Games: Soul Calibur 3

[Cataferal]

My first impressions from a days play:

Pros
+Graphics are exceptionally beautiful, matching GIII and MGS3's optimal standards.
+Music makes me tingle with joy: fun, well composed and produced with high production values.
+New characters are well designed and implemented.
+Character creation mode is decent, and should add lots of replay value.
+Core gameplay and style remains unchanged.

Cons
-New modes are average/tedious to play.
-Characters moves were unnecassarily changed for SCII, now they've gone further.
-Some cheesy voice-acting detracts from the epic build-up and atmosphere of a fight.
-AI that doesnt encourage the use of guard impacts or stun attacks; it now predicts your moves too easily, reducing single player modes to a hopeful button mash.
-No online play to actually show off your custom character.
-Core gameplay and style remains unchanged.

I'll be blunt: so far im not convinced this is the ultimate Soul Calibur package Namco hyped it up to be. Sure, theres a lot to unlock, the graphics and music are truly astounding, and the new characters were brilliantly designed, but for a game that clearly takes pride in its wealth of single player modes, they sure botched up that AI. Considering even on the lowest of difficulties it can predict your moves with systematic ease, i fear all Namco's efforts to inject originality back into the series have been undermined.

There is one mode, however, that redeems this technical banality, and thats the classic Versus mode: sadly the only social feature of the game. With this, there is at least some long-term justification in monotonously slashing through the single player modes, and needless to say, this is where the most fun is to be had. So let the rule of thumb be heard: if you have friends who share your enthusiasm for this series, your going to be spending hours on end with this game. If not, then this title will certainly give you a bang for your buck (or 50), but misleadingly, your not going to be getting much out of this game if you dwell on the single player modes alone.

<font color="#345E81" size="1">[ November 10, 2005 09:03 PM: Message edited by: Cataferal ]</font>

[mrwiseman]

The AI will always predict your moves because it knows them (after you've entered them, that is). In fact, it's the only sane way of predicting them; doing what a human does, i.e. observating how it's going and thinking what's most likely to happen would be insanely hard to program and insanely CPU intensive. A random factor is introduced so the AI "fails" prediction. More advanced AIs will intentionally delay their prediction. And most advanced AIs will start trying to predict what you're doing based on what you've entered so far, so the second you press forwards it selects all the moves that start with forwards and decides to initiate the move that will cover most of your possible moves multiplied by their damage and the frequency you've performed them in the past. Of course, if they did this (which is quite easy to implement), you'd be screwed, and random, less predictable button mashing using less effective moves would have more chances of winning than a serious player trying more complicated combinations the AI will always succeed to block. This is what the Mortal Kombat 2 AI does (difficulty mode adjusts the probability that it'll "fail" to predict your move). If you played that, you should have noticed how shitty and infuriating it gets.

It's often hard to tune this kind of thing, although being an essential feature in a fighting game, it's an important drawback.


The real thing to do would be to implement a hard AI (not as in "tough", but as in "hard artificial intelligence", Google it if interested). A neural network connected to a series of inputs: what character are you and the AI, where are you both, your life, the time left, your 3 previous full moves, and your current part-of-move. Then the AI tries anything at random (anything can be no action too), and values the outcome of the idea as (AI_life_new - AI_life_old) + (your_life_old - your_life_new). This way it'd learn like a human. At first, it'd be completely retarded. It'd require some training, so a pretrained easy mode should be bundled with the game. After 2000 combats, it could be playing better than the world's best Soul Calibur player, yet it'd never be as frustrating and repetitive as soft AIs. Of course, they should have to include some kind of AI reset, or preset Easy, Medium and Hard states, otherwise after you've played for a while you're done for.

<font color="#345E81" size="1">[ November 10, 2005 09:25 PM: Message edited by: -Wiseman- ]</font>

[Cataferal]

Not all AI has to work around the players input. A well designed opponent AI should be able to function independantly, and react in a variety of ways to the visual movements on screen only, just as a human player would. This would take a long time to design, and would probably be more exploitable, but it'd be more realistic. The alternative, as exampled here, is just too mechanic to enjoy playing against, and you just dont improve as a player.

[mrwiseman]

A well designed opponent AI should be able to function independantly, and react in a variety of ways to the visual movements on screen only, just as a human player would.

That'd be so processing intensive it's impossible to do in real time, even in the new generation, not to mention very hard to write. "Seeing" is one the hardest things you can ask for. What the human brain does for vision is amazing, and requires hundreds of millions of neurons to work together. It's hard to explain why you can't do that with a computer without taking dozens of pages here, as I'd have to explain how a computer works in the first place (and anything but a very detailed explanation wouldn't work), so I won't go into detail, but I'm afraid you won't be seeing that in the next generation, or two generations after the next. In fact, I'll be very surprised if that's possible before 2015 - and I'm probably grossly understimating the problem myself. In fact, it gets more complex as graphics evolve. Higher resolutions and anti-aliasing make it orders of magnitude more CPU intensive than it already is. With current the technology, and assuming somebody actually programmed such an amazing piece of software, it would take one minute to recognize where are you and what are you doing in a single frame (and again, this is most likely a gross understimation).

<font color="#345E81" size="1">[ November 10, 2005 09:46 PM: Message edited by: -Wiseman- ]</font>

[Cataferal]

Wow, i didnt know that, but what you said does make sense, and explains why it hasnt been done yet. What if we were to take a step back, and let the AI look at only the wire-frame of a model and their weapon, cutting out all the other visual elements? Perhaps im not appreciating the sheer complexity of this, but im sure there'd be some way to simplify this idea and make it work on some level or another.

[mrwiseman]

Looking at the wireframe would be exponentially simpler, as pattern matching would not be required. It would be even simpler to look at the vector data, which is 3D (a wireframe rendering is 2D and only motion tells what's the depth of every bit). It's still too much for the new generation, but if things keep going at this rate, the generation after the next would be fast enough in theory to do that. Then we'll have to find somebody willing to program that, don't look my way .

[phaq]

no, ps2, and i dont plan on getting one or a ps3 in the future. in fact im skipping the next generation(s) and going backward. besides all that, fighting aint my top genre anyway. if i want create a character, wrestling games have been doing that for a long time.

[Cataferal]

But wrestling games suck. They dont even have ninjas. How the fuck can a game be good without ninjas?

[Dreamer7000]

Wow, those are a lot of cons and pros up there. Since I don't own the game nor will be able to play it yet, I can only say that I'll give it a try surely. Tira's really catchy for me, and the fact that Talim is in there, specially if she's cuter makes me even more sure.
The character making ability was also a great idea imo. I love to create, and hell even play MY characters.
A few months maybe, and I can see myself and my pals inserting the disk to my PS2 excitedly.

[mrwiseman]

It's a great idea, but underexploited. They could've made body parts selectable, and if this messed animations up, they could've made them deformable from a basic "average Joe" mesh (multipoint, several ways, within ranges), which shouldn't cause problems. Skin color and detail maps could be chosen, and faces generated with something like Singular Inversions' software which kicks ass.