Roblox is pulling a DLSS 5, offering yet another example of a tool mistaking itself for an artist

Is there a word that tech companies relish more than democratize? It sounds so good, doesn't it. Slam dunk marketing copy: Everybody loves democracy! (Billionaires, avert your eyes and whistle innocently). In our mind-numbing AI era the word is doing more heavy lifting than ever, serving as the go-to phrase to suggest that any creative act that once involved pesky barriers like curiosity, practice, or hard-earned expertise is now accessible to all.

That's the heart of a new pitch from Roblox today: Roblox Reality, the company says, is a new AI platform aimed at "Democratizing photorealistic, multiplayer gaming." The general idea seems to be that this is the "good" version of Google's "AI worlds" presentation, which spat out videos that vaguely looked like videogames but obviously would have no grounding in, you know, game design. Roblox is instead pairing "the photorealism of Video World Models" with the underlying Roblox game engine, so that any game designed in Roblox could theoretically replace its basic, blocky graphics with jaw-dropping, hyperreal GRAPHICS.

In other words: Roblox creators will be able to slap an AI slop filter onto their games, though perhaps tactically the blog post above never actually uses the phrase "AI" to explain the technology.

In other other words: Roblox is doing a DLSS 5.

As with Nvidia's disastrous debut of an upscaling algorithm that overwrote the work of game artists to look "better," Roblox is a toolmaker no longer content with merely supplying the canvas and the paintbrush. The move makes sense from a technologist's point of view: you're only going to sell so many more paintbrushes (or in this free-to-play ecosystem, attract so many new painters) by making the brush better or easier to wield or whatever. The way you hyper scale is by promising people your new Super Brush can do all the painting for them.

AI boosters love to frame this sort of thing as democratization because it makes it easy to label opposition as gatekeeping, as if the ability to create art was born from unevenly distributed talent rather than from passion and perseverance. Generative AI promises a shortcut to arrive at the same end result without needing to put in all that work, but it misses the point that the process is where style and meaning and creative growth all come from.

As game designer and critic Austin Walker elegantly put it late last year, "Most of us did not choose to write or draw or design so that we could put products on the shelf. We chose to do it because it’s a set of skills we enjoy exercising, that give us fulfillment. It’s like asking a pro pitcher to just let a machine pitch for them. Fuck off and give me the ball."

The creators of Roblox Reality have identified a real problem—that making triple-A, highly detailed games is prohibitively expensive and impractical for all but the richest companies today. But that's still a problem technologists can tackle without suggesting the computer should just make up the graphics itself. The whole endeavor is also predicated on the idea that realism makes games better, or more appealing, when style almost always wins out over sheer detail. I don't think the people working on Roblox Reality even understand the draw of their platform if they see "photorealism" as a real benefit to the kinds of games that attract millions of players to it.

Are more people going to play Grow a Garden if it looks less like a plasticy Lego playspace and more like someone typed the prompt "cute colorful tradlife springtime homey garden" into Sora and then let Roblox Reality layer that skin over the game's skeleton? Roblox's biggest hits have much more in common with the "friendslop" games breaking out on Steam every few months—the likes of Peak, Repo, and RV There Yet? that find expression and comedy in their aesthetic choices. Peak would not have exploded in popularity if its blobby climbers looked like Temu Nathan Drake, which is the best any generative AI system that's digested and puked back up vast amounts of game art could hope for.

And in the same way that "asset flip" has become a recognizable aesthetic for derivative 3D games built in Unreal or Unity, I think players will be able to sense the artifice inherent in any game built under the assumption that only the core game systems need to be hand-designed, with Roblox Reality's video rendering slapped on top. It promises "breathtaking realism, secondary motion, natural dynamic environments, and fluid physics," but each of those promises comes with baggage.

How do the "fluid physics" affect the player's perception of motion and momentum and how a character feels to control? If something in a game looks like a realistic physics object, will they be disappointed if they can't solve Half-Life 2-style puzzles with it? How well does "breathtaking realism" pair with the sound effects? Imagine Nathan Drake again, but he makes the 2D Mario sound every time he jumps.

Even if Roblox Reality works and somehow doesn't look like absolute ass—a real longshot, based on the evidence we've seen from DLSS 5 and Sora and other generative tools so far—the novelty of millions of Roblox games all suddenly chasing a commodified photoreal makeover will fade quickly. The noise of that visual overstimulation will ultimately just stand to highlight how much the games still made by human minds and hands will stand out—like Over the Hill, an off-road driving game focused on the "minimalist serenity" of truckin' around, that already has a million wishlists on Steam.

"'Democratize' offers a synecdoche for an optimism that tech’s social goals and financial imperatives are aligned," critic Lora Kelley wrote for Mother Jones in 2022, just before AI drove the term's usage to new heights. The truth, so far, does not bear that optimism out, given how hard tech companies have been pushing AI in places its value isn't yet evident.

Engineers will no doubt keep trying, but you can't solve for art, no matter how hard you try.