AI-Driven NPC Conversations in MMORPGs Demo using World of Warcraft!



Read more about World of Warcraft ➜ https://worldofwarcraft.mgn.gg

Here is a short demo using state of the art AI from OpenAI’s GPT API and a leading realtime voice acting AI API using the gpt-3.5-turbo model generated dialogue text as input. Both the conversation’s text and the voice are generated in realtime using AI.

These technologies are combined in a custom World of Warcraft client I’ve written from scratch to demonstrate the viablity and power of applying AI to game development for creating an immersive open-ended dialogue and conversational system with NPCs in an MMORPG.

Nothing is scripted but there are a couple short few-second edits to remove the “dead silence” as GPT’s API is intended for streaming text and not simply just waiting for an entire paragraph of text. Waiting for that is required for turning it into voice though. The GPT API sometimes takes awhile to respond as well currently as it seems to be under high load. But really only a couple seconds are edited out for demo purposes.

As for what the AI “knows” other than general World of Warcraft knowledge the AI is fed some information about the targeted NPC, their name, their faction and whatever available gossip text they contain in-game to hint to their motivations and style of speech. Additonally, they’re told about their current location as well as the player’s class and race. They’re given explicit directives to roleplay and stay in character as well.

I’m working on a crossplatform MMORPG in Unity3D that I hope to ship sometime this year and I’ll likely integrate the work of this AI demo into it too! So stay tuned.

source

28 thoughts on “AI-Driven NPC Conversations in MMORPGs Demo using World of Warcraft!”

  1. Whipped up this demo over the past 3 days. If you want some technical information about how it's achieved check out the description! I'm super excited to see AI applied to MMORPGs in the near future for creating more immersive worlds. I'm working on an MMORPG myself and am excited to integrate stuff such as this into to the project.

    Reply
  2. This is amazing!!! I hope to see more of this!!
    It's the first time ever that I've seen something as immersive as this, only thought about it in dreams tho. Can't wait to see what dwells in the future with ai in gaming!!

    Reply
  3. Okay, when he was talking about The Crossroads it sounded like a wiki fan page read from a point of view of… well, anybody really. But when speaking about Runetotems, not only he referred to them as "We" but also added something about himself, so I'm sold. However, I wonder what would have happened if you didn't explicitly said "you" in the question, would he still refer to them as "we are" or replaced it with "they are".

    Reply
  4. Cool indeed, but the NPC is wrong about the crossroads being a place where travelers from all corners of Azeroth cross. Crossroads is in Kalimdor. Furthermore, it isn't in a region where Horde and Alliance are in constant conflict.

    Reply
  5. I always wanted to design something like this. Hopefully as things grow it'll become faster at responding. Eventually perhaps it could listen to your own voice through a mic, feed it to the character. In fact I think this sort of this is only inevitable.

    Reply
  6. Would it be possible to have an AI like that but trained on a much smaller data set, only the information about WoW ?
    If so, would the AI require much less resources and be much quicker ?
    Additionally, could you train different AI on different infos, as to reflect the state of knowledge of this particular npc ? This would give NPCs actual point of views. Could you even feed them information that actually are beliefs, which would reflect on the culture of this NPC ?
    I'm guessing you could also work on each NPC to give them some sort of personality traits ? This one is very talkative but they don't necessarily have to be.

    This is very exciting, and I can't imagine how much more exciting it must be to RPers haha

    Reply
  7. Seeing what enthusiasts can cobble together on their own with co-opted general purpose neural network models, I can only imagine what a fully-staffed development studio might be able to do with more finely-tuned dedicated systems. I'm surprised, and perhaps a bit dismayed, that we're not hearing about these sorts of projects from major studios already. But I think it's ultimately inevitable.

    Reply
  8. Could you speed this code up by waiting for a sentence to complete and then send it to be converted to voice? Unless you don't get the data streamed back from the API, but one large text packet?

    Reply
  9. I am convinced then games with built in AI are the future of gaming, especially MMO. Not just for dialogues. It could generate quests and content, even dungeons in moments, making procedural never boring worlds we never could have.

    Reply

Leave a Comment