Tuesday, July 31, 2018

Simulation and AI

At DevGame, a Devstream on two different types of simulation, for process and for effect.
When you use simulation for process, you almost always have a situation where the results are not going to be realistic. The process is complicated and it is intrinsically inaccurate. It doesn't matter whether you're talking about an AI attempt to replicate human intelligence, whether you're talking about an attempt to replicate an infantry firefight, or whether you're dealing with something like a football or soccer game, in all of those cases you're dealing with multiple layers of abstraction, and every abstraction, every assigned variable is going to be different than the real world

Even if you build a very complicated model using very accurate statistics, the small errors, the small differences, are going to multiply so that by the time that you get to the end result, you're not going to end up with very realistic numbers.
Also on DevGame, a reader raises an interesting question about whether the transformation FPS games is related to r/K selection theory.

Labels: ,


Blogger The Kurgan July 31, 2018 5:15 AM  

Would a “cheat” not look at overall statistical outcomes of a process and then “sandbox” the process within those parameters?

Possibly with a tiny “leak” possibility from one sandbox to another for those one in a million or ten thousand or whatever lucky shots?

I ask in ignorance because I certainly don’t design games but it would seem to me a relatively obvious solution.

Possibly though this is thought of as “impure”?

Blogger VD July 31, 2018 6:03 AM  

Would a “cheat” not look at overall statistical outcomes of a process and then “sandbox” the process within those parameters?

That's exactly what they do. But it's hard to do that precisely because of all the interoperating parts. You can't just put a hard cap on the score to keep it realistic and ignore the number of shots on goal, time of possession, and many other statistics. EVERYTHING has to fit together. And therein lies the challenge.

Blogger Ken Prescott July 31, 2018 6:35 AM  

Trying to get every little detail "right" ends up with the Campaign for North Africa, and you still don't get it right anyway.

Blogger AnvilTiger July 31, 2018 6:57 AM  

Simulation works very well when the parameters are well defined. For example, you can create a simulation that play "Hunt the Wumpus". It works fine. The key for simulations is that the the boundaries e.g. rules are clear. In such situations simulations work very well.
This relates to business processes. If you can model it in BPMN you are good to go. 90% of the time in the real world simulations are very useful.

Blogger Azure Amaranthine July 31, 2018 7:09 AM  

It depends a lot on how complex the simulation must be, as well as how well the real-life scenario is understood. If the scenario is entirely fantasy, for instance, the programmer(s) or designer(s) can put in whatever values or mechanics they feel would make the simulation most entertaining/interesting with no regard (or even abhorrence of) the fact that the outcomes of reality-based simulations over finite timeframes are generally neither clean, nor pretty, and often not reliable either.

In addition, real life scenarios are effectively never fully understood, even in the cases of seemingly simple scenarios. There are always more statistical tails.

Blogger maniacprovost July 31, 2018 7:09 AM  

I immediately think of computational fluid dynamics and FEA. The smaller scale the simulation, the more precise the individual elements, the better. But the elements are just abstract mathematical models. They don't actually attempt to simulate atoms or some foolishness.

Blogger Harambe July 31, 2018 7:13 AM  

What type of game are we talking about, out of interest?

Blogger Azure Amaranthine July 31, 2018 7:15 AM  

A simulation of reality is always a tradeoff between simplicity and accuracy, and that's even before you have to deal with unknowns. Artifice intrinsically differs from nature.

Blogger John Raptis July 31, 2018 7:18 AM  

You want a K-selected game? Look at all the single-life high-investment games that have been coming out (Don't Starve, FTL, all the other "roguelikes"). Be optimistic, GenZ males aren't having fun with the "infinite lives, infinite participation achievements, click to win" model of game design that took over from about 2000-2012.

Blogger Azure Amaranthine July 31, 2018 7:21 AM  

Or any survival game with permadeath.

Blogger McChuck July 31, 2018 7:54 AM  

@3 Ken - At least it had the Italian Pasta Ration rule

Completely O/T - This morning, while discussing a political campaign in Virginia, WMAL (DC talk radio) discussed Chuck "love is real" Tingle as a premier writer of monster porn.

Blogger Johnny July 31, 2018 8:02 AM  

I tend to think in terms of statistical analysis rather than simulation. And that is always backward looking. While they don't think of it in terms of cheating, it always is because allowing that the data is correct, you have perfect knowledge of the past. But alas, the future may be driven by different forces.

Blogger wreckage July 31, 2018 8:13 AM  

I enjoyed the stream.

Blogger Ken Prescott July 31, 2018 8:33 AM  

@11 - "At least it had the Italian Pasta Ration rule"

And it turned out that rule was wrong, anyway (the rations included cans of tomato sauce, and they cooked the pasta in that, no extra water required). That was Berg being a sperg.

