Far Cry 2 and Watch Dogs Legion game director Clint Hocking believes the conversation around AI has been spoiled in games by online discourse and bad-faith CEOs.
Alongside discussing some of the issues stealth games face with hyper-realistic visuals, Hocking discussed the current trend of AI in games on an upcoming episode of the FRVR Podcast.
LLMs and algorithmic-based generation has existed in video games for decades. While generative AI is its own can of worms, ethical AI in games has existed in the form of procedural generation, and many games use generative processes to fill open worlds and maps with grass and foliage assets instead of hand placing them.
Hocking’s own work with Watch Dogs: Legion saw characters and their backstories generated on the fly with the use of an internally-trained algorithm. Names, jobs, and character quirks are assigned to an agent on the street that you can then control. This method even resulted in a minor controversy where a character was spawned as a “paediatrician that “ended a personal relationship with a patient”. Not so dissimilar from the issues commercial LLMs face today on a much larger scale.
Watch Dogs: Legion’s five-year development cycle came just before the controversial explosion of generative AI, but the team was looking at ways of generating content on the fly. “During development, we did talk about is there a way to generate some of this stuff,” Hocking said. “Is there a way to generate some of the audio? And we speculated that there would be soon, but it was do we start trying to work on these sort of pie-in-the-sky ideas and hope we can get them to come together and then fail and now have play as anyone?”
While Legion does use algorithms to assist in its ‘Play as Anyone’ novel gameplay idea, those are all locally trained models. In the end, most of the work required brute force via “adequate project management and discipline”. However, even with the modern use of Generative AI, distanced from any controversy, simply generating content does not make a good game. It needs intent, it needs to have a purpose, and it needs to service the human-made product, not the other way around.
The game designer explains that using AI tools or implementing AI into games still requires careful thought and intent. “Obviously, if all you did was, “hey, I can walk up to anybody in London and talk to ChatGPT’, who gives a sh*t about that? I don’t. That’s not a game, right? That doesn’t make anything interesting.
“You still have the same problem of how would I dynamically generate an agent at runtime that represents this person in a way that’s meaningful and represents this person’s knowledge of the world. And then, you know, in the modern context if you were doing that, how do we do that without boiling a bathtub full of hot water every single time you ask a question to every single person in London? You could literally destroy the Earth with the amount of energy that one instance of the game could consume. We wouldn’t want that.”
Hocking explained that there are ethical ways of using artificial intelligence in games. While a more extreme example, Arc Raiders’ AI voices were locally trained on the voices of hired actors. Embark also paid the same actors to re-record AI lines as the game continues to evolve. In another example, Ubisoft’s Ghost Recon: Wildlands used a pathfinding algorithm to generate realistic road networks between villages.
One of the larger issues now with game development using complex algorithms is that the ongoing discourse against unethical AI—tools such as ChatGPT which is built off the back of millions and millions of pieces of stolen work—is that ethical AI uses may be caught in the crossfire.
“One of the difficulties today is that the discourse around AI is so inflamed,” Hocking said. “And it’s so inflamed, frankly, by people who don’t know anything about it on both sides of the argument. The internet is a hate machine and it makes it very difficult for people to have the interesting and important conversations.”
Hocking worries that a game that made use of AI in an ethical way to design mechanics that no one had ever seen before—such as how Watch Dogs: Legion turned every NPC into a playable character—then it may be harmed by the lack of a nuanced conversation. “It would suck if someone were to make… some really interesting, really cool game that used these technologies in a way that everybody agreed was ethical, but that a bunch of people who didn’t know what the f**k they were talking about decided to destroy that person and downvote their game on Steam and destroy put them out of business because of things they don’t understand”.
The rampant explosion of AI has muddied the waters significantly over the last few years, and the AI boon has largely dismantled the average person’s relationship with algorithm-based tools. There are ways of using LLMs ethically, but that requires time and patience, and
“Anytime there’s one of these booms that’s on the edge of an inflationary hype bubble, like NFTs and whatever was before that… There’s a bunch of very unethical tech-bro-wannabe CEO dudes who get a bunch of funding from somewhere and don’t know what they’re doing and make a giant mess of the thing and generate a lot of bad faith,” he continued.
“They’re just wandering around looking for what’s the next internet. What’s the next mobile boom. What’s the next NFT. So that if I can get $2 million from my rich uncle, I can turn it into $200 million and then I can have my yacht in my private island. These people are gross and that’s one side of the equation and they have a very loud voice. And it sucks. And it creates a lot of bad faith, a lot of bad blood, and it’s hard to navigate important conversations, never mind unimportant [ones].”
Unfortunately, AI can’t be put back in the bottle, and that digital Pandora’s Box has far more worrisome issues outside of gaming. For the future of gaming, there are ethical ways forward, and we’ve already seen some of them. The issue, then, is managing the bar of what AI is good and what AI is bad. For example, Nvidia’s DLSS 5 is bad, whereas DLSS image reconstruction and ray reconstruction is good. However, the conversation requires meaningful and difficult conversations, and hopefully some form of regulation.
