At the recent 2024 Worldwide Game Developers Conference (GDC), Ubisoft unveiled "NEO NPCs", the first generative AI-powered gameplay prototype. It can change the way players interact with NPCs in the game, enabling unscripted dialogue, event memory, emotional understanding, and cooperative assistance.
We had the pleasure of interviewing Ubisoft's Vice President of Technical Production during the GDC event. He also had an interesting discussion on the content of NEO NPCs and Ubisoft's future research and development in generative AI. Here are the details of the interview:
Q: What do you think the new technology of NEO NPCs will bring to players?
A: What's exciting about this technology is the value it brings to players. The NEO NPCs we're showing today are just one aspect. It's a character that you can really interact with, strategize, and discuss, and you can also start to see some of the potential for native gameplay.
But the impact of this technology on gamers goes far beyond that. In other words, it's a smarter world for players, and they'll interact more deeply with the game world they're in. At the same time, narratives can become more adaptive and responsive.
It's a very broad question, about a vision that defines a new gaming experience. But that's where the excitement is, isn't it? At this year's GDC, we're going to be doing something with these technologies for players.
The core R&D team of NEO NPCs
Q: So, will this new technology be implemented in the new games we know about Ubisoft in the near future?
A: Another exciting part is that the set of technologies you see for NEO NPCs may be used by our future triple-A games, as well as by simpler games. What our team, and our department, does, is about really putting the technology together so that we can show our potential. We then had discussions with all brands and games to decide how they could leverage and integrate these technologies. It's very important to us that the creative team of each game can decide whether or not to adopt these technologies based on the brand IP and what they want to bring to the game.
Q: So, are there any specific games that are planned to use this new technology, such as the new Far Cry?
A: We don't have much more to announce at this point.
Q: We all know that sometimes a new technology can change the gaming industry. So is the use of this technology limited to NEO NPCs, or will there be more use cases or potential plans?
A: It's true that what we're very proud to share and show today is, of course, NEO NPCs, but we're also trying to apply this technology to different situations and scenarios. The creators will prove the technology, it's just that we're trying different ways to see and capture its potential and see its impact on the game, and it's definitely not limited to those NPCs.
Q: How do you think generative AI will impact the gaming industry?
A: The impact will be very large. I think any industry is being disrupted by AI today, right? We've heard about journalism, insurance, finance, and so on. So the video game industry is also going to be disrupted on a massive scale, and it's already being disrupted. Interestingly, because video games are not real-world, sometimes games can also be used as test scenarios for other industries.
I think it's going to change the way we make games, and it's going to change the player experience. When we're defining this change, it's very important that we have a very clear understanding of what the drivers are – what kind of experience is going to be for the player, what is it like for the creator, what is the experience for the creator, and the success of the change is going to come from the value it brings to the creator, and we're centering on those two very important focuses.
Q: How do you think this generative AI is influencing our thinking and direction? I'm not just talking about today's situation. We're talking about AI impacting all kinds of jobs, how do you see its impact on jobs?
A: As we mentioned earlier, first we need to refocus on this particular experience. I think it's very important to base the discussion on a specific case. Just like the team that is now delivering prototypes for games, there are data scientists, engineers, machine learning specialists, and you can see some creators.
There's no doubt that the new way of making games is how we assemble teams to make these things. And the real question behind this is how do we upskill our people so that they can be trained to master these technologies and get to the level that we need today. and how these technologies can help us improve the quality of our games and increase the value of our own games.
Q: In China, there are a lot of fans of Ubisoft games. Will your new technology make Ubisoft's games look more appealing in the Chinese market? For example, if Chinese players speak Chinese to NPCs, will they get a response?
A: This is a very interesting potential use case. What we need to check is what kind of quality it can achieve, and we also need to verify to make sure the answers are correct. We need to establish a process to ensure the quality of this type of interaction.
But I think it's interesting that you're precisely defining a whole new type of experience and the added value of the game, and that's really exciting. Maybe when you speak Chinese to NPCs during the experience, you'll get a surprise because they will answer and understand what you're saying.
So for your question, I think it's quite possible. These techniques should already be used for localization, and I think it will bring a whole new level of learning, and I invite you to try it out for yourself in the demo.
Q: Can you share a little bit about some of the other ways this technology has been used in older games, not just about NEO NPCs? I'd like to hear about what you guys discussed in the brainstorming session.
A: It's hard to give a complete vision. First, we need to deliver a prototype that shows the value it can bring to the player. It took us 12 months to explore and figure out how to use this scenario to show the gameplay and context of the game. So there's no one specific and simple answer to that, that's my first point.
The second point is that I would be surprised even though I have been in the industry for 20 years. Judging by the speed at which AI tech companies are evolving, even today, I have learned something unexpected from Open AI. It's been fantastic. So the vision is like a new question every year, every month, every two weeks: What's next?
Q: So this technology is just being developed for Ubisoft, or are you planning to share it with other developers? Is there a specific timeline?
A: We develop our own technology, which will also be applied to our own games and different brands. We don't have a timeline for the commercialization of this technology yet.
Q: In the near future, perhaps not so near, how do you see the benefits of this technology, and will it increase efficiency and reduce costs for companies?
A: Our goal is to surprise players with new experiences and improve quality.
Q: Is there anything you would like to say to Chinese players about this technology?
A: It's hard for us to keep track of the news from the gaming industry every week. I would say that our current focus is probably more on the US and European markets.
Q: When the game is released, players need to be connected to the internet to get responses from the AI provider, right? Does that mean players need to be forced to be online and have a more stable internet connection?
A: You may see this in the demo. First of all, the calculations are done on the NVIDIA platform, not on your personal computer. At the same time, large language models are being miniaturized at an astonishing rate. We're already doing research and work exploring various solutions for precomputing, local computing, and more, as the field is evolving at an astonishing rate.
The only question is, will it be in 6 months, or will it be 18 months, will it be localized, or will it be completely hybrid? These are the questions that our team will be working on in the next 2-3 years. Hybrid AI, proximity remote processing, or completely local processing, we can do this using different model pre-training methods.
So, all of these issues are the next steps for our team to face. Today, the demo showcased something interesting for the players. Can we use that to build a fun game? If we think it's fun for players, we can start working on it. Technical feasibility is coming, and it will definitely come because AI is evolving at such a rapid pace.