'We're running at a f**king wall, and we're gonna crash'—CD Projekt's lead quest

[viper37]

'We're running at a f**king wall, and we're gonna crash'—CD Projekt's lead quest designer on big budget RPGs

During an RPG designer roundtable organized by PC Gamer, Cyberpunk 2077 quest director Pawel Sasko and other veterans talk about the "staggering" complexity of triple-A games.

Recent PC Gamer editorial "The cinematic BioWare-style RPG is dead, it just doesn't know it yet" caused quite "a commotion" between the lead designers at developer CD Projekt Red, Cyberpunk 2077 quest director Paweł Sasko said in a recent PC Gamer roundtable interview (opens in new tab). "Everyone actually, after reading this article, said: we mostly agree, actually, with the thesis. At least when it comes to triple-A, we are just running at a fucking wall, I think, and we're gonna crash on that wall really soon."
Sasko's comments above kicked off a discussion on the technology behind today's big-budget games and the expectations players have for them. The "wall" Sasko referred to is the ballooning complexity and expense of making games like Cyberpunk 2077. Before it released, I think it's safe to say many RPG players assumed that if CD Projekt had done such a fantastic job with The Witcher 3, it should be able to do the same with Cyberpunk.

But Cyberpunk's more immersive first-person presentation ended up making it a vastly more difficult game to build in ways that aren't obvious to most players.
The triple-A crisis

"Witcher 3 has so many fucking tricks," Sasko said, explaining one in particular—the way it would often cut to black to stage scenes or transition between bits of a quest, letting the developers spawn or despawn objects, and change the weather or time of day. "Sometimes there's a scene of a guy behind a bar, and he's like, submerged waist-up to the terrain because we didn't have animation. So he's just sitting there. But he looks perfectly fine in that scene, and it looks like he actually matches and everything works.

"Then you look at Cyberpunk. No cuts, no black screens, you're 'in' V all the time. Staging is in-person. It got so incredibly more expensive to generate branches. Adding branches to Witcher 3 was so easy in comparison to Cyberpunk. This article really sparked that discussion."

CD Projekt's designers felt strongly that the "no-cuts" first-person perspective was important for the game, but Pawel said they need to find a solution for "scalability of narrative: you want your story to be long, but also be broad, so we try to provide all the branches and choices and consequences." Doing that with their current tools requires an enormous budget. Disco Elysium, he pointed to as a contrast, was able to add narrative branches incredibly cheaply, thanks to the text-heavy, top-down presentation.

Former Dragon Age creative director Mike Laidlaw said the primary challenge, to him, is controlling player expectations. "As soon as you're delivering something that starts to be cinematic, you then are essentially inviting comparison to the most cinematic things. So you are kind of keeping pace with Naughty Dog or Cyberpunk," he said. Laidlaw witnessed the rise of cinematic RPGs firsthand at BioWare, and said that while working on games like Jade Empire and Star Wars: Knights of the Old Republic, they had simple tools for quickly creating most conversations. 
"We would drag a stage into a level and that essentially had lights and cameras and a set of animations and stuff. That became a tool that we could use to put a symbol to a conversation. We would call those 'bronze' conversations, but a 'gold' conversation would be, we'd put some art on it. You get someone to do slow zooms, rotates, and Dutch angles and everything. And you had to budget those, so we would very consciously be looking at our scenes and going 'well how many golds can we afford?'"

"I do think what you guys tackled with Cyberpunk was so phenomenally aggressive and I'm terrified at what that must have done to all of you," Laidlaw said, bringing it back to Sasko's comments. "It was like the unbroken perspective of Half-Life 2 but with all the branching and stuff from all your titles. It was a phenomenal achievement, I think, but I imagine expensive in a deeply personal level on top of the money." 
[...]

[/quote]

[Syt]

Have yet to listen to the talk, but I was shocked that Josh Sawyer is now looking like Ned Flanders.



I used to watch his Twitch streams a bit, and I'm more used to him looking like this:



Or this:

[The Brain]

How soon will AI do the heavy lifting?

