Develop asks how AI is developing on new consoles and PC, and where it can go from here
As the leap in graphical capabilities becomes less noticeable on new consoles, developers are looking for other methods of standing out and dazzling players.
One of the biggest advancements touted in new generation consoles is in artificial intelligence. With the vastly increased amount of RAM and the touted capabilities of the cloud, AI is becoming a new turn-to for developers offering more immersive, believable experiences.
And it’s no surprise to see AI taking an increasingly central role in new games. The hardcore players are looking for ever-stauncher challenges – just take Dark Souls for instance – while story-driven games such as The Last of Us and BioShock Infinite require clever AI and pathfinding to generate an emotional connection and ensure players aren’t simply infuriated by the presence of another character. Gone should be the days where your AI-controlled companion steps in front of your gunfire, standing and smiling oblivious as enemies swoop in.
AI taking centre stage
Gilles Mazars, CTO of the Masa Group, which last year released its new AI middleware Masa Life, believes clever AI can be a game-selling feature this generation.
“Contrary to previous generations, the new generation of consoles are not really marketed on rendering, as there is not actually a big leap forward compared to the previous generation,” says Mazars.
“AI has been presented as a possible ‘winner’ for the additional power the PS4 and Xbox One bring to the table. We believe AI can benefit from this in certain areas such as environmental queries (raycasting, for example) that will be offloadable to GPUs and crowds of NPCs. However, more horsepower doesn’t mean better AI per se.
“The recent spotlight on AI due to this new generation of machines will encourage games developers to spend more time on character behaviour. We might see better characters because the focus has drifted away from rendering.”
Matthew Jack, CEO and lead architect of Moon Collider, a Scottish firm that is powering the AI behind space sim and crowdfunding sensation Star Citizen with its Kythera tech, agrees that new advances in hardware is helping artificial intelligence reach new levels, and believes the big enabling factor in particular is more and faster RAM.
“This helps us manage a larger number of AI characters and background creatures and to maintain more detailed information about each of them,” he says.
“It also allows us to store more information about the game world, so that AIs can reason about bigger worlds and respond to them more accurately.
“There is also potential to make use of more CPU cores, especially on PC. AI is very suitable for offloading onto other cores, so it’s easier for AI to take advantage of this than, say, graphics.”
However, Epic senior programmer Mieszko Zielinski says that, in his opinion, developers have already reached the maximum level of AI smartness in the first and third-person shooter genres, and further raising the bar could ruin the fun.
He does say though that there is potential in using the cloud to offload AI computation power and, thus, improve intelligence in other genres, such as strategy or racing (see ‘Invasion of the car snatchers’), but believes this will require adjustments to the way AI is handled by game code.
“Cloud AI computations are a perfect fit for calculation-heavy strategy games. I can also imagine harnessing all that additional computation power to simulate rich worlds to a high detail for RPGs,” he explains.
“There’s a lot that could be done with cloud computing provided one doesn’t mind the additional preparation and data handling burden that comes with it.”
Rival Theory CEO William Klein, whose firm offers free AI toolset Rain for creating AI-based characters in Unity, says it isn’t clear yet how useful such computations could be, but admits it does have potential.
“Offloading AI to an external server is important if the problem your AI is solving requires lots of computation or large data sets. It isn’t clear that those are bottlenecks for better AI in games,” he states.
“We see the cloud opportunity as having more to do with connectivity and distribution. For example, our Sentio product makes heavy use of Microsoft’s cloud for managing AI lifetimes across games.”
A next-gen leap
Given that many games already offer seemingly intelligent foes and companions, will any advances in AI be clearly noticeable to players? After all, unlike graphics, the changes will not jump out to players in an instant.
Jack says the biggest changes will be in how AI handles dynamic and physical worlds, allowing both enemies and player companions to handle changes to the environment created by the player, allowing for destructible environments, for example.
“When you blow a hole in a wall or push a boulder, AI systems have to either recalculate their map of that area to respond, or do some very expensive querying of their surroundings all the time,” states Jack. “Either way, this was usually too expensive for last-gen consoles, but now it’s a pretty obvious use of the new power.
“Another leap we’ve just begun to see is to high-quality companion characters. These have been a long time coming. But games like The Last of Us and BioShock Infinite are showing the way and they didn’t need
next-gen hardware to do it.”
Mazars believes the new generation of games consoles offer an opportunity to develop more engaging enemies that can be used by developers in more interesting and unique ways.
He explains that an AI designer’s job is not to create NPCs smarter than the player – similar to Zielinski’s earlier assertions about hitting a peak in AI – but to offer a more exciting way to engage with game characters and friends. Experiencing the seemingly serendipitous string of quests in Skyrim or the early Fable titles could be one indication.
“What you will notice is not harder-to-beat enemies but a more progressive level of difficulty throughout the game, with NPCs adapted to your learning curve,” says Mazars.
“This improvement of AI will be a significant step forward for the player, offering a more progressive gameplay without those typical demotivating stages in the games when starting a new level.
“For sure, with a new hardware performance boost, and the distributed computing architecture of online games, AI will be more powerful, and we will be able to run more NPCs simultaneously. This will undoubtedly open up new game opportunities. But, to make sense for the player, we think this trend has to be continued, and conducted simultaneously with a better integration of AI in the creative process and the huge contribution of game designers to AI development. And this change has already begun.”
Klein believes AI is becoming one of the most important aspects of any game, and says being better integrated into that creative process and becoming a central part of the emergent experience is the next step for AI.
“The biggest challenge for AI in games is adapting how we think of it and how we use it to drive user experience,” he says. “The traditional view is that AI fills in the gaps of our virtual worlds characters and creatures that we interact with through pretty simple mechanisms. Rival Theory’s perspective is that characters in games are at least as important as the games themselves, and that we should be doing more to enhance interaction and gameplay through them.”
One of the biggest new challenges facing the development of smart AI, says Jack, is the changing nature of how some titles are made.
Games on Steam Early Access, for example, will require at least some form of clever AI opponents and companions early on in alpha stage to keep players entertained and happy, even when on the understanding that the game is yet to be finished.
Developers are building increasingly bigger worlds too, with sprawling MMOs such as Star Wars: The Old Republic and the Cloud Imperium’s upcoming space sim Star Citizen, which features mile-long ships and spans dozens of solar systems.
“It’s always been fast-paced and iterative, but there’s a push towards involves players earlier and earlier in the process, shorter development cycles and responding to feedback quickly. At the same time, ambitions for game design are only growing,” he explains.
“Soon it won’t be acceptable for a designer to wait a year or two while AI programmers build all the systems to support their vision, before he can try out his ideas. But that’s generally how it’s done. AI programmers have few existing libraries or tools to draw upon, apart from what can be re-used from previous projects.
“That’s part of how we see Kythera – developers benefiting from the features developed for other games. AI programmers have to ‘stand on the shoulders of giants’ to keep up with design and build towards those grand visions.”
But despite the challenge, Jack sees these changes as an opportunity for developers to offer new experiences, and offers leading examples from indies and triple-A studios who are bringing the focus back to gameplay innovation, rather than simply trying to dazzle with stunning visuals.
“We’re seeing more games where innovative AI techniques are fundamental to the design, like The Last of Us and the Assassin’s Creed series,” he says. “It will be the interactive crowds, the playful wildlife, the companion characters that will wow future players.
“And after that, it’s taking a much more active involvement in the complex and ever evolving worlds of MMOs, such as Star Citizen. Making AI fully-fledged citizens of these worlds is going to be the next frontier.”