07-05-11 - Huh-

I watched the 20 minute Carmack interview (thanks Nino) which was pretty damn excruciating, god dammit I want text transcripts, text text text, text is for information, video is for fucking TV news. Anyway, a few things struck me as strange.

One is that he talks about how important it was to get away from dark corridor shooters with monsters jumping out at you, and yet - you can watch the main gameplay trailer they've released and admire all the ... dark corridors with monsters jumping out at you.

But the main thing that made me go "huh?" is when he's talking about how they need to avoid redoing the entire engine for every game, and what they might save development time on in the future, he says something like "the AI, the animation, is basically good enough" (so they wouldn't be changed for future games).

Uh, what? Maybe if you want to make games where mindless monsters pop out on scripted paths and they animate around awkwardly and unnaturally, then yes, AI and animation are done, but in a more general sense, then no, they're not even remotely close. Time would be much better spent if they never rev'ed the graphics engine again and instead focused on AI and animation.

Valve is a good example; their graphics engine is actually pretty archaic now, but their animation system is very good, and their games look great because of it. Motion is hugely important; and of course valve's animation is still way behind where it should be (animation needs to become more code-driven, less canned, so that it can be more dynamic to weight transfer and surface and other world interactions, and also just more varied, more emotional). I think motion is maybe the most important step in the uncanny valley. If you have very natural motion even on a stick figure it looks startlingly real.

(Valve's characters still look like puppets that play one animation, then the next, sort of like the mechatronic Country Bear Jamboree kind of thing; they're very clever to use robots or toon shading to hide the uncanny valley).

Nobody is even close on AI. I don't expect game AI's that can talk to you or learn, but the goal should be very simple : playing a networked game against AI's should be just as fun as playing against humans. This is the "Turing test" for game AI if you like; if you shut off voice chat and play your Halo or Starcraft or whatever vs. an opponent and don't get told if it's AI or human, you shouldn't be able to tell. The AI should surprise you and experiment and sometimes make mistakes and make you laugh and impress you and do all those things that humans can do. Of course it should be able to, our sights are way too low.

Software developers often get stuck in the abstraction, they wind up comparing to their peers and forget about the absolute target.


nothings said...

Was there anything else interesting in it? I tried watching that a week or two ago and bailed right away.

cbloom said...

Not that I recall; it was mostly about philosophy and the compromises for 60 fps. A lot of saying how great the content team is. Not very much about the technology.

BTW so far I'm very unimpressed by the unique texturing. It certainly looks like the content guys had the inevitable problem that you just can't make that much unique texture, so of course they actually build the levels the same was as always - by splatting tiling textures and such.

ryg said...

"BTW so far I'm very unimpressed by the unique texturing."
I think it's a big step in the wrong direction.

Anything that amplifies the already totally out-of-control amount of data wrangling and content creation necessary to make a modern "high-budget" game by an order of magnitude isn't a solution, it's actively making the problem much worse. Especially since it ends up forcing your artists to do all texturing work in your own custom tools. Which, if ids previous tools are any indication, are probably seriously crappy.

cbloom said...

Yup, I mostly agree with that.

It seems to me that some kind of simple procedural texturing system makes much more sense.

My favorite is "texture by example" because it lets artists just work with the paint tools that they know and love, rather than working in some weird glitchy custom texture scripting system.

cbloom said...

Now, high-res "unique" precomputed lighting *does* make good sense, since that's just a matter of hey throw a bunch of CPU hours at it and make the lighting look really good.

Precomputed lighting still is the best solution for most games IMO, despite the sex appeal of dynamic global illumination.

Nino Mojo said...

I think it's a good thing that Carmack talks the way he does to interviewers and the gamer populace anyway.

cbloom said...

Yeah, for sure. Though I think he used to be even better about sharing technical details, and it feels a bit more like marketing now. To be fair, almost everyone has clammed up a bit since the old glory days when game developers actually talked to each other.

Miles said...

"I think motion is maybe the most important step in the uncanny valley."

I tend to agree with this and I just found this soft science result confirming the same thing:


old rants