Ubisoft’s Paris Studio has showcased its early development build of generative AI-powered NPCs that can deliver real-time conversation with players without putting pre-determined answers within their answer database.
Dubbed as the “NEO NPC” project, it is powered by NVIDIA’s Audio2Face application delivering physics-accurate mouth and facial expression movements plus Inworld’s Large Language Model (LLM) for an authentic conversational experience.
Quoting Narrative Director Virginie Mosser’s words, she went from building a character’s back stories, lore, hopes, dreams, personalities, past experiences, and anything that is relevant and using those to write dialogues to them but now, instead of static text and phrases, the characteristics and “points” of NPCs are directly fed into the AI model.
From there, the LLM will understand its position and role and then respond with valid answers that adhere to the virtual world and landscape but this isn’t the “be all, end all” solution because total freedom is given to the player in terms of starting conversation, the model still needs to have guard rails to keep in check and not just blatantly respond with anything that is “out of topic or scope”.
The team also has to fine-tune the model to avoid bias and stereotyping in terms of the response it generates to make sure they are what they are and are not supposed to be.
As for when will Ubisoft integrate the idea into future games, they insist on crafting it to be as flexible as possible so that smaller budget titles to famous triple-A IPs can both benefit from the technology.
