Football Sims and Sliders

[albionmoonlight]

The more I that think about how hard it is to program a football sim with a robust AI, the more I wonder if I am not getting further from the idea of "the more customizable the better."

My point, which I hope to illustrate with a brief example, is that so many factors in a game relate to other factors, that it actually takes a lot more work than one realizes to make slidable settings work.

Say you have an injury slider (maybe one of the more innocuous sliders suggested). If you set injuries to off or to a really low value, you will know that quality backups are not nearly as important as quality starters. If you set it to really high, you know that you had better make sure that your #3 QB is decent because he will probably see some game time in a season. The question is--does the AI adjust for the injury sliders, too? If not, it becomes too easy for the gamer. If so, the whole dynamic of player evaluation (the main thing the AI does, IMO) changes. Considering that only 24 players of the 50+ on a roster are starters, the value associated with back-ups is a huge part of player evaluation. That value is directly related to injury frequency. The AI will need to tweak almost all that it does to accomidate sliders.

Also--changing injury rates (and, therefore, the rates that backups see the field) might have effects on how the game engine develops young players and how players are revealed to the gamer (and the AI teams) as boom/bust players. So the programmer may decide to change those values in relation to the injury slider (say--make playing time less important for backups when injury rates are really low). Of course, drafting players (another huge area in which the gamer goes head to head with the AI--and a part of the game that many people feel could be expanded (at least they do around April)) depends a lot on how players are developed. After simming a few seasons, humans will know how quickly they can plan to plug in a RB, OB, etc. into the starting lineup. One expects that the AI should know this, too. However, if this value is dynamic based on sliders, programming the AI to do this well becomes a lot harder.

Of course, as these values all change in the AI's mind, so too does trade value of players and picks. I won't go into all of the potential sliders and how they all can relate to each other (I think that people get my drift), but it seems that trying to program an AI to manage all of them would be like trying to solve one of those algebra II problems where you have four variables, but only three equations to combine to figure them out.

So, in theory, I agree with the slider crowd. In practice, however, if it comes down to making a universe that is static enough for the AI to stand on its own two feet and put up a good fight or a customizable universe in which the AI is not as strong across the board because so much energy had to be spent preparing it for all of the eventualities--I think that I will be happier in the first universe.

[Ben E Lou]

albion,

This has been my contention for QUITE some time and one of the primary reasons I don't want a high amount of customization. The more options, the more the AI needs to be able to adjust to those options.

Its the AI, stupid.

[Maple Leafs]

Sliders to allow for an already solid AI to be tweaked to the gamer's preferences = good

Sliders as a crutch to allow programmers to avoid having to create a realistic AI = bad

I don't want to have to fiddle endlessly with sliders to get the thing to work right. I want a default setting that's pretty good to start with... if you want to tack sliders onto that, go ahead.

[sabotai]

The problem with many game AIs is that their AI is just an elaborated form of "if..else if...else" statements.

It would be very possible to create AI to handle the slider values. It would also be, for a single developer doign the whole thing, very difficult because it woudl mean having to do other AI things than just "if...else if....else".

If the outcome of the slider movements are known, then it would be possible to use the "if" AI. Because then you could program them for that situation.

Like with the injury and backups you mentioned. It would be very possible to have the AI take into account the injury slider and the value of backups.

However, I feel this is an artificial resolution to the problem. It would also mena that you would have to be aware of every single possibility so that you could program your AI to handle every single possibility. Which is nearly an impossible thing to do.

As SkyDog says, It's the AI, stupid. And in order to have AI handle variable settings like that, the best solution is to create an AI that thinks and act dynamically. Which, for a solo programmer, would probably be a hard thing to do since they have to also design and program everything else.

It comes down to how much time and work a developer can put into the AI. Unfortunatly for us, chances are, a solo developer (or even a small team) probably couldn't put that out kind of time and effort it would take to create sunc an AI to be able not only to think and act dynamically, but to also plan and foresee the future.

We can sit here and look at settings and go "Well, changing this setting will have this effect." Once we create an AI that can do that same thing, then wou'll be able to cusomize the world. Until then, it'll be either no consideration for slider values or just more "if...else if...else" statements to handle the forseeable outcomes.

[Bonegavel]

Not to minimise the difficulty, but suppose you have 3 settings for the injury slider: High Normal Low.

Ok, you as the player know that if it is set to HIGH, you are going to have to make sure you have good backups. In the AI routine, you check for the slider setting:

if(injury==high){
backupPriority = 100;
}
else(injury==normal){
backupPriority = 60;
}
else
backupPriority = 30;

On HIGH, the AI knows that it better concentrate on backups. On normal it is 60% of a worry etc.

This may be oversimplified (first of all, not all AI players will have the same percentages- you will want some variation), but you must make sure that your sliders are understood by your AI. Don't just throw in a slider to affect numbers that the player will understand its impact but the AI is unaware. Make it aware.

[Bonegavel]

Sab and I posted at the same time with relatively the same answer (except he had a bit more philosophy).

I hate to keep bringing it up, but Galactic Civ has a great idea by recording the playing habits of players, and using the basic strategies as AI. You could possibly have infinite AI opponents with this approach.

[Coffee Warlord]

The problem is...the more options you have, the more bloated the AI code becomes, because you have to alter your logic for each and every bit of computer controlled options out there.

So instead of telling the AI to react this way...you have to tell it to react 10 different ways per item.

It adds up. It slows things down. A lot.

[Bonegavel]

for me, it comes down to this: I want an AI that is consistent with my perceived reality of what is being simmed (i.e., if I'm playing an american football sim, I want it to simulate numbers much like the nfl).

At the same time, I want it to make mistakes on occasions. I also want the AI to be unpredictable at times. I would like the AI to NOT remain the same over the course of time. If the game spans 100 years, it better not be doing all the same things in year 99 as it did in year 2 (of course, some fundamentals will remain unchanged).

E.g., the game of football has changed much sinced its inception. I also, don't think it is possible to predict where football is headed in the future. But, the AI could simulate past trends such as Run dominated offenses could show up for a few years, then sway back to Pass oriented, etc.

I guess, my bottom line is that my perfect sim is different from your perfect sim and the only way to give everyone their perfect sim is to allow customisation.