If SI only overhaul one section of FM for FM15, what would you choose?

[x42bn6]

The real question is; can artificial intelligence ever be intelligent enough to compete with humans regarding the arbitrary decision to replace a player that is happy and playing well?

The answer is; no.

Therefore, the solution is to mimic human behaviour with under-the-hood robotic behaviour. Forcing the AI to make that decision may be "an utterly awful idea", but it is sadly the only one that will ever function.

I don't see why the answer to this question is "no". We have a lot of processing power in this generation of computing (we had a lot in the past generation, too). This scenario is not exactly the most complex task in the world - you don't need to consider 50 transfer "moves" in advance.

Consider this - the AI is already able to plan things somewhat: It has a shortlist of its own, which is why you see "Wnt" icons on your screen. The decision to sell a player will just add another "branch" in the AI's thinking - to sell the player. A utility function (which is a function of money, squad quality, time, etc.) can be used to value the utility in selling the player versus a few extra steps ahead (i.e. selling a player and buying a replacement - one step ahead). This isn't a difficult AI problem at all - it's a simple decision tree.

[BiggusD]

I don't see why the answer to this question is "no". We have a lot of processing power in this generation of computing (we had a lot in the past generation, too). This scenario is not exactly the most complex task in the world - you don't need to consider 50 transfer "moves" in advance.

Consider this - the AI is already able to plan things somewhat: It has a shortlist of its own, which is why you see "Wnt" icons on your screen. The decision to sell a player will just add another "branch" in the AI's thinking - to sell the player. A utility function (which is a function of money, squad quality, time, etc.) can be used to value the utility in selling the player versus a few extra steps ahead (i.e. selling a player and buying a replacement - one step ahead). This isn't a difficult AI problem at all - it's a simple decision tree.

If it is that simple why has the problem been prevalent for 20 years now, and only marginally improved in FM14?

Here's Real Madrid's squad 2022, ordered by appearances:

Admittedly, it is not as bad as in FM13, but I have payed attention to their first eleven since I started the game in 2017 (especially against me) and they have largely relied on players already at the club in 2013 - or who has been there a very long time. Lewandowski, Ronaldo, Bale, Isco, Ramos, Casillas, Wanayama and Eriksen have all been fielded to boredom against me.

True, they have been rather active on the thransfer market, rarely buying for less than £60m, but it has had little impact on the first eleven.

I must point out that the transfer market is suffering from nearly every player being either transfer listed with a knockdown or not for sale unless you pay stupid money. The "stupid in stupid out" counter-argument to the oft-repeated "AI clubs refusing to sell for ridiculous money" complaint does not hold water since logically no small club is going to say no to an amount of money that could buy the whole league system, yet in FM they do just that. I really don't think the solution to "talent-hoarding" is to make clubs refuse to sell, but to make it impossible to identify great talent before they have actually proven themselves.

A more rigid AI transfer policy could rectify this if done properly, alongside a general improvement of the decision-making process. If there is an actual processing date for future transfers (i.e 1st of June and 1st of December - both a year in advance) instead of trying to mimic opportunism, and that policy is enforced, then yes the manager in charge would probably affect the squad more than before. Too much, you guys say - I say it's high time!

Of course, such a radical change as this would require tuning. It is likely that some sell/buy decisions would be awful, but with this rigidity as a framework nuance could be added later - the SI way...

[milnerpoint]

If it is that simple why has the problem been prevalent for 20 years now, and only marginally improved in FM14?

Because its not as simple as he makes out, it rarely is, but thats not to say there isnt a lot of room for realistic improvements.

[x42bn6]

If it is that simple why has the problem been prevalent for 20 years now, and only marginally improved in FM14?

Nobody has thought about it? Delays? Mistakes? Fear of change? Not a priority? There are a lot of possible reasons why things don't get done in software. Difficulty is just one of the possibilities.

A real transfer policy builds on squad policies, such as desired squad size and usage of youngsters. A lot of poor clubs can't afford to do one-in, one-out, so don't - which is why your idea is a bad one.

Instead of enforcing transfer numbers, the game should simply be capable of valuing its squad utility (value being a function of squad quality, bank balance, etc. - a utility function, in other words) and predicting its future utility, and using those numbers to dictate its transfer policy (which will affect the predicted future numbers). For example, if a club has many key players approaching the age of 30, it knows the utility will drop dramatically in a few years time, so it must do something about it (i.e. promote youngsters, sign young players who can grow into those roles, or consider stop-gap signings in the future).

Because its not as simple as he makes out, it rarely is, but thats not to say there isnt a lot of room for realistic improvements.

I'd love to hear why you think it is difficult.

The idea I've posed uses decision trees and utility functions, well-known and widely-used concepts in AI. It is a task that can be made easily-parallel, as each branch can be computed in parallel. It involves adding and multiplying numbers together, and our CPUs nowadays can do billions of these per second. SI have a history of historic transfers (their database), and can therefore "travel back in time" and use these to compute the utility function weights for balance.

Your statement might have been true in the 80s or early-90s, where we would have struggled to compute these in a reasonable amount of time.

