Today, Warren Spector called upon John Carmack and Tim Sweeney, two of the industries major programming superstars, to make AI better. And the man does have a point about AI, although talk is cheap in his case. Poor AI is detrimental to the progress of realism – and it’s down to Uncanny Valley.
Warren Spector talks a good game and often makes good points.
As I said, today at Gamescom he called upon fellow developers and programming superstars John Carmack (of id Software) and Tim Sweeney (of Epic Games) to throw as much of their talents behind AI as rendering pretty graphics. To quote him;
“”I’ve been actively trying to shame some of my fellow developers, specifically John Carmack and Tim Sweeney. Can you imagine what games would look like if those two guys spent as much time working on non-combat AI as they do on rendering? Can you imagine what games we would have if John Carmack decided he wanted to create a believable character as opposed to a believable gun?”
To be fair, he does raise a good point in this.
Realism is a wonderful aim for videogames but we have this problem – and we call it the Uncanny Valley. The more real something looks and the less real it reacts, the more we notice how unrealistic it truly is and the more repulsed we are by it. This is a very real (*groan*) issue for the progression of video games in the current market – when you see the likes of The Last of Us and Beyond: Two Souls, these are fantastically realistic games aiming for immersion in a world and with characters that are believable, that we can relate to and feel for. However, without competent AI to immerse the enemies of the world in, these worlds can never truly capture the unpredictability of a real world encounter. As you become accustomed to the game, and its set AI pathing, you begin to find yourself able to predict and counter the most grand of enemies. At this point, the magic is often lost, because we’ve seen through the graphics and learned that underneath, the AI is the same old predictable if/when arguments repeated ad infinitum.
There are games that manage to use this to their advantage, of course – Dark Souls. I keep using Dark Souls as a reference of good game design and it is because it just is a good example of it. The game is challenging, often brutish, and the predictability allows a player to learn to overcome these obstacles and progress deeper into the game. Everything is set up as a hurdle to leap, sidestep or smash right into – we notice the AI is basic, but it’s because the AI is basic that it allows for some spectacular moments and a feeling of self-worth and pride in beating the odds. It doesn’t matter in Dark Souls case because it’s designed – by purpose or by accident – to take advantage of that. It’s not especially realistic – although it is pretty – but it’s a game designed around its limitations and takes them and makes them something of a strength.
By contrast, Assassin’s Creed 2 is a game designed around speed and action. You are supposed to try not to be found, or spotted. And yet, if you are, the game suddenly has a problem. It’s doing so much – the controls, the engine, the enemies, rendering – that it sort of seizes up when it comes to combat, and you find only one or two enemies at any given time coming in for an attack. This makes combat manageable, of course, but it isn’t realistic. It doesn’t feel natural. And the illusion of being a glass cannon assassin is immediately and irrevocably lost, and it becomes more of an action brawler than a real demonstration of parkour acrobatics. Sure, you can still play the game as it is intended, but the pretense and illusion is lost when you don’t feel like there is any danger in a confrontational fight.
And it’s in this that we must make a distinction – some games have basic AI because it works. But some games have basic AI routines because the game itself cannot manage so much effort.
The push for more graphical grunt is intrinsically linked to the need for new and advanced AI routines. We’re not talking about making a GLADoS here, but the industry has coasted along on the same base efforts for some time and it is starting to feel a bit samey and droll. Making new worlds for us to explore is fantastic, but all for naught if your intent is to populate them with mysterious creatures that don’t act very intelligently at all. The moment we see through the graphics and spot the same old AI concepts that we’ve become accustomed to over the years, the illusion of progress is dismissed from our minds. Sure, we may still enjoy the games – very much so – but you can’t get that magical giddy feeling back again of how new it all is when you suddenly feel that it isn’t very new at all.
Thing is, this is something that the industry should be working on and I assume IS working on. I can’t believe that this isn’t the sort of thing that people in these studios aren’t already aware of.
Uncanny Valley is an issue that gamers have dealt with for some time – even moviegoers have had the pleasure of seeing movies such as Final Fantasy: The Spirits Within, a movie that almost entirely encapsulated the Uncanny Valley effect. It wasn’t a bad film. It had great voice work, a really nice plot and good characters. But for all the good and right the movie did, it was still animated characters in a hyper-realistic style, and the longer some people sat in front of it, the more they noticed the design and animation and the more they felt uncomfortable. It’s a well-documented effect and at some stage, we all go through it. Moreso now in gaming, where worlds and design and characters are as realistic in their looks as they have ever been.
But if they don’t behave, breathe and have those little ticks and imperfections that define a human being, it’s hard to “relate” to them as a living being. They are a character in a narrative inside a video game, nothing more and nothing less, and all the voice acting and all the script writing and the hours spent designing and crafting this character are wasted as we don’t feel guilt for our mistakes – game over, start from last save point, let’s go again.
And it’s why Mr. Spector is absolutely right. However, as a games developer himself, self-deprecation or not, it’s something he should also be working on. AI doesn’t need to evolve to predicting out every move, but it does need to evolve beyond the traditional arguments that have always defined it. In a stealth game, you hide in shadows and an NPC comes so close that they SHOULD see you – but they don’t, because the AI is designed to ignore you in that shadowy area. Illusion broken. A fighting game where characters slice into each other with weapons and come out without a scratch on them. Illusion broken. An action game where only one or two enemies ever engage at any point, even when surrounded by a dozen or more of them. Illusion broken.
These are some of the traditional things that could be improved on. AI will always be harder to push into realism than graphics, because we’re great at recreating what we see and what we are. But we can’t create WHO we are – the emotion, empathy or deviousness that comes from experience and living. These things are organic traits and we don’t really have the capacity to push into such grand delusions. But fixing many of the AI problems could be as simple as looking at these small lists of things we see and find unrealistic – like hiding in shadows merely inches from an enemy – and making it more realistic. If the enemy has seen us hide in the shadow especially – the guy then not knowing where we’ve gone is especially strange.
As we push into more realistic graphics with more graphical grunt, the more we are destined and indeed, doomed to notice such failings at such a core level. The more real something is, and the less real it acts, the more we are aware that it isn’t real. And it defeats the very point of having more realistic graphics if, at the end of it all, we’re going to be more acutely aware than ever before that something isn’t real. Progress cannot be one-pronged anymore. AI does need some attention, if only to fix the things we already know are broken.
But it won’t be easy, and this is why it probably won’t happen very fast. Because it isn’t easy, and it’s easier to show off a pretty game than to say a game has great new AI. People like pretty, but they don’t often think about the AI until they are playing the game – at which point, the developers have essentially already got your money so f*** you very much goodbye. You’ve already been conned out of your cash. That realistic shooter isn’t realistic at all, but you’ve spent the money and played the game. Effectively, you’ve been had.
Better AI may be a necessary future. But it’s a future that many developers will be keen to avoid – it’s a lot of work for arguably very little gain right now.
But as graphics get better, one day we will reach that point where we are repulsed by how real/unreal something is. We will react, and it will be a big reaction. Then the industry will have no choice but to do something to rectify it. But they’re happy to wait until they get to that bridge first, no point preparing to cross something that isn’t in your line of sight at the moment, is it?
Until then, we wait. One day it will happen.
Just unlikely to be in the next few years.