The Creators of Baldur's Gate 3 Clarifies Its Application of Machine Learning for New Divinity Game

The studio behind popular titles like Baldur's Gate 3 and Divinity: Original Sin has recently teased its next major project, sparking immense anticipation within the player base. However, follow-up statements from the studio's figurehead have introduced nuance to the conversation, focusing on the team's stance toward machine learning.

AI as a Creative Assistant, Not a Substitute

In a recent message, the studio's founder outlined that the team is using machine learning for specific ancillary functions. These encompass enhancing presentation materials, producing early-stage concept art, and writing placeholder copy.

Notably, Vincke made clear that the shipping content in the game will be authored exclusively by actual creatives. "Larian is developing everything manually," he affirmed.

Our studio is constantly expanding our roster of concept artists and are currently assembling dedicated writer rooms.

Given that concept art is being explicitly called out — we right now have twenty-three visual developers and have roles to fill for more artists.

Each initiative we do is additive and focused on letting our team spend more time on the creative process.

Every ML tool used well is supplementary to a artist's workflow, not a substitute for their talent.

Responding to Feedback and Defining the Path

The admission of using AI initially generated backlash among portions of the community. In reply, Vincke provided more detail on social media.

"Our team utilizes AI tools to research ideas, just like we use search engines and art books," he wrote. "During the very early planning process we use it as a basic framework for layout which we then swap out with authentic concept art."

He added, "Our studio recruits creatives for their creative vision, not for their capacity to replicate what a AI generates."

Three Pillars of Practical Application

Vincke had in the past broken down the studio's practical strategy to machine learning, grouping its use into key functions:

• Handling Monotonous Jobs: Areas like motion capture cleaning, dialogue cleanup, and pipeline-specific tasks like adapting animations for different models.
• Rapid Prototyping ('White Boxing'): Using tools to speedily create basic versions of gameplay ideas to experiment with concepts prior to complete development.
• Future Potential for Gameplay: Investigating how machine learning could eventually enhance emergent player agency, specifically in managing dynamic reactions in a detailed game universe.

He explicitly stated that central narrative domains — such as music composition — are not departments where the team is cutting human involvement. In fact, Larian is recruiting more in these very positions.

"Our studio is neither shipping a game with any AI components, and we are certainly not looking at trimming down teams to substitute them with artificial intelligence," Vincke concluded.