How do you Rate the Progress of Artificial Intelligence in Games?

Alex J. Champandard on October 9, 2007

It’s a bit of a controversial topic lately, so it seems perfect for this week’s developer discussion! Remember, you can win an T-Shirt at the end of the month for interesting comments, so join in by posting below!

Games have been around for quite a few decades already. Over the years, there have been visible improvements in the quality of the graphics, the sound, and even the physics simulation. However, can the same be said for AI?

From the perspective of a game player, are you satisfied with the current level of AI in video games? Did you expect quicker or slower progress? What do you expect from game AI in the future to add to your playing experience?

Join the discussion.

Game Characters Chatting

Next week’s discussion: technological improvements of game AI.

Discussion 6 Comments

dphrygian on October 9th, 2007

As a player, I expect two things from AI. First, it should provide a fair and sufficient challenge, and second, it should provide a reasonable facsimile of the behaviors of a human being (or alien/zombie/robot/etc., as the case may be). The first expectation is the easier one, and it's the rare game these days that doesn't satisfy it. The second expectation is more challenging. I believe that this simulation of intelligent behavior is the primary difference between game AI and more purely academic problem-solving AI. It's a problem space that involves not just programming, but design, art, and audio as well. The reason it is so challenging is because, as human beings, we have well-established expectations of the behavior of other human beings. (Incidentally, low expectations is probably why no one ever complains about zombie AI.) Game AI is virtual acting, and if it isn't just right, our psychology rejects it. I don't think we've approached the Uncanny Valley in commercial game AI yet, but more progressive experiments like Facade hint at it. Part of the problem lies with game design (for shooters in particular). If an enemy AI only exists to put up a fight for five seconds and then die, there's not much time or space to make him a believable person. The appearance of team tactics can add a lot to this--in fact, verbal communication between enemy AIs can create the illusion of teamwork even if there's not much going on under the hood. I believe you mentioned this in the article on Thief recently, but giving players information about what the AIs are "thinking" is critical. An AI may have an absolutely top-notch decision system under the hood, but if the player doesn't understand why it makes the choices it does, the effort is wasted. The AI could have moved randomly and the player wouldn't know the difference. Friendly AIs, such the player's squad in Brothers in Arms, are also effective for creating the illusion of thought, because they are designed to stay alive longer than the typical shooter baddie. This gives them time to perform more behaviors than just run, shoot, and die. It also allows players to make an emotional connection to the characters. Bioshock's Big Daddies and Little Sisters are another good example of long-lived and emotionally interesting AIs. The player can follow them and observe their behaviors without needing to engage the Big Daddy in combat--and when they do choose to fight, it is a more meaningful fight for having made that connection to the characters. As I see it, the AI programmer's job at this point is to provide the tools to enable these kinds of progressive AI designs. In this regard, I think the state of the AI scene is looking good. Hierarchical state machines and planners are allowing us to implement more complex and interesting behaviors in a systematic and extensible way. There's a lot of exciting potential in these systems, and even little things like the enemies knocking over objects to create ad hoc cover in F.E.A.R. can add a huge value to the player's enjoyment of the AI.

diegix on October 10th, 2007

I think AI is right now just keeping pace with all the other aspects of a game like graphics and physics. To me, great improvements have been made in the graphics and physics areas, probably because you are sure that the time invested in improving the graphics of a game is going to be seen in the game. However, that's not always the case in AI, you can spend tons of work doing a funny and believable AI, but that's not going to be seen in the screenshots and not so much in the videos that sell games. In the past generation AI had to improve a lot, 3D worlds, more complex games, open worlds certainly are things that AI has learn how to deal with and it's working. In this generation it's more about adding tons of detail to the same games with had in the past and the goal for AI is mainly have an AI as good or even not so good as in the past generation working in this huge and richer environment.

Sergio on October 11th, 2007

Seconded. Both [URL=]Warren Spector[/URL] (of Deus Ex fame) and [URL=]Damian Isla[/URL] (lead AI for Halo 3) mention that with the added complexity of game worlds for next-gen games, we're more often than not lucky to get the same behaviours we got a few years ago for the much less powerful PS2. So in a sense, AI is progressing rapidly just to be able to keep up. I just hope we are not about to hit a technological wall that will make world's complexity to grow exponentially in terms of AI, reaching a point where we just can't catch up. It doesn't help that perception also changes as visual quality and realism increase. I remember thinking that Doom was impossible to surpass graphically. Enemies then only needed a simple animation to look real. Today, the posts in this blog about Siggraph are proof that we need better technology for animation than current research can provide, if we want characters to blend in photorealistic environments. I still think we'll eventually get over this, and overcome the wall if we reach it. But I don't expect order of magnitude improvements like we've been seeing for graphics and physics.

alexjc on October 12th, 2007

David: Great points. Are you happy with this progress though? :-) Diego, Sergio: Ah yes, this one is becoming a bit of a trend lately. Progress in graphics and physics amounts to a regression for the AI! So either we can hire more AI coders, or tune down the complexity of the world... It seems we'll need much better technology to get over this hump... But we can discuss diminishing returns on technical investment in more depth next week! Alex

Andrew on October 13th, 2007

It's keeping up for most first person shooters where the level of intelligent behaviour is pretty easy to fake, and where good pathfinding pretty much makes it intelligent :) I'm interested also in the more strategic AI's and sadly, I don't see the same kind of progress (even pathfinding progress), since even newer RTS games, the AI is easy to defeat since it uses no real tatics or map intelligence. A real shame really! (I have, however, not played M:TW 2 yet). So, strategy AI is improving to encompass new gameplay and mechanics, but not at the rate of graphics or other parts. A players perspective really, but I do hope to see AI working on the more lifelike things, and games like Spore might up the ante like The Sims and Black and White did.

avok23 on January 29th, 2008

TBH only action games have improved in ai. Most people still make games like they were in the snes era. Racing games are still all about the fastest cars, same with sports.

If you'd like to add a comment or question on this page, simply log-in to the site. You can create an account from the sign-up page if necessary... It takes less than a minute!