Research can be speculative or applied. Artificial Intelligence research is often both, trying to solve real-world problems while at the same time testing theories about how the human brain works.
Game Face. Photo By: Phillip Toledano
A branch of the AI research crowd are interested in games both as a testbed for theoretical work and as a market for applied AI. Unfortunately, these are conflicting goals.
People play games to be entertained, and any AI present in the game must contribute to this. I personally think that AI enhances player enjoyment when it is both surprising and relevant. That is, it should result in an experience which feels new, yet which is consistent in the current context.
This regrettably suggests that AI is synonymous with NPCs, which is a mistake that both game players and researchers make. There are plenty of opportunities for non-NPC AI in games, and yet there is scant research being done in these areas. I’m referring to things such as
- a cinematic camera that responds realistically to game world events and player movement;
- dynamic set pieces, including chase sequences and fights;
- story events that fit the overarching narrative but which adapt to a sandbox environment;
- an audio score that foreshadows unscripted events and announces the presence of hero characters;
- large-scale crowd and vehicle simulation;
- adaptive character animation and movement;
- accurate matchmaking algorithms for multiplayer online games;
- elegantly handling dropouts with automatic AI takeover;
- automatic navmesh generation from a polygon soup;
- predicting player behaviour to counteract controller and network lag; and
- automatic exploit detection and prevention.
The problem is that the role of the (usually lone) AI programmer on a game development team often involves many tasks that get in the way of performing research, including asset acquisition, audio and animation integration, data production, tool implementation and support, multithreading support, optimisation, debugging and so on, leaving a perfect opportunity for academia to supply the research chops. What’s needed are robust, efficient, designer-tweakable techniques that are easy to debug, and which scale with available CPU and memory. Sadly these requirements are not a priority for researchers, and yet researchers remain perplexed that game developers don’t use some of the inefficient, unpredictable techniques that they develop.
You see, the problem is that your neat little algorithm might perform well 95% of the time, which may be a great improvement over the state-of-the-art, and which may justify publication, but 95% is not good enough when you have an audience of 5 million game players (as hundreds of thousands of them will see broken behaviour).
But the biggest point of contention between game developers and researchers is that we gamedevs think that cheating is acceptable. After all, a game is just a Turing Test, with the player deciding whether intelligence exists based on the behaviour they perceive, so why not use all available information to deliver on that promise, instead of placing artificial restrictions on what data can be used based on whether or not it would be available to a human player? It just doesn’t matter how the behaviour is achieved – we’re not looking for insights into how the human brain works – it’s all down to player experience. This behavioural approach is out of favour with researchers (and has been ever since Chomsky defeated Skinner), but is the core of pragmatic game design. Perhaps never the twain shall meet.
4 Comments
An interesting post, even for someone like me who knows next to nothing about AI other than my interaction with games. The less I notice the AI, the better.
Lately I’ve been playing Red Dead Redemption – about 22 hours last time I noticed the clock. Now I don’t often play first person shooters, so I am no expert, but I did notice the AI in this game for all the right reasons. The camera responds realistically. The audio score is fantastic – it foreshadows unscripted events in a way subtle enough that if the event doesn’t, happen by player choosing, it doesn’t feel as if it was out of place. There are a LOT of cut scenes, which usually annoys me, but they are a joy. They also don’t feel awkwardly removed from the in-game action. Great cut scenes that look nothing like the in-game action are also annoying.
Like a good song, I was a tad disinterested after the first few hours of gameplay but it got better and better. I prefer it to Bioshock and GTA4, my previously favourite FPSs. Give it a whirl if you’ve not already done so.
Just read your previous post. Guess you won’t be playing Red Dead anytime soon. Not many people I know can find the time to finish such games. Even I struggle, and I don’t have little ones. Will give LIMBO a shot. I’m a sucker for a good platformer. A very average platformer can keep me at it for hours upon hours. A good one is more akin to an addiction than a game.
P.S. Half-Life 2 WAS repetitive and boring. Part of the genius of RDR is that the repetition makes the game that much more immersive – riding a horse through the valleys and mountains along streams and in snow – it is supposed to be slow and repetitive.
I’d love to experience Read Dead Redemption, and I probably will. I just know that I won’t be able to commit to completing it. It would be great if the experience of playing RDR could be boiled down to four hours.
These games aren’t first-person shooters, by the way (well, Bioshock is, but not GTA4 or RDR). They’re sandbox / open world / third-person action-adventure games (SOWT-PAAGs). Yep, that’s the official acronym for the genre.
Oh, yeah – not technically FPSs. When I said I’m not into FPSs, I meant “shooters”. Shooting stuff is not usually my cuppa. More of a platformer/puzzler me…
It would be cool if games like RDR could be distilled into four hours of fun. I loved GTA4, but I really did get sick of it 3/4 of the way through the story. But I couldn’t stop without finishing it after putting so much time in.
Long play times are not a new thing, right? The first couple of Super Mario games took me ages to get through and were adventure like. Some games were damn hard back then and took a long time for that reason, but SMB and 2 were long, from memory.
RDR is a bit different. Like I said above, much of the mood of the game really is developed through the slow, drawn out action. Good thing is you can opt out of much of the travelling if you wish. Many of the scenes come almost straight out of Sergio Leone masterpieces and similar, which are all very long movies with similar motivations I suspect.
There must be a reason studios don’t offer distilled versions, but I think they’d go well. Maybe research tells ‘em that too many people would forgo the longer (more expensive) versions, making the dev of those longer versions unprofitable?
You can have my copy of RDR soonish if you wanna give it a spin. I’m off for an hour or three of kickin up sand; yeehaw!
Post a Comment