Skip to main content

AI-generated NPCs represent the worst parts of video games

Mindless tasks and shallow characters? Yes, please! Said no one, ever.

Nvidia and Convai announced a big step forward in generative AI for video game development during Nvidia’s special address at CES 2024. That development is Convai’s AI platform, a tool that slots into Unity and Unreal and lets developers create NPCs with “revolutionary conversational, perceptual, and action generation capabilities.” Except it sounds more like a big waste of time and money instead of a revolution.

“With these new releases from Convai, users can now ask NPCs about their surroundings, with real time actions,” a Convai representative said in a press release. “For example, characters can perform complex actions such as fetching items, following or leading characters to various places, and even execute complex tasks all dynamically requested and generated during gameplay.

“This opens up endless possibilities for developers and gamers to create the next generation of games. NPCs interact with each other even when the player is idle, producing countless options for how these characters perceive and engage with their surroundings.”

“Cool” new tech like this is why actors and voice actors went on strike in 2023 and demanded protections against AI. Companies that plan on using this platform have yet to disclose where they’re getting the automatically generated voices from and how, if at all, they’ll compensate actors whose voices they use. Setting aside that massive ethical and legal problem of where developers who use Convai and Nvidia’s other tech are getting the voices for these interactions from, the implications are, frankly, insulting.

Convai and Nvidia are trying to sell us gourmet meals, but their exciting new tech is just pasting together recycled bits and pieces that’ve sat on the shelf for decades and dressing it up with fancy words, hoping we don’t realize what it actually is. At best, it’s an extension of the Battle Pass mentality. You see, the big draw, as Convai pitched it, is that generative AI gives you more stuff to do and more ways to do it. In other words, it’s just another way to keep you interacting with a game for longer – the same strategy that studios tried emulating with live-service games, throwing millions of dollars out the window before realizing that busywork isn’t fun and shutting their projects down.

AI-generated NPCs can have conversations with each other! Yeah, that’s been a thing for a long time now. Walk through the streets of Sector 7 in Final Fantasy 7 Remake, and the sounds of ambient conversations rush at you from every direction. Eavesdrop on NPCs in Baldur’s Gate 3, and you might just learn a juicy little secret. In these and other situations, the conversations are rooted in specific times, places, and cultures that have relevance to the broader narrative and your role in it. They don’t just exist for the sake of filling a void with noise.

The conversations in Convai’s cyberpunk ramen bar demo are unique every time, and I’m struggling to imagine a narrative where that’s useful or even desirable. The demo video shows the bartender and a cybersecurity expert talking about whether they’re AI constructs – AI must’ve scraped that one from a middle schooler’s C-grade project – and the security expert refutes the idea, agrees with it, and has a neutral reaction to it in different scenarios.

Hope, a Baldur's Gate 3 dwarf, stands in the middle of Raphael's foyer wearing a quizzical expression

Imagine giving up unique scenarios like Hope's salvation in BG3 for some AI-generated guff pulled from a computer's hat. 

Maybe that’s exciting to you. To me, it just indicates a shallow character with no anchor tying them to their world or even their personality beyond that moment. The next time they speak, they’ll be a completely different person. It’s the equivalent of talking to someone who can’t commit to anything or who says something different depending on who they’re speaking with. That’s a frustrating and pointless exercise in real life, and I don’t fancy doing it in a video game.

The prospect of these so-called dynamic conversations where NPCs react to your spoken prompts doesn’t sound appealing either. I never dreamed of a scenario where I could ask Baldur’s Gate 3’s Lakrissa about the weather in the Druid’s Grove or speak with Balthazar about his tailor before he attacks the Nightsong. It’s a gimmick, and a poor one at that.

There’s a reason that human writers create specific scenarios and lines of conversation for their characters. Focusing on a handful of topics means they get treated with proper depth, and it cuts out uninteresting garbage that adds nothing to the experience. These practices are basic concepts, and I’m baffled how many AI proponents just don’t see that – or, more likely, they ignore it so the shareholders keep investing.

Convai also boasts that you can send NPCs to get items or receive random quests from them. Neat. Animal Crossing did that 20 years ago, and the quests then and in the games that evolved that idea further weren’t all that exciting or meaningful. There’s a reason we didn’t like the empty, MMO-style fetch quests in Final Fantasy 16, and that reason is because they’re boring.

An NPC quest in Animal Crossing is shown, where a pig asks a human to fetch an item for them.

No, I won't fetch your used handkerchief from the lazy squirrel, because I value my time.

These dynamic NPCs talk to each other and display emotions, which sounds innovative until you think of studios such as Nihon Falcom or RGG who craft entire substories around their NPCs that span multiple games. Even Insomniac got creative and let two voice actors ad lib an entire scenario about parenting in Spider-Man 2 that went viral. AI, with its practically non-existent grasp of emotions and human experience, could never do that.

The idea of endless possibilities is an illusion. You either get something vapid and worthless or a deeper story that needs a human writer or editor’s guiding hand that shapes it into something interesting. That second point isn’t just conjecture, either. The quality of the, presumably unedited, conversations that Convai showcased in its demo was atrocious.

At that point, you’re wasting time and resources on checking and revising the rough scenario – a process that would go much faster with a proper writer’s room – and you need a human to do the job anyway. You may as well just not use AI at all.

Hopefully, studios realize these limitations before overinvesting in it, laying off workers, and making life unnecessarily difficult for people who just want a stable living. I’m not optimistic on that point, though, not after watching certain sectors of the industry compete to see how quickly they could slice off their own limbs over the last two years.