, from most other games.

We would need hardware to play on that are at least on par with the very best modern chess computers (you know, those that can take on and sometimes beat human chess grand masters), to be able to have anything like "good" AI in FM.

Chess required (important: not requires) monster computers because chess's biggest problem is a searching problem. Every chess AI has the same rough steps: Look at the board, look at every possible move up to a certain depth (with pruning of course), and decide what is best.

And no, you don't require ridiculous computing power to create a good chess AI. Deep Blue, the most famous of chess AIs (that beat Kasparov in 1997), has has a peak of 15.4 GFLOPS (http://i.top500.org/site/830). You can match that with a high-end i7 nowadays: http://www.sisoftware.eu/rank2011d/show_device.php?q=cca598d1bfcbaec2e2a1cebcd9f990a78ab88ebe8ec5e384b994a583f1ccfddbb28fbe98f0cdf8dea69baa8ce98cb181a7d4e9d1&l=en GPU processing is even higher. Modern-day supercomputers can likely beat chess grandmasters nowadays simply by virtue of Moore's Law alone, but Deep Blue would still beat a lot of IMs and GMs in chess today.

Having said that, chess's search problems are nasty because it is EXPTIME complete - the more paths you evaluate, the space and time required to solve a deeper problem increases exponentially.

Do we need a similar AI for Football Manager? Probably not. Chess is a list of discrete moves, but, say, football matches are not. Chess has perfect information, while football transfers don't (you might not know the opposing bids' warchests).

Things like squad-building sound exactly like what Markov decision processes are meant to do and don't require stupid amounts of processing power (after all, they were messing with these in the 1950s). Using this allows the processes to reduce to Markov chains, which have lots of analytical properties (i.e. easier to balance).

Football matches are closer to multi-agent cooperative simulations as tree searching is less important (I don't think even Barcelona can plan a 30-move goal from scratch - much of the thinking and decision-making is done on-the-fly) and there's a large element of randomness (i.e. a certain player's passing is not always perfect). However, there are a few cases where you could use tree-searching - planning two-passing or three-passing manoeuvres, or set-pieces. Otherwise, there's of course no easy way of doing this - it will likely involve 22 players as individual agents with specific traits (i.e. "runs with ball often" => player will be in the "dribbling state" more often than most). Programming Game AI By Example has a chapter dedicated to 5-a-side (that can be generalised to 11-a-side) which isn't actually that complex - the difficulty comes in fine-tuning it, making sure the state machine has the required states to drive the game and making sure the logic is correct. At first glance 22 agents is nothing in terms of processing power needed nowadays, although I guess it's different when the gameworld could be simulating tens of games in batch.

[Coentrao]

Personally I have zero interest in playing FM online.

I'd like them to just make the AI better.

I feel the same way.

[mantralux]

Other games has good AI? Yes, but AI needed in a game like FM is in another league entirely, from most other games.

Yes = less.

In FPS games for example, the AI must be able to instantly adapt to your actions. The AI doesn't know your next move, and must therefore react with speed to any decision you make, be it inside the box or outside the box thinking.

Football Manager has the advantage of time. It can take a whole processing cycle for the AI to respond to your actions, it doesn't have to happen instantly.

I'm talking about squad balance now, not match AI.

[CaptainPlanet]

AI wouldn't be such a talking point if it at least did the basics well in terms of squad balance.

The AI getting things like tactics, training or motivation wrong is far down the line, just stop them from buying 4 world class goalies each on £100k a week.

[Ackter]

Wait, so you think FPS games have good AI?

Every type of character in an FPS has exactly the same ruleset. None of them actually make decisions, pretty much all of their AI is "if a happens, do b" (and that's being generous, as half the games out there just have "do a then do b then do c repeat". FM is slightly more complex than that...

[Svenc]

The AI is most definitely weaker than a human mind, and needs a lot of improvement.

However, when a company with the resources of EA can't even get close to challenging FM, let alone challenging a human, it says a lot about how difficult it is to get right, and even more about the chances of a competitor ever being good enough to challenge.

From 2001 onwards, all EA did themselves was getting ahold of the chief designers of the German "On The Ball" series, watch, and sit back. That is a little exaggerating, but EA never put that many ressources into a management game. In 2005ish there was a split between EA Canada and the subsidiary in Germany that was headed by those "On The Ball" team members, presumably because EA didn't think the investment was worth it anymore. The independent outlet that is Bright Future, based in Cologne, has been responsible for the development of the FIFA Manager series for pretty much as long as it is has been around. It certainly is nice for EA to have around, since it pretty much dominates the German speaking management market due to a various of reasons, but it is by no means imperative and something that is granted a quadruple of resources - the core of the 3D match engine at work dates several FIFA Soccer iterations back already. Bright Future are also a smaller studio than SI. But as fun as the games of old were back then, its chief designer, which is the same guy Electronic Arts lured away from "On The Ball" back then, has a nasty track record of failing to portrait the actual football in his football management games.

