News

4

Merceron laments lack of progress in AI

Square Enix's technology director on why AI is one of game development's greatest challenges, and the six-year-old FEAR remains a high watermark.

Julien Merceron, worldwide technology director at Square Enix, believes that AI has been largely left behind in the current generation's charge towards realistic graphics and physics, telling us that it is such a complex discipline that it will be some time before we see significant advances in games.

"I think for many generations of platforms, people have been saying AI is the next thing," he tells us. "From my perspective, not a lot has happened. I don't think the leap will be as big as people believe, because it's a very complex area.

"It's not that AI's difficult to program," he explains. "It's that AI can screw up so many things in the game. It has dependencies on physics and animation, level design and so many aspects that all the changes you make in AI have repercussions on other aspects."

Part of the problem, he says, is that gamers infer deficiencies in AI from completely different things. "So many people expect so much from AI: to be entertained, to seeing behaviours that feel natural, to seeing good animation," he says. "People are going to say the AI is crap because the animation is crap. Or the navigation is badly generated.

"There are so many things that people attach to people related to AI that it's going to be very complex. It's an area that feels way more complex to address than, for example, taking a big leap in rendering, animation or physics."

Given that Merceron has one eye on the future - he expects AI to be allocated more resources as processing power increases - it seems curious that his personal high watermark for AI is a six-year-old console game. "My reference is always the AI in FEAR," he says. "It was so awesome not because it was the smartest AI, but because the NPCs were communicating so well about what they were feeling and doing.

"So, because you could read through this, you thought the AI was so smart. Not because it was truly smart, but because it was entertaining." It is tricks like these, he says, that we may have to rely on while the complexities are dealt with, and the processes streamlined. "It's about the runtime techniques," he says, "because even if we can make an NPC communicate with a player in realtime, we still need to make sure that the creation of an NPC like that will fit into the timeframe of the project.

"If it requires 50 guys over ten years, we know we can't do it. It's the realities of production and development, as well as having the right tools in place. Frankly, I don't think a lot of progress has been made in those areas so far."

Comments

4
jaks's picture

A technology director lamenting the lack of good AI in games. Funny because I've been lamenting the lack of good technology directors that won't spend any money on good AI programming.

NGTO1's picture

It's funny cause all I care about is the lamentation of the women.

Stuz359's picture

AI hasn't really been advanced in games in 15 years. Developers aren't willing to research it which leaves it to universities to advance AI research.

CaptainQuint's picture

I had to create an account just to point out that whilst the AI of FEAR was pretty damn good (and in many ways still is) it absolutely does NOT remain the benchmark, regardless of the bull this man would have us believe. I've played all the relevant shooters (seemingly and interestingly the genre where AI matters most) and by far the most impressive and convincing AI of the lot is found in the Halo series, culminating in the fiercely "intelligent" AI routines seen in Reach.

I wonder, for all their heft and fanfare; will the latest CoD and Battlefield iterations remember to include anything more than rudimentary AI behaviours this time around? I remain unconvinced. But the publishers of those games care little about things like believable AI - since they're very much aware of the fact that much of the mainstream who purchase those titles couldn't care less about AI either. An adrenaline bursting shooting gallery is apparently very popular at the moment, so why bother investing in resource heavy AI in the first place?

Revealingly, AI in mainstream games will only be vastly improved upon the day we see big name review outlets calling games out for their awful AI in their hugely influential reviews. And by that I mean actually knocking points off a review score for such unacceptable "givens" instead of quietly glossing over them...