[Tamas]

How soon will AI do the heavy lifting?

I was told by a game dev friend of mine that ChatGPT is already a useful tool to assist with coding, they recon that eventually their job will become of giving high level instructions to and review the work of similar AIs. 

[Josquius]

The interesting stuff with AI I've seen is with regards to npcs.

[Jacob]

How soon will AI do the heavy lifting?

Depends what you mean by "heavy lifting". Making big AAA games is a lot of work obviously, but the biggest challenges in my experience involve creative risk, production risk, and marketing risk and AI isn't immediately poised to make an impact there yet.

I was at GDC and saw some presentations on AI for contexts related to RPGs and videogames in general. There's definitely some strides being made, and I suppose it'll come eventually but I think we're 5-10 years out at the earliest for AAA type games - and it could be longer.

I wouldn't be surprised if we saw some AA or indie type games leveraging AI in simple cases in the next few years. So, say, some studio getting excited and kicking off production now leveraging ChatGPT and whatever relatively easy to stand up AI tools and using them and putting out some decent games in a single dev-cycle from now (so 2-3 years for indie/ AA). Whether they're successful or not remains to be seen.

But for games aspiring to the top of the line AAA it's going to take longer I think. We're at the stage where a number of individual game creation functions can probably be done reliably by AI in research environments. Tweaking that so it's actually useful in production environment of an actual game is going to take some doing (so "how do we get the AI to get us what we actually need" vs "this is the output we get from the AI and since we're going to use AI, I guess that's the output we're going to use" - basically AI as a useful production tool vs as a marketing bullet point).

Then there's the challenge of putting those useful AI outputs into the game-dev pipeline (which is bespoke for every AAA game, even the ones based on Unreal). This will be complicated if mulitple AI systems are going to be used (say writing, game animation, game ai, cutscene animation, texturing, level design and so on). A big part of what complicates top tier AAA games is that different parts of the game dev pipeline has very different needs and focal points and don't necessarily interface nicely with one another - and that's not something first generation AI is going to necessarily help address, even if the AI can create great things for individual functions. Then, of course, teams will need to learn whether using AI for any given function is actually going to help, or whether it's merely a gimmicky distraction.

So... I'd expect there are probably some AAA games now that are looking at using AI functionality for limited functions for dev cycles kicking off now... so we may see some of those projects come to fruition in 3 to 5 years (a standard dev cycle more or less). BUT that's assuming best case scenarios where the code base and production pipeline is ready to take those, as opposed to creating massive knock-on effects in legacy code bases (and most existing franchises are built on legacy code bases) undermining viability.

... which I think is interesting and important, but I wouldn't call that "the heavy lifting".

Maybe somewhere out there there's someone getting together a top tier team and connecting them with their top tier AI researchers and starting a fresh franchise right now, and maybe they'll strike true... in which case we'll see a game sooner. But given what I know of executive decisionmaking for game dev, I think it's likely too high risk for anyone to sign off on, even if AI is the current hotness.

This is all mostly just gut instinct but I think we're looking at:

~2 years to see Indie/AA games with interesting AI developed components.

~3-5 years to see AAA games that have talking points about how they used AI constructively in their dev cycle (and hear stories of projects that fucked themselves by chasing AI gimmicks and losing sight of the actual game dev).

~6-10 years to see AAA games where AI has been significant enough that there are clear scope/ quality/ cost differences.

[The Brain]

Thanks.

[Syt]

Thanks for the insight, Jake. I admittedly haven't followed the GDC much (though I occasionally watch some of the presentations from designers from GDCs past on YT), besides Yahtzee's videos on The Escapist.

[viper37]

How soon will AI do the heavy lifting?

Define "heavy"

Galactic Civilizations IV: Supernova is in pre-release.

It comes with an integrated ChatGPT module to build your story as you go.  It also builds race portraits and ship designs based on existing GalCiv lore.

It's a start to add variety in a game that lacks it.