Last week, Ubisoft became the latest publisher to join the generative AI bandwagon with the goal of creating previously unimaginable experiences. At least that's the opinion. During a closed-door briefing attended by Game Developer, the company presented a research experiment called “Teammates” which uses generative AI technology—in this case, Google Gemini—to turn NPCs into “living companions” that can remember player information, follow contextual commands, offer task information, disseminate knowledge, adjust settings, or simply chat.
Following an in-depth briefing, a hands-on demo and a series of discussions, we had the opportunity to meet with key members of Ubisoft's generative AI team to learn more about how Teammates works, why Ubisoft relies on third-party LLMs like Gemini, and whether this technology can be used ethically and purposefully. We started the conversation by asking why Ubisoft decided to use technologies like Gemini and ChatGPT when the owners of Google and OpenAI were facing legal battles (thank you, Bloomberg) for how they trained and implemented their artificial intelligence tools.
There is a sense that large artificial intelligence companies have adopted an “ask for forgiveness, not permission” approach when training their models on information gleaned from the Internet, including the work of numerous creatives who are not exactly thrilled about the situation (via Reuters).
As Ubisoft itself admitted during our hands-on demo, linking to third-party LLMs can also cause technical issues. For example, the company suggested that we might experience some glitches during the session due to the Gemini 3 update that was released a few days before.
Ubisoft's director of generative AI gameplay, Xavier Manzanares, said concerns about being locked into one specific model prompted the company to create a robust middleware platform that allows it to deploy in the blink of an eye.
“It's a modular platform because we didn't want to be tied to one model or another. So essentially we wanted to be able to choose what to use based on our needs. Today it could be a model based on Open AI, it could be Claude or Gemini,” Manzanares said. “Tomorrow it could be something else. This may be due to the innovation of some really cool models that we want to trade for. Or it could simply be because we want to have it in-house and want to build the model ourselves. [we knew] when it comes to clues, it's just confusion.”
Manzanares said that each of the LLMs has a “different form of intelligence” and is unique in terms of how they handle performance, which is why Ubisoft feels it's vital to create an adaptable system. Ubisoft's director of data and artificial intelligence, Remi Laborie, added that the company is not betting on a model, but on a broader “creative vision.”
“New models will appear tomorrow. The most modern. Open source. They will open up opportunities, and we are going to determine whether smaller models can solve our problems,” adds Labor. “So we have to be pragmatic about where to start, but also make sure that this is our vision and that we can tailor selections and focus on smaller levels.”
“It's a very cool tool, but it can do bad things.”
But what about the aforementioned ethical and creative considerations? Ubisoft's narrative director Virginie Mosser, who worked on the characters present in the experiment with teammates, said she had lengthy discussions with the AI team about how to combine her writing brains with their technical expertise to “keep the creator at the center of the development team.”
In one case, she explained that this meant being upfront with voice actors about the nature of the work they were doing and rewarding them fairly. It also, she added, meant getting used to “a new way of writing.”
Mosser acknowledged she still has some concerns about using generative artificial intelligence, but said she wouldn't use the technology if she didn't think it was capable of evoking emotion. “I don’t want my work to be flat and cold,” said Mosser, who also said she had a lot of misconceptions about generative AI, stemming from people viewing it as a shortcut that could be used to quickly create content.
On the contrary, she believes that if creative people work together with AI teams, they will be able to create characters and stories that really resonate. “That’s the real goal for me because that’s how we can make a difference in the future,” she continued. “Gather a lot of creative people into AI teams and bring them together today to ask the question: 'What does the future look like for gamers?' We've barely scratched the surface and realize there's a lot of work to be done with the emotional and relational systems, but we're focused.
“To me, we still have the choice and the ability to say, 'We want to add more emotion here.' This is an iterative process. We work a lot on character and content, and sometimes it doesn't resonate.”
AI companions can be controlled using voice commands in Teammate/Image via Ubisoft.
Manzanares said the process reduces the risk of producing “regular slop.” Labor agrees and emphasizes the importance of feedback from internal teams and testers who will be open about when something needs improvement.
“We have internal tests where we can ask, 'What do you think of this character?' How do you describe this character? How did you feel playing this character? We make sure they're feeling something, otherwise it goes back to the drawing board and we get together with machine learning experts and storytelling teams and say, “Okay, what's wrong?” Is it a matter of training? Do we need a new process to ensure that we have more examples and more assessments to ensure that this is something we can measure or something we can’t.”
Labory said the AI team would like to have clear and obvious criteria for success, but admits that sometimes characters and gameplay elements either click or fail because of subtle feel. He added that it is this aspect of the process that brings the AI team together. “Through this interaction and working on something close to design and creativity, we discover something new in our work and in what we create,” he continued.
To that end, Manzanares says Ubisoft approaches all of its AI experiments by starting with “something important.”
“Having potential prototypes is great, but if we don't see the code behind it and don't know how it integrates into the potential project or our pipelines, it's not for us,” he explains.
The goal, he says, is not to simply create flashy videos, but to give Ubisoft's internal teams the tools to help them produce interesting video games.
“It’s a wild world now. There are many extremes here. [around generative AI]. Really, on our end, it's not just about creating experiences and middleware, but also trying to teach everyone what's going on behind the scenes,” he said.
“It's a very cool tool, but it can do bad things. Everyone should know this and we should be humble about this fact. If we hadn't done the work we did, it would have simply created generic content that wouldn't differentiate between experiences.
The game developer attended the Ubisoft press conference in Paris, with tickets and accommodation provided by the French publisher.
