PDA

View Full Version : Musings On AI-Run Teams In The MP Environment


Ben E Lou
04-08-2014, 08:01 AM
It has been interesting to watch teams put on full AI control in MP leagues. Those teams have been quite a bit more competitive than most probably would have guessed, especially in free agency. A few observations/explanations as to why I think this is:

The AI is willing to do what most humans aren't: build the best team possible--even if it means spending on older players--to try to improve its prospects from something like 6-10 to 9-7. Many human owners with sub-par rosters will avoid players with more than 8ish years of experience like the plague. (FWIW, I tend to suspect that strategy is less effective in this version...) Check out these signings in the GML as examples: "AVAILABLE" had a pretty good FA1... (http://www.fof-gml.com/forum/showthread.php?t=1350&p=5766&viewfull=1#post5766)
The AI never misses a free agency stage. This is an obvious one, but at times it's important. Humans get busy, forget that they didn't hit "export," etc. etc. etc. AI-run teams participate in every offseason stage.
The AI doesn't lowball, and humans forget that fact. This one tripped me up once before I adjusted, and I've noticed that some others haven't. An older player wanted something like $9M per year. I offered something like $7M per year, figuring "there's a good chance that no one else will pay that much, so I'll bet I can save a little cap space here." An AI team offered him a deal right around his asking price, and I was without the player.
The AI never gets "lazy." Instead, it always cuts the fat to maximize cap space. There was a team that went on AI control in IHOF this offseason. It had a good chunk of cap space before going on AI--maybe $35M or so. With that kind of space, some human players wouldn't have bothered to expend the time and energy to go through every single contract on the roster to see who was overpaid, and who could save money on renegs. However, the AI cut 9 players (all overpaid guys..all looked like "good" cuts), and renegotiated 10 contracts. And speaking of renegotiations...
The AI never forgets to do its contract extensions (as long as the team is on AI in FA1:1.) Almost every year in most leagues there is at least one human-run team that lets its players walk into FA because the GM simply forgot to extend contracts. Usually it's a non-playoff team where the human GM has partially checked out, just waiting to get to the offseason. He doesn't want to do his last-year extensions yet, though, just in case someone gets a debilitating injury. But as the last few weeks of the season wear on, he's less and less engaged, ends up sleeping through the postseason, and has multiple 60+ players in their prime whose contracts just expire.
The AI appears to evaluate every single player in the pool rather than overlooking some, as humans do. As thing as a couple of points above: the AI is never "lazy." It never says to itself "I've already looked at three defensive tackles rated over 60; I'm not bothering with those other three guys in the 52/52 to 58/58 range."

None of this is to suggest that it's better to have AI-run teams than human-run teams. But I have come over to the belief that in most cases a team on full AI control will be more competitive than a team run by a medicore-at-FOF human owner who isn't fully engaged. And I suspect that overall, AI-run teams in FOF7 are more competitive than maybe 20-40% of teams run by human GMs...at least for now as MP owners who don't play SP are still adjusting to the game. The AI is by no means perfect, but it's clearly leaps and bounds better than FOF2K7.

claystone
04-08-2014, 05:49 PM
interesting read.

I do agree that we (humans) do get side track every once in awhile, be it work related, home stuff or just plain lazy. I'm trying to change that with the way i have been running my teams with this new version of the game and the leagues i'm in. I being more active even with my busy work life.

I guess the "AI" is giving wake-up calls to everyone of how to make a better team and compete...;)

Sharkn20
04-09-2014, 04:13 PM
Gteat read!!

Wake up lazy duddes!! Sarah Connor's Chronicles are here!!

garion333
04-09-2014, 05:11 PM
Shark, you are insane. I think you're Skynet.

Sharkn20
04-09-2014, 07:05 PM
Shark, you are insane. I think you're Skynet.

World will end someday :cool:

mrtourette
04-10-2014, 02:57 AM
Good points, and helpful tips for us inferior humans to better ourselves.

Any observations on AI drafting? I've noticed many draft choices that sit somewhere between 'unrealistic' and 'daft', my pet hate being a team drafting a QB at the top of the 1st round two years in a row. Would never happen in real life but does in the game, and I was guessing because how the AI assesses the roster and best available draft choices doesn't take into account the potential real life fallout from drafting a top QB two years in a row (and that maybe we should learn to disassociate ourselves from such an emotional situation and just judge our roster on the numbers).

Kodos
04-10-2014, 09:03 AM
Phew. I was feeling bad for leaving Toronto ownerless. They're probably better off without me.

Ben E Lou
05-16-2014, 11:39 AM
The AI is willing to do what most humans aren't: build the best team possible--even if it means spending on older players--to try to improve its prospects from something like 6-10 to 9-7. Many human owners with sub-par rosters will avoid players with more than 8ish years of experience like the plague. (FWIW, I tend to suspect that strategy is less effective in this version...) Check out these signings in the GML as examples: "AVAILABLE" had a pretty good FA1... (http://www.fof-gml.com/forum/showthread.php?t=1350&p=5766&viewfull=1#post5766)Another interesting (at least to me) angle on this that I've noticed lately has been the scenario where a league has only one or two AI-run teams, so the older players are split among several AI teams. Over at the IHOF, one AI-run team loaded up on them this offseason and has improved from 9-7 last year to 8-2 so far this year.

gstelmack
05-16-2014, 12:34 PM
I still think you have to look at this post-TC though. A good example from WOOF is:

LT Bob Clements Player Details (http://www.fof-woof2.com/playercard.php?playerid=76)

13 year left tackle, we have to offer a 3 year deal at something like 4+ mil cap room (I tried to get him to take less, he won't even look at the offer), but the AI can offer a 1 year deal.

