Ubisoft: AI is the "real battleground" for new consoles
Developers "extremely limited" by the current generation of consoles
Ubisoft's Yves Jacquier believes that improved graphics will not be a "strong feature" of the next generation of consoles.
Jacquier, executive director of production services at Ubisoft Montreal, has been working with students at Montreal University following the creation of the Industrial Chair on Learning Representations for Immersive Video Games earlier this year.
Ubisoft is investing $200,000 a year for the next five years with the aim of developing "other ways of thinking" about game production. Jacquier demonstrated the group's work in the field of procedural AI, telling GamesIndustry.biz that innovation in this area is vital to the future development of videogames.
"AI has always been the real battleground. The challenge is that, if you see an AI coming, you've failed. And that's a problem we have to overcome as we create the impression of flawless, seamless worlds."
What's the value of making something more realistic and better animated if you have poor AI?
Yves Jacquier, Ubisoft
In the past, consoles have been marketed principally on the strength of their graphical capabilities. The Wii broke this trend, and Jacquier believes that the next generation of consoles will all follow suit.
"In general the industry expects that graphics will not be a strong feature any more... Obviously, graphics are better for marketing purposes because you can show things. AI you can't show."
"Our challenge with the PlayStation 3 and Xbox [360] is that we're extremely limited in what we can do. It's a challenge for the engineers to provide nice graphics and nice AI and nice sound with a very small amount of memory and computation time."
"We think that the next generation of consoles won't have these limits any more. Games might have more realistic graphics and more on-screen, but what's the value of making something more realistic and better animated if you have poor AI?"
Jacquier's comments chime with those made by id Software's John Carmack in a recent interview with Eurogamer.
"The better games get the harder you have to go to give a delta people care about," Carmack said. "That's going to be a challenge for the next-generation of consoles, to show that the pack-in title is going to look more awesome than what you get on the current ones that people will want to go spend $300 on a new console."
I'm pleased companies are aiming for more realistic AI, as this will make games more immersive and fun to play, as well as making it appeal to more people.
One of my programmer acquaintances is very down on the state of AI and AI knowledge and skill in general. He also, quite correctly, notes that increasing AI capability results in an exponential requirement in processing power. It's not a problem like graphics were simple things can be parallelised and where cheap units can be crammed in together on future chip dies.....
Last time i had the conversation it was obvious that AI is currently akin to static lighting as opposed to, say, ray tracing. If we can bump it up to a complexity level of dynamic lighting then i suppose we'll have hit a milestone - but that's going to require a lot of work and also some figuring out of what's important to an AI. Soon we might have AI actors asking the "director" what their motivation is. :)
Many machine learning and local search optimisation techniques are far better suited to games. If significant time and effort is spent to develop the domain of the AI agent, and then for search/testing/improvement, it can lead to a final product that carries an action policy: it knows what action to take based upon circumstances. The benefit of these approaches is that the training/testing is done during development rather than during play, leading to a final product that is computationally very cheap to run (enemy players in F.E.A.R are a great example of this).
Dynamic as you suggest or online learning agents are now a big research area and games are receiving a lot of exposure. However they can still consume more CPU than is acceptable in some instances, and also require a lot of tailoring to ensure the AI does not behave outwith strict parameters you wish to define (e.g. gaps in logic or even bugs in code can be exploited by AI much better than humans).
So i suppose im saying, whats the point in having great Ai if you have poor animation that cant clearly convey state changes?
However I also feel Physics are also important. Such as objects having real weight, mass, volume and durability. Instead of a brick being a textured polygon box to look like a brick, it can have the actually properties of a brick, be broken into pieces and still have the inside parts and have the weight, durability of a real brick. Id also like to see real clothes on characters. Clothes that are actually resting over and reacting to a characters movements.
Instead of having predetermined destructable enviroments, having them act naturally, would allow players more choices when confronting a situation.
I know all this is just a fantasy, but not long ago, the things we can do now in gaming were nothing more than wishful thinking.