Put short: Not gonna happen so soon.

I'd like to see some competition myself though, even if it were just for different spins on the same basic ideas. There is one currently in the making, presumably with Gianluca Vialli involved in some form. It looks more like a cross between The Sims and On The Ball more than anything though.

[Spart]

Personally I have zero interest in playing FM online.

I'd like them to just make the AI better.

Couldn't agree more. Football Manager is, for me, the greatest single player game around due to the personalised "world" that develops throughout saves, and the stories that are organically created and play out as time goes on.

I don't want to have to deal with other's people's out-of-game ego's and needs. I want an immersive, complex and realistic football sim ready for me to lose myself in on my terms, whenever I want.

[x42bn6]

Wait, so you think FPS games have good AI?

Every type of character in an FPS has exactly the same ruleset. None of them actually make decisions, pretty much all of their AI is "if a happens, do b" (and that's being generous, as half the games out there just have "do a then do b then do c repeat". FM is slightly more complex than that...

Not true. Quake 3 uses goal-based AI and pathfinding (which means you can design a custom map and the AI will generate paths off it). Half-Life 2 uses paths, hints and schedules.

Newer FPSes doubtless build upon these.

[Ackter]

Pathfinding is not complex, and in the cases of custom maps it'd be a one-off calculation when the map is generated.

[x42bn6]

Pathfinding is not complex, and in the cases of custom maps it'd be a one-off calculation when the map is generated.

Alright, Quake 3 uses goal-based AI, while Half-Life 2 uses paths, hints and schedules.

Still more than a bunch of "if" statements.

[edgar555]

Oh x42 I do love your posts.

But Ackter is right, FPS AI is on the whole terribly simplistic. When Halo was released the AI blew everyone away and no one's really bettered it since and even that isn't exactly tough.

[Ackter]

Alright, Quake 3 uses goal-based AI, while Half-Life 2 uses paths, hints and schedules.

Still more than a bunch of "if" statements.

Yes I agree. But anywhere near as complex as the AI that's going on in FM?

Remember the sheer number of calculations that go on every time you press continue in FM. Even with only one league loaded and a small database that still involves thousands of players and staff. Now I'm not saying every single one of them gets calculated every time but even a small percentage of them is a huge amount of work, and certainly more than what's required for the 50 or so AI characters you'd get loaded in memory at any one time in an FPS.

[x42bn6]

Oh x42 I do love your posts.

But Ackter is right, FPS AI is on the whole terribly simplistic. When Halo was released the AI blew everyone away and no one's really bettered it since and even that isn't exactly tough.

It's not simplistic at all. Take a look at Valve's own Source SDK documentation: https://developer.valvesoftware.com/wiki/Category:AI

It certainly seems simplistic in that humans quickly reach and surpass AI levels, but we're talking about the underlying complexity of the AI (which is why everyone is saying that FM's AI is "complex"), not how easy it is to beat them.

[Ackter]

Incidentally, one of the best looking pathfinding code I've seen is in the game Viking on the 360. It can involve hundreds of enemies on screen at once, and they're all reacting to what's around them and finding ways to get to places. But even that is simplistic.

On a related note, Skyrim's pathfinding is horrible.

[edgar555]

It's not simplistic at all. Take a look at Valve's own Source SDK documentation: https://developer.valvesoftware.com/wiki/Category:AI

It certainly seems simplistic in that humans quickly reach and surpass AI levels, but we're talking about the underlying complexity of the AI (which is why everyone is saying that FM's AI is "complex"), not how easy it is to beat them.

Ok, I understand your point. I still don't think people should be comparing FPS AI to something like FM. In my limited knowledge they seem very different beasts.

[x42bn6]

Yes I agree. But anywhere near as complex as the AI that's going on in FM?

It could be. Look at the Source SDK.

Remember the sheer number of calculations that go on every time you press continue in FM. Even with only one league loaded and a small database that still involves thousands of players and staff. Now I'm not saying every single one of them gets calculated every time but even a small percentage of them is a huge amount of work, and certainly more than what's required for the 50 or so AI characters you'd get loaded in memory at any one time in an FPS.

