Ubisoft, a video game publisher known for developing successful open-world franchises such as Assassin’s Creed and Watch Dogs, has recently introduced an AI-powered tool called Ghostwriter that uses machine learning to “effectively generate first drafts of barks.”
“Barks” are the spoken phrases and other noises that non-player characters (NPCs) produce in response to specific events within a game. These events may include enemy combat dialogue, conversations between other NPCs, ambient background noise in crowded areas or casual greetings directed at players as they move through the world.
Typically, game developers have to manually write each bark, which can be time-consuming and laborious. However, this arduous task can be avoided with the use of Ghostwriter.
To use Ghostwriter, writers begin by defining a character and the type of dialogue they want to generate. The tool then offers a few variations, each with two options for writers to compare and choose from.
The missing context from this headline is that the tool was created in collaboration with writers and is just about creating more variations for human-written 'barks', the short, repeated lines given to NPCs.
— Ubisoft (@Ubisoft) March 22, 2023
The process involves pairwise comparison, where the preferred option is selected, and Ghostwriter learns from the decision. Over time, the tool becomes more effective and accurate as it processes thousands of selections made by writers.
Ubisoft explained that the objective behind Ghostwriter is to help game writers streamline their workload, freeing up time to concentrate on more significant aspects of game development.
In a video release, Ubisoft mentioned that the Ghostwriter tool was created in collaboration with narrative teams to enable them to accomplish repetitive tasks more efficiently, allowing “freedom to work on games’ narrative, characters, and cutscenes.”
Ben Swanson, the R&D scientist who developed Ghostwriter, said integrating the tool into game production was the most significant challenge. To address the problem, the team created Ernestine — a back-end tool that allows anyone to create new machine-learning models in Ghostwriter.
Swanson’s journey developing Ghostwriter
Per Ubisoft’s official blog, Swanson became interested in the creative applications of Natural Language Processing after taking a course on Digital Literature with two creative writers from Brown and Rhode Island School of Design during his Ph.D. in Computer Science at Brown University.
This experience introduced him to generating art with generative models, and he has since been exploring the potential of combining this technology with creative writing.
The scientist continued to pursue this interest while working at Stadia Games and Entertainment at Google in 2019 and later at Latitude at AIDungeon, where he conducted further research on machine learning and published a paper on the topic in 2021.
Swanson joined Ubisoft in 2021 after watching a Game Developers Conference (GDC) presentation by the Watch Dogs team at La Forge Montreal. La Forge is a program that fosters collaboration between academic research and video game production at Ubisoft. It enables university researchers to use Ubisoft’s resources and data to innovate and create projects like Ghostwriter.
The California native expressed great optimism for the future of Ghostwriter in video game production and the role it can play in the industry. By taking advantage of its user-friendly interface and powerful AI infrastructure, the dev teams who integrate Ghostwriter into their production process can enhance their games and develop more complex narrative designs while retaining full creative control over their work.
As a writer, having to edit AI-generated scripts/dialogue sounds far more time consuming than just writing my own temp lines 🤷🏼♀️. I would far prefer AAA studios use whatever budget it costs to make tools like this to instead hire more writers. https://t.co/VKYPeMHiwY
— Alanah Pearce 🔜 PAX East (Saturday only) (@Charalanahzard) March 22, 2023
Swanson recently presented Ghostwriter at this year’s GDC in a “Machine Learning Summit: Natural Language Generation for Games Writing” on March 21 in San Francisco.
During that talk, he recommended that developers who want to incorporate large language models consider using paid APIs from companies such as AI21labs or OpenAI. Swanson also suggested that tools like gradio, jupyter, Firebase and HTML5 could be used by developers to create workflows that can interpret and execute data.
Roblox, Unity testing the water on AI tools
Ubisoft is not the only company exploring the use of AI tools for game development. At the GDC, Roblox Corporation unveiled its AI tools, including Material Generator and Code Assist — currently in beta.
Game engine Unity has also been building new AI technology to provide game creators with AI-powered game development tools. The company has introduced a beta signup for its AI program and a newsletter to inform game developers about Unity’s AI-related updates.