Should the AI learn from you/other AI teams?

I think there is a lot of us who need to try real hard to make the game more interesting for us, but I'm wondering that maybe the best way to make the game more difficult is to have an AI that learns.

AI should try to adapt the methods (tactics, what type of players to choose, transfer strategies, training) of a succesful team. Obviously, when talking of tactics, they shouldn't simply take your tactic one for one, but they might try to take the basic formation and try to make it work in the same way.

This would also force the human manager to adapt and change occasionaly.

Wouldn't that just mean that all the teams would end up using the same tactics?

(Almost certainly yours.)

Yeah, it would be nice if football could suffer metamorphosis in FM, problem is that is very hard to do it

"Should" is a pretty strong word there.

I think it would be great if there were some large-scale learning algorithms applied to the FM "AI". I think it is reasonable, having done some evolutionary AI programming myself once-upon-a-time, to suggest that some sort of advance in this area is to be hoped for in the near future.

From my limited experience I imagine the difficulty is actually the processing speed requirements rather than a programmatic hurdle. However, a yearly or quarterly "Time out to consider the data" could be both feasible with current speeds and not too huge an ask.

You will note, however, that most games, nearly all games really; Most games have no AI of this sort at all. What we call "AI" in games is entirely conditional and procedural. What you suggest is often demonstrated in even undergraduate computer science programming but hasn't found its way into many games as of yet. This perhaps suggests that games are possibly good enough already, for most people, without any intelligence whatsoever. And that shouldn't be overlooked. Just because companies can make AI adaptive or actually intelligent, perhaps there are business reasons why they do not.

If FM were "bot-able" in the sense of being able to apply user-made strategies to otherwise autonomous entities (AI teams) via scripting languages then you would likely already see robust and life-like bots designed by the community (and hopefully by themselves a bit too) able to deputize for a more universal AI as you are suggesting. However, again, FM's database is really really big. The data a bot might want to handle is potentially massive. But there is still room for some sort of extensible use of data if FM were to be redesigned in this direction. I find that an unlikely prospect, but a worthy wish. Barring user-driven autonomous agents I think FM's scope probably precludes the introduction of AI as you suggest. Any new processing speed will be used for higher priority things like match sophistication, depth, etc.., rather than additional calculation load in the form of learning or heuristics, unfortunately.

I would not say that the AI should be able to learn from anybody in the game.

It should however in general be better. That's true especially for the squad building AI. The wheeling and dealing in the transfer market is the area where the human users are just so much ahead that after a few years at the latest success should come fairly automatically.

"Should" is a pretty strong word there.

I think it would be great if there were some large-scale learning algorithms applied to the FM "AI". I think it is reasonable, having done some evolutionary AI programming myself once-upon-a-time, to suggest that some sort of advance in this area is to be hoped for in the near future.

From my limited experience I imagine the difficulty is actually the processing speed requirements rather than a programmatic hurdle. However, a yearly or quarterly "Time out to consider the data" could be both feasible with current speeds and not too huge an ask.

You will note, however, that most games, nearly all games really; Most games have no AI of this sort at all. What we call "AI" in games is entirely conditional and procedural. What you suggest is often demonstrated in even undergraduate computer science programming but hasn't found its way into many games as of yet. This perhaps suggests that games are possibly good enough already, for most people, without any intelligence whatsoever. And that shouldn't be overlooked. Just because companies can make AI adaptive or actually intelligent, perhaps there are business reasons why they do not.

If FM were "bot-able" in the sense of being able to apply user-made strategies to otherwise autonomous entities (AI teams) via scripting languages then you would likely already see robust and life-like bots designed by the community (and hopefully by themselves a bit too) able to deputize for a more universal AI as you are suggesting. However, again, FM's database is really really big. The data a bot might want to handle is potentially massive. But there is still room for some sort of extensible use of data if FM were to be redesigned in this direction. I find that an unlikely prospect, but a worthy wish. Barring user-driven autonomous agents I think FM's scope probably precludes the introduction of AI as you suggest. Any new processing speed will be used for higher priority things like match sophistication, depth, etc.., rather than additional calculation load in the form of learning or heuristics, unfortunately.

Thanks for a good, informative answer.

I feel that this would also make long-term saves for those users who are happy with the game more interesting too.

---

As for squad building, I agree. Hence why I think the AI should be able to pick up what kind of players are in a successful teams (in terms of attributes most of all).

I feel that if the AI is capable of picking up on what you do, this would make these crazy and unrealistic achievements like winning the league with maximum points etc much more unlikely. And this ability to learn (or maybe invent) should probably base on the attributes of the managers. At the moment I wonder what is the difference of a manager with all attributes between 5-15 and someone like Jose Mourinho.

Of course this could be a pipeline dream, but I'd like to think that this is possible in the future. Because I really wonder for how many more possible challenges I can set for myself in future FM's.

Although I wouldn't say the AI learns as such, the tactics do change to reflect successful teams in a league.

I'll use an example from my FM11 save - I was Lazio manager for around 10 seasons during which time I saw Inter initially struggle fighting for any European place before switching to a 442 with 2*DMs. This proved to a good counter to the narrow formations a lot of other teams used in the league and they had a lot of success over the next 4-5 seasons. Slowly over time other teams switched tactics, some copied Inter's formation whilst others looked for ways to counter it. When I left Lazio some teams were starting to find a way to counter Inter's formation by packing the midfield in a 451.