[BiggusD]

Nobody has thought about it? Delays? Mistakes? Fear of change? Not a priority? There are a lot of possible reasons why things don't get done in software. Difficulty is just one of the possibilities.

A real transfer policy builds on squad policies, such as desired squad size and usage of youngsters. A lot of poor clubs can't afford to do one-in, one-out, so don't - which is why your idea is a bad one.

Instead of enforcing transfer numbers, the game should simply be capable of valuing its squad utility (value being a function of squad quality, bank balance, etc. - a utility function, in other words) and predicting its future utility, and using those numbers to dictate its transfer policy (which will affect the predicted future numbers). For example, if a club has many key players approaching the age of 30, it knows the utility will drop dramatically in a few years time, so it must do something about it (i.e. promote youngsters, sign young players who can grow into those roles, or consider stop-gap signings in the future).

I'd love to hear why you think it is difficult.

The idea I've posed uses decision trees and utility functions, well-known and widely-used concepts in AI. It is a task that can be made easily-parallel, as each branch can be computed in parallel. It involves adding and multiplying numbers together, and our CPUs nowadays can do billions of these per second. SI have a history of historic transfers (their database), and can therefore "travel back in time" and use these to compute the utility function weights for balance.

Your statement might have been true in the 80s or early-90s, where we would have struggled to compute these in a reasonable amount of time.

Yes sounds like a nice idea, but even this one would be thwarted by SI's attempt at creating an opportunistic AI that tries to get bargains and won't do much if it can't. There needs to be a decision to act at some point, and my idea (1st commandment: there shalt be a replacement in every window, or thou'st burn, heathen!) will ensure that.

If there is no money to sign for, the replacement for "aging player X" would have to be either "free player Y" or "own youngster Z". Don't apply the worst possible meaning to "one-out-one-in".

Even if there is no need for a change at the calculation date 1st of June/December, there might be in two, three or four years. Programming a fantazillion if-sentences to give the AI a ramshackle long-term planning capacity is not necessary, when the same function could be served by having a stronger enforcement of the future plans in tthe first place.

[x42bn6]

Yes sounds like a nice idea, but even this one would be thwarted by SI's attempt at creating an opportunistic AI that tries to get bargains and won't do much if it can't. There needs to be a decision to act at some point, and my idea (1st commandment: there shalt be a replacement in every window, or thou'st burn, heathen!) will ensure that.

The utility function will take into account time (as I mentioned above). When a transfer window shuts, a club's transfer options naturally shrink drastically, so a form of "negotiation risk" or guessing the amount of time to negotiate a deal would be enough. Indeed, an AI might proactively go out and start the discussions even before selling - just in case.

If there is no money to sign for, the replacement for "aging player X" would have to be either "free player Y" or "own youngster Z". Don't apply the worst possible meaning to "one-out-one-in".

The thing is, it's not necessarily one-for-one. For example, a player who is 5th choice but happy to collect his wages might not need replacing at all, as he's 5th choice. Or a player's departure might be replaced by two youngsters competing for that spot, so it's two-for-one. Or an amateur club that has no money nor youngsters to promote might just have none-for-one.

If you like, there is no need to replace the player, and they might not replace the player with a single one. The departure of a player doesn't necessarily require the AI to take an action.

[HUNT3R]

I think most people would be surprised by the complex nature & efficiency of the transfer AI in FM, it's certainly not a simple one-size fits all system as we have to create logical processes for a wide variety of clubs with different types of managers while operating in various leagues & nations which all have their own styles to take into account.

Reading some of your posts especially over the last few months, it does sound like the AI is a lot more complex than I ever gave it credit for.

I'd like to see better long-term decisions by the AI though. I see clubs with aging players (30+) with no sign of replacements being brought in and prepared for when the older players retire. Although, saying that, I have no idea how much more complex it can get before it starts pushing the minimum CPU requirements up or slows the game down too much

[x42bn6]

I think most people would be surprised by the complex nature & efficiency of the transfer AI in FM, it's certainly not a simple one-size fits all system as we have to create logical processes for a wide variety of clubs with different types of managers while operating in various leagues & nations which all have their own styles to take into account.

The way the AI thinks can be the same; however, each AI would have different goals and possible actions based on its circumstances. For example, in a league where only 25 players can be registered, the only difference between a club in this league and another in another league is that the utility of 26 players is a lot less in the former than the latter.

You don't have to create a variety of processes for the permutations you've stated above. Such a thing defeats the purpose of the editor. The AI has to be generic. It will be able to understand the restrictions of each club based on its competitions and tailor its goals, actions and scoring algorithm correspondingly.

I do agree that the transfer AI in FM could always be improved, the day it doesn't need improving is the day that we can all sit back & let technology do all our heavy lifting because Artificial Intelligence will always lack the flair required to provide a perfect simulation of the sometimes utterly bonkers world of football.

How many people would have said that Celtic's recent managerial appointment was unrealistic & would never happen or if I go back a few decades that Leeds United would never sell Eric Cantona to Man Utd & certainly not for that fee?

