Rebellion, the team behind Sniper Elite and last year's hidden gem, Atomfall, has no expectations "to ever use gen AI on the screen," according to CEO Jason Kingsley in a conversation with GameSpot. "This isn't what we think we should be doing." Still, the CEO added that he feels having an online conversation on the topic has become "very difficult."
"The online discourse, I think, should be more nuanced. I think we should be able to have a better, more meaningful conversation, and discuss ideas. Because some people are totally against any form of AI at all. Other people are totally for every kind of AI possible. And I think the reality has to be somewhere in the middle."
Kingsley said that the idea of "using [gen AI] as part of the tools chain and the process, helping speed up iteration, it may have its practical uses [...] For example, back in the day, one of our big, successful levels in Sniper Elite was a viaduct with a big gun on the top of it. You could take a screenshot of that, and maybe a designer says, 'What would this look like in the snow? Would this play differently in the snow? Would it play differently at night?' We didn't do that, but you could. The designer could put it through a gen AI and say, 'Turn this from an Italian landscape to a snowy landscape, and let me have a look at it. And that means they can explore ideas cheaply and quickly to see whether it's worth exploring with human beings more."
The trick, he argued, is finding ways to use gen AI tools without them outright replacing humans. "I think we are in one of those weird times in history where nobody's quite sure how it should be used and how it shouldn't be used." He went on to outline how a gen AI tool and human laborers might co-exist.
"I think when it's used properly, I think it should act as a force multiplier for creativity, you know? So, for example, if you want to do collision boxes, you want to take a landscape, a huge landscape, like Atomfall, and you need collision boxes put around everything. A person can do that, but it's going to take weeks--and before machine learning and AI, a person did it. Now that person could make more objects in the game, or put more design components in the game, and let the machine do the collision boxes, for example.
"Is it right for an AI agent to be a player, and do QA, to a certain extent? Instead of 50 people playing the game, can we get 50,000 agents playing the game really quickly as well, to find those edge cases and those bugs and fix them before shipping? I think maybe yes, but that wouldn't remove the QA jobs; it would just mean we could do more QA more cheaply. We'd still keep the people doing QA, because people are really good at messing things up. QA teams are brilliant. The best QA people are brilliant at finding edge cases and messing with your game, and they should be empowered, and they should be thanked. I mean, people don't often thank QA, but QA is hugely important when it's done well. So yeah, I think the jury is still out, and I don't think anybody's quite sure. I think it'd be very dangerous to say you should use AI this way, or you shouldn't use it that way. I think it depends. There's an ethical side to it. There's an unethical side to it, and I'm not quite sure where it will land."
Kingsley also acknowledged that the use case of pairing AI and human workers is a bit easier for Rebellion, as the company is privately owned and therefore not beholden to public shareholders. In workplaces more obsessed with fiduciary duties, that's where AI tools often threaten to outright replace human workers to improve the bottom line.
"The company is owned by me and my brother. We don't have any investors. We don't have any [venture capitalists] or anything like that. And all the games we do with our own money, and we put them out to the market through partners like Steam, Sony, Microsoft, and Epic. In some ways, I think partly because we haven't ever bothered to try to raise money or anything like that, we haven't needed to sort of do the whole dog-and-pony show about corporate success or otherwise."
Rebellion's Alien Deathstorm is tracking for a 2027 release. You can read about how the studio employs a few clever tricks to stand out in the crowded Steam store.
Source
"The online discourse, I think, should be more nuanced. I think we should be able to have a better, more meaningful conversation, and discuss ideas. Because some people are totally against any form of AI at all. Other people are totally for every kind of AI possible. And I think the reality has to be somewhere in the middle."
Kingsley said that the idea of "using [gen AI] as part of the tools chain and the process, helping speed up iteration, it may have its practical uses [...] For example, back in the day, one of our big, successful levels in Sniper Elite was a viaduct with a big gun on the top of it. You could take a screenshot of that, and maybe a designer says, 'What would this look like in the snow? Would this play differently in the snow? Would it play differently at night?' We didn't do that, but you could. The designer could put it through a gen AI and say, 'Turn this from an Italian landscape to a snowy landscape, and let me have a look at it. And that means they can explore ideas cheaply and quickly to see whether it's worth exploring with human beings more."
The trick, he argued, is finding ways to use gen AI tools without them outright replacing humans. "I think we are in one of those weird times in history where nobody's quite sure how it should be used and how it shouldn't be used." He went on to outline how a gen AI tool and human laborers might co-exist.
"I think when it's used properly, I think it should act as a force multiplier for creativity, you know? So, for example, if you want to do collision boxes, you want to take a landscape, a huge landscape, like Atomfall, and you need collision boxes put around everything. A person can do that, but it's going to take weeks--and before machine learning and AI, a person did it. Now that person could make more objects in the game, or put more design components in the game, and let the machine do the collision boxes, for example.
"Is it right for an AI agent to be a player, and do QA, to a certain extent? Instead of 50 people playing the game, can we get 50,000 agents playing the game really quickly as well, to find those edge cases and those bugs and fix them before shipping? I think maybe yes, but that wouldn't remove the QA jobs; it would just mean we could do more QA more cheaply. We'd still keep the people doing QA, because people are really good at messing things up. QA teams are brilliant. The best QA people are brilliant at finding edge cases and messing with your game, and they should be empowered, and they should be thanked. I mean, people don't often thank QA, but QA is hugely important when it's done well. So yeah, I think the jury is still out, and I don't think anybody's quite sure. I think it'd be very dangerous to say you should use AI this way, or you shouldn't use it that way. I think it depends. There's an ethical side to it. There's an unethical side to it, and I'm not quite sure where it will land."
Kingsley also acknowledged that the use case of pairing AI and human workers is a bit easier for Rebellion, as the company is privately owned and therefore not beholden to public shareholders. In workplaces more obsessed with fiduciary duties, that's where AI tools often threaten to outright replace human workers to improve the bottom line.
"The company is owned by me and my brother. We don't have any investors. We don't have any [venture capitalists] or anything like that. And all the games we do with our own money, and we put them out to the market through partners like Steam, Sony, Microsoft, and Epic. In some ways, I think partly because we haven't ever bothered to try to raise money or anything like that, we haven't needed to sort of do the whole dog-and-pony show about corporate success or otherwise."
Rebellion's Alien Deathstorm is tracking for a 2027 release. You can read about how the studio employs a few clever tricks to stand out in the crowded Steam store.
Source