The Creators of Baldur's Gate 3 Details Its Implementation of Machine Learning for Next Divinity

The team behind popular titles like Baldur's Gate 3 and Divinity: Original Sin just teased its next major project, sparking a wave of anticipation within the player base. However, follow-up remarks from the company's lead designer have added nuance to the conversation, touching on the developer's philosophy toward machine learning.

AI as a Creative Assistant, Not a Substitute

In a new statement, the studio's founder explained that the team is utilizing generative AI for certain preliminary tasks. These include developing presentation materials, generating initial concept art, and drafting placeholder copy.

Importantly, Vincke made clear that the shipping assets in the game will be crafted entirely by actual creatives. "Larian is writing all the content ourselves," he stated.

Larian is constantly growing our team of writers and are currently putting together dedicated writer rooms.

As visual development is being particularly called out — we currently have 23 artistic staff and have job openings for more talent.

All our efforts we do is additive and aimed at letting our team spend additional energy on actual creation.

Any AI system applied correctly is additive to a creative team workflow, not a replacement for their craft.

Addressing Concerns and Clarifying the Vision

The news of employing this technology initially generated concern among some the fanbase. In reply, Vincke provided additional clarification on public forums.

"At Larian, we employ AI tools to explore references, in the same way we use search engines and physical media," he explained. "In the conceptual brainstorming phase we use it as a basic framework for layout which we then replace with authentic artwork."

He continued, "We've hired creatives for their unique talent, not for their ability to replicate what a machine suggests."

Three Pillars of Practical Application

Vincke had earlier broken down the team's targeted approach to machine learning, grouping its use into key functions:

• Automation of Tedious Tasks: This encompasses polishing mocap data, voice editing, and technical processes like adapting animations for different models.
• Fast-Tracked Experimentation: Using systems to speedily create simple models of mechanics to experiment with concepts ahead of full implementation.
• Long-Term Aspirations: Exploring how machine learning could in the future facilitate new forms of reactivity, specifically in creating player-driven narratives in a detailed game universe.

He clearly stated that core creative areas — such as music composition — are not areas where the team is reducing human input. On the contrary, Larian is expanding its staff in these exact roles.

"Our studio is neither releasing a game with machine-made assets, nor looking at reducing teams to replace them with artificial intelligence," Vincke concluded.