You can have imperfect AI too. In chess, for example, you can have one AI algorithm but a weaker AI simply looks ahead less, or chooses moves more at random. Similarly, you have one transfer AI algorithm where a weaker manager is a poorer judge of player ability, transfer values and/or negotiation and therefore will be less capable or "optimal" when it comes to transfers.

[Amarante]

The Chess AI is fairly simple, depending on the difficulty level it calculates all possible moves that can be done and the consequence of this move and then act accordingly.

For the Transfer Ai to be improved it starts with simple steps, Players that are being sold that would not leave the club any time soon. Clubs turning down bids for players even though they are in debt and they clearly can allow the player to leave. The Transfer market is not broken just needs to be fine tuned, For that to happen the AI have to understand how to build a squad

[scott MUFC]

the differance between Away and Home performances.

Home, sleek pasing defenders keep thier shape, players run at the opposition

Away, lots of dithering, no players running past men, poor passing, keepers getting butterfingers.

stuff that should happen randomly, I never see at Home.

if both teams are using the same tactics then the players should behave in the same way. nothing to do with nervousness etc, it just happens.

and on the transfer front I had a silly one on one of my saves, where we got to Febuary, and Sunderland get deducted points for going into adminstration, and transfer list some players.

now for me if they were in that position, why are they not transfer listing in January.

but defo the 3D engine, it is the best it has ever been, but a lot of stuff looks odd, like keepers holding the ball behind the line, keepers catching when surrounded by players, lots of tackles from behind, players falling over nearby tackles, 2 players run into each other and make little tackles and gets nowhere.

[cerud]

the differance between Away and Home performances.

Home, sleek pasing defenders keep thier shape, players run at the opposition

Away, lots of dithering, no players running past men, poor passing, keepers getting butterfingers.

stuff that should happen randomly, I never see at Home.

if both teams are using the same tactics then the players should behave in the same way. nothing to do with nervousness etc, it just happens.

and on the transfer front I had a silly one on one of my saves, where we got to Febuary, and Sunderland get deducted points for going into adminstration, and transfer list some players.

now for me if they were in that position, why are they not transfer listing in January.

but defo the 3D engine, it is the best it has ever been, but a lot of stuff looks odd, like keepers holding the ball behind the line, keepers catching when surrounded by players, lots of tackles from behind, players falling over nearby tackles, 2 players run into each other and make little tackles and gets nowhere.

Don't get me started on two players running into each other. Had one game where two of my own players ran into each other and both ended up with broken legs. Oddest injury event ever...

[Loki679]

I know we have PPA but does the game have PCA? Possibly the players CA could be hidden to the AI as PA is with PPA and PCA used to assess transfer viability and squad building requirements. This could be coded to take age more into account when the AI looks at its squad so that while a player may have a CA of 150 still at 30 years of age the AI sees it as 130 or something.

[Illeana]

The training system needs a revamp IMO.

More depth to it giving the player more choices for what needs training.

[shirajzl]

Transfer AI may be thoroughly complex, but it completely falls apart in instances when obscure clubs reject a sum of money that would finance their existence for the following 20 odd years.

I can accept the general ideal of "clubs refusing to sell", but in some situations it just doesn't hold water no matter how you spin it.

[scott MUFC]

I would just like to know the price, Every player has a Price, to me they are not reasonable, I try and buy players who sell for 20-30m IRL, lets say Shaw, Southampton want over 50M for him.

gets annoying when you want a player and the scout report shows, Sale Price -

you enquire not for sale.

Interestingly if you buy between windows the price seems to freefall.

and Russian Youth intake just gave 9 players... 8 are premiership star potential?

[noah08]

Modeling the multiplayer aspect of the game off of Out of the Park Baseball multiplayer

[DE-Z]

Training really needs to go back to how it use to be. There was nothing wrong with it back in FM12 and now I really struggle to get the most out of my team when Training use to be my 2nd most enjoyable aspect. I use to love coming up with programmes and mixing them up throughout the system to get the best of my team where as now one of my favourite aspects is none existent.

[holteender115466]

The GUI.

It has been horrid the last couple of years and this years has to be the worst.

[Fond Foat]

The AI squad building. Too many small clubs from small nations managed by humans become too successful too quickly. This is mainly due to the ability to build a better squad than the AI can. I also think a part of that is your reputation and your clubs reputation increases too quickly.

[skybluedude]

It would be hard to pick only one, but I would go for better feedback on tactics, and better description of roles, mentalities, philosophies, player insctructions etc... in game. Not like million times mentioned, to have to read 3rd party blogs in order to understand the engine.

Defiantly this!! 250 hours later and I am finally starting to understand this version, certainly can't play current FM's like the older games.

[jamez_b]

Tactical and Player Analysis - I'd like to see more news items with Journalists analysing high performing teams tactics or certain players. Also to be able to hire analysts to give you useful data and insight from the previous match and offer up suggestions of what is working and what isn't? If a certain player isn't performing be given insight into what could be affecting his performance.

It would however need to go hand in hand with more player management options. Currently if a player looks nervous or complacent during a match which then leads to a poor performance you only option is to drop or second guess what to say in the press conference to stamp out complacency or nerves.

We're all managers after all and we need to be able to manager our players as much as we can manage the way they play.