An FPS is close to real-time, while in FM, you can afford to wait until other games have finished simulating. An FPS would never have the workload of what is essentially lots of batch processing in FM, so it's misleading to compare the number of calculations. If we look at the match engine, perhaps a fairer comparison would be an 11v11 CTF; and if FM simulates, say, 10 games (i.e. the final day of the Premier League), you'd look at 10 simulations of 11v11 CTF games to see how long they take.

A possibly more accurate comparison would be RTSes, too. Here, for example, in this Starcraft example, both sides are controlling around 100 units each:

The AI for an individual Starcraft unit is fairly rudimentary (just dynamic pathfinding I believe), but it's a lot more complex than 11 on 11, and has a lot higher overheads (i.e. base maintenance and macro).

[Ackter]

The FM match engine generates 45 minutes of football in approx. 2 seconds at the start of every match.

FPS have 5 people on screen who need to decide whether to duck behind cover or not.

[Ackter]

Yup, Star Craft, Total War etc are complex as hell. These are the sorts of games you can compare with FM.

(Though you should have picked an AI vs AI vs AI match from Starcraft 2 rather than two human controlled teams as an example)

Especially Total War.

[x42bn6]

The FM match engine generates 45 minutes of football in approx. 2 seconds at the start of every match.

FPS have 5 people on screen who need to decide whether to duck behind cover or not.

5? Why would you only play 5 AIs? Tribes, for example, can have good old 32v32 carnage-fests. Some maps in Counter-Strike: Condition Zero support 32 player games.

And it clearly does not take 2 seconds for the AI to decide whether to duck or not. It takes 2 seconds because the player is involved - it's a real-time game. Processors nowdays perform billions of floating-point operations - they'd have no issues processing an entire AI vs. AI game in seconds as well. It would be like arguing it takes around 15 minutes to simulate a football match in FM because when the user hits continue and jumps into the match, tinkering with tactics and ranting at the referee, full-time occurs in 15 minutes.

There are AI bot leagues that simulate thousands of games per day to see which bots are better. I don't know if there are any FPS ones, possibly because it's a difficult thing to program, but I know there is a Brood War one somewhere.

[x42bn6]

Yup, Star Craft, Total War etc are complex as hell. These are the sorts of games you can compare with FM.

(Though you should have picked an AI vs AI vs AI match from Starcraft 2 rather than two human controlled teams as an example)

Especially Total War.

That link is two bots playing against each other. The bots are custom-written for an AI competition.

There's no point in using Starcraft 2 because it lacks the modding capabilities of Starcraft. In fact, Blizzard won't let you write bots for Starcraft 2. So Starcraft 2's AI is probably even worse than the AI posted in the link, because Starcraft 2's AI cheats (i.e. it gathers resources at a higher rate than the player).

[Ackter]

The sheer amount of calculations that need to take place in that few seconds far outweighs the 'decisions' needed to be made at any one point in time during an FPS game. Also remember that an FPS isn't loading and calculating everything at once - it does it in batches whenever you cross certain points of the level.

Also Tribes probably not the best example, seeing as it's a bit rubbish. It's pretty much got worse with every release.

[Ackter]

That link is two bots playing against each other. The bots are custom-written for an AI competition.

There's no point in using Starcraft 2 because it lacks the modding capabilities of Starcraft. In fact, Blizzard won't let you write bots for Starcraft 2. So Starcraft 2's AI is probably even worse than the AI posted in the link, because Starcraft 2's AI cheats (i.e. it gathers resources at a higher rate than the player).

Ah right, the description didn't make that very clear.

[x42bn6]

The sheer amount of calculations that need to take place in that few seconds far outweighs the 'decisions' needed to be made at any one point in time during an FPS game. Also remember that an FPS isn't loading and calculating everything at once - it does it in batches whenever you cross certain points of the level.

It doesn't do it in batches. Whenever the state changes (which is usually all the time) or the next cycle of processing comes round, it reevaluates what it needs to do. In Quake 3, it reevaluates its goals whenever it takes damage (as damage is part of the state engine), or an(other) opponent appears in its field of vision, or a game event happens (i.e. enemy flag captured => protect capper), etc. This is why it's a real-time game - it processes things in real-time.

Certainly, specific events might trigger a lot of batch processing, such as flags being captured, but by-and-large the AI operates in real-time.