He's a great tackle now, but will he still be after TC? Most humans are looking longer term than just next season, and don't want to mortgage the future on these expensive veterans that might tank.

I got lucky in WOOF last year, I sent a big one-year deal to WR Jeff Eberly with 13 years of experience, and he tanked at Ex1 (not TC) and did nothing for me. Glad I only had the one year deal.

I do look much harder at 8-10 year vets than I did in FOF7, because it seems like guys retain their ratings a bit better after that point, but spending 3 year deals on old guys that might already be declining is something I'll only do when in "win now" mode, because it can KILL your cap the following season.

digamma
05-16-2014, 12:46 PM
Over at the IHOF, one AI-run team loaded up on them this offseason and has improved from 9-7 last year to 8-2 so far this year.

To be fair, this team had a pretty damn solid foundation.

lastcat3
05-21-2014, 06:38 PM
my pet hate being a team drafting a QB at the top of the 1st round two years in a row. Would never happen in real life but does in the game, and I was guessing because how the AI assesses the roster and best available draft choices doesn't take into account the potential real life fallout from drafting a top QB two years in a row (and that maybe we should learn to disassociate ourselves from such an emotional situation and just judge our roster on the numbers).

I think the Browns do that virtually every year.

Ben E Lou
05-23-2014, 03:33 PM
I still think you have to look at this post-TC though. A good example from WOOF is:

LT Bob Clements Player Details (http://www.fof-woof2.com/playercard.php?playerid=76)

13 year left tackle, we have to offer a 3 year deal at something like 4+ mil cap room (I tried to get him to take less, he won't even look at the offer), but the AI can offer a 1 year deal.

He's a great tackle now, but will he still be after TC? Most humans are looking longer term than just next season, and don't want to mortgage the future on these expensive veterans that might tankAnd as I said earlier, humans are wayyyyyyyyyyyyyyyyyyyyyyy over-cautious on this. Clements was a good example to check there. He was a whopping 69/69-->67/67 in TC. ;)

I just compiled numbers from four leagues. The results, while pretty compelling, will probably continue to be ignored by most, but oh well. *shurg*

PLAYERS EXAMINED: all players in year 10 and up who went into the preseason rated 50/50 or better.

SAMPLE SIZE: 630 players

AVERAGE CHANGE: -2.85

Let me say that one again for the contingent that will eventually read this post and not take appropriate action: on average, across four leagues and hundreds of players, the average year 10 and up veteran is dropping less than 3 points per season.

Removing punters, kickers, and QBs from the sample (because they have longer careers than any other positions and therefore would be expected to drop less), the numbers change, but not by a ton: the average change in the non-P/K/QB group is -3.42.

Some notable numbers from the entire sample:
475 players (75.4%) were -5, -4, -3, -2, -1, 0, or gained points. Putting it another way, there's roughly a 75% chance that your old starting-quality player is going to be pretty close the same player he was last year.
The older they get, it appears the *less* likely they are to have a big drop, even for non-QB/P/K. 52 of the 630 players were non-QB/P/K and in year 13 or greater. Their average drop? -2.69. Two point flipping 6-9. (Typically only stud players who are rare athletes make it that long, so my assumption is that the game is coded that rare athletes don't suddenly become ordinary apart from the ol' VSOD.)
Well over half of the players were in the -2 to +2 range. Yes, 382 (60.6%) of these guys were just -2, -1, 0, +1, or +2. So more often than not, these guys are basically unchanged.
Only 53 out of the 630 players (8.4%) dropped 10 points or more. Restating that one: there's less than a 10% chance of your veteran FA signee having a 10-point drop.
Seems like I hear a lot of "ZOMFG, HE'S GONNA GO -20 IN TC." What are the chances of that actually happening to a veteran starter-quality player? Well, in this group of 630 players across four leagues, it happened exactly 6 times. Less than 1% of the time.So, circling back to the original thought of the thread, and seeing AI-run teams in several leagues doing pretty doggone well by signing older FAs while humans stand by and watch, I think it's safe to say that a big reason for that is that JG programmed the AI to be aggressive in signing older, starting-quality players because it's pretty rare for older, starting-quality players to become tackling dummies overnight in FOF7.

garion333
05-23-2014, 04:10 PM
I hear what you're saying, but I can't read this and think that most RBs have already had their biggest drop by year 10.

That and I can't help feeling like we need to dive deeper into the numbers before we can make qualitative statements. We know positions age differently, but is a 70 rated player more likely to go -10 than a 40 rated player?

I took your advice about all this and have been approaching older players with less hesitancy, but I seem to be getting beaten out by backloaded contracts. That's a separate issue and perhaps not an issue at all, really.

Ben E Lou
05-23-2014, 04:27 PM
We know positions age differently, but is a 70 rated player more likely to go -10 than a 40 rated player?
I didn't pull the data for players rated below 50, but my guess would be that the lower rated guys are more likely to drop, or simply don't stay in the league long enough most of the time to make it to year 10 very often. I can pull those guys later this evening.

gstelmack
05-23-2014, 06:04 PM
If it really is 3 points per season, so 9 over the course of the contract, then yeah my small sample size has burned me.

Just double-checked Eberly, and he was a 12 point drop and retired for that one-year deal.

I'll adjust my thinking and see how it goes going forward.

Julio Riddols
05-23-2014, 07:21 PM
I got burned on a veteran signing in the CCFL, but that was clearly a VSOD. Turned around and did the same thing in the RNFL and that guy hung in there and is playing relatively well. I've felt recently like older players make for a more stable team in general, but that is purely anecdotal, as are most of my FOF observations.