Blogger Encephalitis July 31, 2018 9:14 AM  

@4 - "90% of the time in the real world simulations are very useful."

Until they fail spectacularly, usually because of some non-modeled variable. 'Black Swans' and all that. And then you lose all your money, the levees break, war breaks out, but you can stride confidently along saying "It was a nine sigma event! None of the simulations ever predicted it!"

Blogger Volpack July 31, 2018 9:24 AM  

The map is not the ground.

Blogger The Kurgan July 31, 2018 9:29 AM  

That actually sounds like a very interesting problem to solve.
I think I could do the abstract math part even if I don’t have any coding ability.
I mean I’m almost sure I could tabulate a nested statistical analysis that would do a decent job.

Is there a sort of simple/basic problem you would suggest I try to model (perhaps one that is well explored by game designers, so I could verify my version after the fact, but it would all be new to me and my approach would most likely have nothing to do with theirs) to see if my perception of this is accurate or way off?

Blogger McChuck July 31, 2018 11:01 AM  

@14 - I know. That's what makes it hilarious, and helps to prove the rule.

Blogger Revelation Means Hope July 31, 2018 12:13 PM  

applicable to my real life job. Asked constantly to help create a workflow to help in the healthcare delivery. But the variables are so large that I can see why they've called me in to help.
Doctors may be smart, but they aren't always wise.
It is certainly an engaging and interesting challenge, each time.

Blogger Converting Dollars into smoke July 31, 2018 2:12 PM  

you're not going to end up with very realistic numbers. 42 ????????

Blogger John Bradley July 31, 2018 3:51 PM  

When you use simulation for process, you almost always have a situation where the results are not going to be realistic. The process is complicated and it is intrinsically inaccurate.

But we're all going to die when the oceans rise 10ft over the next 100 years, because if there's one thing SCIENCE! totally understands, it's the weather...

Blogger Silent Draco July 31, 2018 6:08 PM  

It calls for a world toolset like WarpIV to collapse the alternate process lines back together, and make a bounded set of outcomes (probable, likely, possible, maybe).

42 was always the answer. The question was correct. The world changed. Do the math, and pay attention to the base.

I used Adams' answer to introduce some systems engineering papers and presentations. Wbat's almost always missed: did you answer the question you began with? Did you refine or adjust elements? Were they accounted for? At this point near the end, there were always a number of engineers looking very uneasy.

Blogger Jon Mollison August 01, 2018 12:33 AM  

You want Black Swans, take a look at the Black Swan generator called Dwarf Fortress. I've had entire functioning complexes with 60+ dorfs survive goblin raids, wandering colosi, and all manner of threats crash and burn because some dopey kid wanted to build a masterwork bucket and threw a tantrum over the fact we had no jade for it.

Any thoughts on the gear work and simulation aspects of that monument to fiddly bits of minutiae, Vox?

Blogger Dirk Manly August 01, 2018 8:07 AM  

@23 Jon Mollison

For those of us who haven't played the game, you've alluded to something, but haven't quite given us enough detail to understand what you're referring to. Would you mind filling in the holes, so we know what you're talking about?

Blogger Jon Mollison August 01, 2018 6:52 PM  

Sure thing. The New Yorker has a good summation comparing it to SimCity - they call Dwarf Fortress its "gifted, maniacal, extremely worrisome younger brother."

Skip to the halfway mark. When you see the ascii art, that's Dwarf Fortress.

The relevance is that all of the dwarfs (or dorfs in DF slang) are miniature AIs that pursue their own needs and desires, and they interact in strange ways. A big part of the fun comes from looking at what happened, noticing trends, and building a narrative around the results. Kind of like when your last chit in a hex-and-counter wargame doesn't actually break, but goes on a tear and rips up your opponent, and you imagine it must be a man driven mad with grief at the death of his foes and rather than run away goes full Rambo on the enemy.

So yeah, that little weird story actually happened to me in the game once. Had a child dwarf struck by a burning passion to build a jade bucket - why a jade bucket? Luck of the draw - and when he couldn't he randomly started attacking other dwarfs, which lead to a brawl and the dwarf constables running over and killing the kid whose parents flipped out and attacked the constables which sparked a full on civil war. It was mass chaos, and I lost 12 hours of fortress design and building. It was hilarious, and I immediately started a new game, but it wasn't the sort of thing you'd ever plan - it just happened as a result of an insanely complex simulation grinding through its parameters.

Blogger Jon Mollison August 01, 2018 7:05 PM  

Also relevant:

Post a Comment

Rules of the blog
Please do not comment as "Anonymous". Comments by "Anonymous" will be spammed.

<< Home

Newer Posts Older Posts