Look at what Valve's Source SDK does:

- https://developer.valvesoftware.com/wiki/Decision_Making_Overview

- https://developer.valvesoftware.com/wiki/Think%28%29

Also Tribes probably not the best example, seeing as it's a bit rubbish. It's pretty much got worse with every release.

The quality of the game doesn't really matter. Tribes's AI has to be even more difficult to program than most FPSes because the bots spend more time in the air than on the ground, so pathfinding and combat is miles more complex.

[Ackter]

Well, yeah it does. If the game is crap, so is the AI.

[Ackter]

Regarding the Valve AI, what's so hard about that? Instead of "If A, do B" it's "If A, B and C are happening, do D". That'll be the same for pretty much every NPC generated.

And I meant the levels themselves are loaded in batches (with the exception of small scale maps like multiplayer maps).

[x42bn6]

Well, yeah it does. If the game is crap, so is the AI.

A pretty boring game could have perfect AI (if the game has been solved, like checkers).

So it is entirely possible for a crap game to have good AI, in the same way that a crap game could have good graphics (just that the graphics can't redeem the game).

[Ackter]

A pretty boring game could have perfect AI (if the game has been solved, like checkers).

So it is entirely possible for a crap game to have good AI, in the same way that a crap game could have good graphics (just that the graphics can't redeem the game).

It's the AI, and the human's interaction with that AI, that is the be-all and end-all of every game. Graphics can compensate to a certain degree, or they can make it worse. But superficial things like graphics are never enough to make or break a game. If a game is good, it's good despite poor graphics. You don't get a good game despite poor AI.

[RBKalle]

Why bringing action games' AI into the picture???

FM's biggest flaw is AI's questionable signings, which can be rooted to Reputation and PA [actual or perceived, it barely matters] having too much weight in the transfers and selection dynamics.

I'm no code expert but assuming the AI evaluates every potential signing by "looking" at his CA,PA, Reputation and (maybe) position, and compares those values with the "status" of the club, couldn't a better balance of those factors be a decent, albeit partial, solution to the issue?

I know it's probably too simplistic, but it wouldn't likely require to rewrite huge parts of the module...

Just make better managers/top clubs be more demanding in terms of quality. Currently way too many mediocre players get purschased by clubs which IRL wouldn't even sign them for free.

[Ackter]

Why bringing action games' AI into the picture???

FM's biggest flaw is AI's questionable signings, which can be rooted to Reputation and PA [actual or perceived, it barely matters] having too much weight in the transfers and selection dynamics.

I'm no code expert but assuming the AI evaluates every potential signing by "looking" at his CA,PA, Reputation and (maybe) position, and compares those values with the "status" of the club, couldn't a better balance of those factors be a decent, albeit partial, solution to the issue?

I know it's probably too simplistic, but it wouldn't likely require to rewrite huge parts of the module...

Just make better managers/top clubs be more demanding in terms of quality. Currently way too many mediocre players get purschased by clubs which IRL wouldn't even sign them for free.

I'd say its the ability to judge their own squad correctly that causes most of the problems. It's what's causing them to look for the wrong players that leads to them having 5 strikers when they only play one up front, for example.

The ability to judge a player definitely needs some work, but that's not the crux of the problem imo.

[x42bn6]

Regarding the Valve AI, what's so hard about that? Instead of "If A, do B" it's "If A, B and C are happening, do D". That'll be the same for pretty much every NPC generated.

Anything that has a state machine of sorts will have lots of conditionals. FM will likely have these as well because a state machine is the easiest way of doing AI (i.e. a player either has the ball or not - one state machine).

Conditionals are just part of any programming language.

The reason why it's not that simple is that it continuously evaluates its position every n seconds (depending on the AI). In Valve's Source engine, each AI has a schedule that it is currently performing, with a list of tasks to perform that schedule. There's logic to plan which schedule to take if it does not have one, and there's logic that reevaluates its schedule depending on what happens (i.e. it is complete, or some condition happens, i.e. it takes damage). It will boil down to conditionals, yes, but you may as well say "every AI has source code" rather than "every AI is just if-then-else".

And I meant the levels themselves are loaded in batches (with the exception of small scale maps like multiplayer maps).

We are talking about multiplayer maps, no? It is just that our "players" are AIs. Single-player maps of course are entirely different but those have different purposes where the user is explicitly involved in some quest of sorts. Single-player maps are rife with hacks and cheats, too, to keep the storyline going (i.e. lock off some areas until the player has reached a specific checkpoint).

In terms of competing AIs, it only makes sense to talk about multiplayer maps. Here, every AI and player has the same goal.