As the games industry evolves, what does the future hold for AI?
As game development undergoes what must be its most tumultuous period of change yet, AI is one of the disciples having to be the most agile.
There’s new approaches to pipelines to consider, the diversification of studio models and integration with parallel disciplines to adapt to, and a wealth of new digital platforms to support.
Despite the scale of the challenges ahead, speak to almost any of the AI space’s leading specialists, and you’re greeted with optimism.
“New hardware is always a fun technical challenge,” says Havok’s AI team lead Chris Elion, who typifies the buoyant energy that defines today’s AI sector.
“In gaming, there’s a constant push to make things smaller and faster, and the shift towards mobile platforms highlights this even further.
"The majority of our code is the same across all platforms, but it’s that small section of custom math or multithreading code that can make a big difference in the final performance.”
That diversification of platforms isn’t just keeping Havok busy. As many players continue to gravitate to ever more elaborate triple-A experiences on console, the on-going rise of casual, social and mobile continues unabated.
“Those two sides of the spectrum require different approaches to AI,” states xaitment’s MD, COO and co-founder Dr Andreas Gerber.
“Right now, casual gaming experiences utilise very basic pathfinding functions. Triple-A titles use some combination of pathfinding and movement, as well as some high-level AI that is often coded by hand.
“That custom code is being written on a title-by-title basis, and can’t necessarily be re-used for additional titles.”
Modular AI solutions have become increasingly popular, partly as they simplify the creation of low-to-high level AI, and ease the re-use of AI across titles.
“With our different modules, we’re well poised to offer AI for every level of game without forcing developers into a one-size-fits-all package,” offers Gerber.
From a purely business perspective, gaming’s increased scope does offer AI technology providers many new business opportunities as they explore more accessible pricing models.
This is a fact which is perhaps best demonstrated by Havok’s new Strike program, which offers a collection of licensing options targeted at a wide range of game developers, from small indie teams with one-to-two people, to larger and more established teams who are in a prototyping phase ahead of securing finance.
Looking forward a little further, all kinds of other opportunities are presenting themselves to pushing the abilities and functions of AI.
“The first big move will be studios standardising on a core set of low-level services for AI,” suggests Elion.
“These services will be things like high-speed nav mesh generation, dynamic avoidance, character steering, etcetera.
These services are flexible enough to accommodate a range of genres like classic first or third-person shooters, MMO, RTS, racing, flight and space simulators and so on, and will operate in all possible environments including fully dynamic or massive open worlds.”
“The other exciting thing is AI’s relationship with animation,” adds Eric Plante, character animation product manager for Autodesk’s games technology group, which works closely with the company’s Kynapse AI technology team.
“To start with high level AI, today you can get a character mesh on an asset store, you can get animations on an asset store, but getting a behaviour on top of that, and a behaviour that’s more easily tweakable is something that I don’t see today.
"That’s important for the hobbyists, pro-sumers, small studios and those doing prototyping.”
Elion is quick to echo Plante’s enthusiasm for the integration of animation and AI technology; arguably a trend that has been gathering pace for many years.
He also brings to light the fact that AI is now developing a closer relationship with other disciplines.
“Other areas that we are sure to see advances in are the interactions between AI and the physical environment, and AI and animation,” confirms the Havok man.
“Havok has made some great progress on mapping out the interaction between low-level AI and dynamic environments.
“Some of the results, which are available to developers today, are very cool. For instance we see a lot of emergent behaviour from characters when they have to not only steer around each other, but also in-and-out of destructible terrain.”
Arguably, AI is to be found anytime the game has to make a decision that will impact the overall experience.
That could be could be something as simple as choosing the next target for a single unit, up to controlling the whole game pacing by choosing when and where enemies are to be spawned.
An example provided by Kynapse senior engineer Mustapha Bismi is that a developer can use AI to interactively adapt the game difficulty based on how the player is performing.
In that context AI can be used at a reactive, tactical level or strategic level.
“Therefore, if one sees a game as a series of interesting choices, made by the player based on what he understands of the game, AI is then the ultimate feedback tool at our disposal to shape the experience,” explains Bismi.
“It can be used to improve the cosmetic side of things, making the entities behave better, thus improving the game immersion.”
Artificial intelligence is, many believe, actually at its best when used in this way; to make the game experience more enjoyable, less predictable, and more interesting to the player. In this context it offers the game designer the ability to change the game on the fly to respond to the player feedback.
Other predictions about the way AI will evolve in the next to five or ten years focus on the way character and narrative in games will change.
Take for example Gerber’s vision: “On the development side, I think we’ll be seeing more studios focusing on creating more intelligent characters that add even more to a game’s storytelling potential.
"For example, characters that can learn from their surroundings and the player’s actions, that can adapt to the current situation and are proactive instead of just reactive."
Meanwhile, on the tools front there is a consensus that a greater push towards ease-of-use, reliability and re-usability is essential.
“Those things make it easier and more cost-effective to create AI at all levels,” says Gerber.
“Features like fast, stable cross platform libraries and easy-to-understand graphical interfaces will really help simplify the creative AI process.”
All this optimism is encouraging, but it would be unrealistic not to recognise the challenges facing AI. For as game environments become increasingly dynamic, even at the most basic level of locomotion and pathfinding, the problems are far from being resolved.
“This is one of the greatest challenges we face,” claims Autodesk’s Bismi.
“If finding a technically valid path in a dynamic environment is already quite hard a problem in itself, it is hardly enough to have a good AI.
"We now require AI that is more sentient about the environment, that chooses paths not only based on how little movement they have to perform to get to the objective, but also based on their understanding of what is going on, avoiding risky areas such as open fields in an FPS.”
What’s more, as CPU and memory budgets for AI related tasks increase, the industry is now seeing more and more AI directing the gameplay experience in itself.
“This leads to new ways of designing AIs and new tools to control, debug, and improve them. We now need AIs able not only to reason about the whole game in order to plan things; they also need to have some kind of memory, modelling past and present behaviours,” concludes Bismi.
None of these challenges are insurmountable, however, and AI remains one of the most captivating spaces in game development.
Out of the box solutions are becoming increasingly flexible, affordable and powerful, and there’s a degree of democratisation of the technology unfolding that is set to benefit studios of every size and consumers with every kind of taste.
The AI of tomorrow, it seems, is shaping up to be smarter than ever.