Discussion
pie-graph

What’s Holding Back Game AI Most?

Alex J. Champandard on September 11, 2008

A little earlier, I posted an editorial which speculated that the problems we had within the field of game AI weren’t necessarily technical. Most of the problems we face in industry these days are solvable, given enough development time — although we could certainly become more effective at it!

Now, you don’t necessarily need to agree that Game AI Has Been Solved™ (I’m not sure I believe that entirely myself), but it makes for an interesting exercise nonetheless, particularly to help identify what the major problems are in our part of the industry. So, technical issues aside, what do you think is holding back the field?

Discussion 13 Comments

zoombapup on September 11th, 2008

Can we add "the lack of will to actually develop more compelling AI characters" to the list? Thats what I see as the problem.

Jare on September 11th, 2008

What about "AI devs distracted by general gameplay and animation scripting tasks"? :) Sorry if this sounds like we're bombing your poll Alex, but I'd hate to lump Zoom's or my point under "General Game Dev Challenges."

zoombapup on September 11th, 2008

Actually you could put both of our issues together as "a lack of developer priority" Essentially, we CAN deliver really amazing AI, but its not seen as valuable enough in production terms to really expend the effort to achieve that. I think its basically that we dont have an ultimately compelling way of evidencing the value that AI can add to a game. But luckily, it only takes one developer to really see it as a USP for us to see some movement in that direction. Research is definitely going to start opening that up a little, but I think the first game to really "get" AI and spend enough time implementing a concept that holds AI as a high priority feature will be the most compelling argument.

alexjc on September 11th, 2008

O.K. Good points Phil and Jare. I added a new poll option: "Low-Prioritisation of AI vs. Other Issues." I wonder what we can do to fix that one! Alex

hellokeith on September 11th, 2008

Lack of priority / dedicated time given to AI is a symptom of the real problem: AI is increasingly less and less valued in an online multiplayer-dominated gaming industry. It is in a company's best interest to tie game value to online player audience size/interest, because it ensures the company's ability to create a new game and/or sequel every other year. A game with great AI and/or COOP extends a game's life, thus reducing the amount and frequency of money the player will spend.

Dave Mark on September 12th, 2008

Not enough obscure metaphors. ;-) (Thanks Alex for giving me a week off!)

alexjc on September 12th, 2008

Arg, I should have split that prioritization option into two... Now it's going to be hard to decipher whether people meant: [LIST] [*]There's little interest in building designs with advanced AI. [*]During development, people get reassigned to game code because it seems to take more time & effort! [/LIST] So those of you who voted for that, feel free to clarify! Alex

scypior on September 12th, 2008

I think that very important fact is that the game are constantly cheating the player. And the success is when the player doesn't notice that he was cheated. That way, using simple technologies you can achieve sometimes more things that using more complicated ones - one example is coming to my mind: RAI thingy in Oblivion. It was so hyped before release, and ended to be one of the stupidest AI in games. To goal was admirable - to have real living NPCs, but they failed (probably because team involved in this was rather small, and they probably put their priorities somewhere else, I don't know). On the other hand, in game like The Witcher, town NPCs are [I]totally[/I] scripted. But in the end it looked much better than in Oblivion. So maybe what is holding the AI is the thing that the players are happy with that simplistic solutions? Maybe they like stupid foes, cause they play the game to relax, not to be frustraded because of fact that the bot is more inteligent than the player :D "[I]The biggest enemy of the game is the player himself[/I]". Sadly there actually is some truth in that sentence...

zoombapup on September 12th, 2008

I think the problem is that there hasnt been a "killer app" to demonstrate the really pervasive use of AI as being commercially useful. Perhaps its really going to be an incremental shift thats going to happen. Unfortunately game designs dont tend to move forward all that much (look at the latest doom games and compare them to the early wolfenstien games in terms of design). For all the graphical advances, we've made absolutely no advances in design (hell, I quit out of playing plenty of modern games because the designs are so cliche and have been done many times before).

Kevin Dill on September 12th, 2008

So I voted for my favorites, but let me hit on a bunch of them real quick... I'll take these in the order of votes they've gotten so far, if only because I don't remember Alex's original order. :) * Low-Prioritisation of AI vs. Other Issues I definitely agree that this is #1. I don't feel like it needs to be split, Alex - both of those things are contributing factors to the overall problem. The big problem is that AI is not taken seriously. With that said - I get the impression (maybe it's a pipe dream) that Rockstar is starting to really take AI seriously at the highest levels, so maybe we can give the rest of the industry a good solid kick in the ass. ;) * Too Little Practical Knowledge Exchange I'm not sure I agree with this one. Sure, I'd like more, but we have several forums for knowledge exchange, and they get a fair amount of use. There is always at least a talk or two at GDC. There is AI Wisdom every year. There is AIIDE. There is Alex's web site. I find that there is more practical knowledge out there than I have time to absorb. If there is a lack of practical knowledge exchange, it's either because people are reluctant to talk about what they've done (they either don't think it's important or just can't be bothered), or because management is suppressing it. Of the 5 companies I've worked at, management only suppressed it at one (and even then they don't suppress it completely - several Rockstar folks have published over the years), so that doesn't seem like the primary problem to me either. So... those of you who voted for this one - your task is to write a paper on something you've done by the end of the year! (Hint: AI Wisdom CFP will be around the time of GDC '09, as will AIIDE) * Insufficient Code Sharing & Reuse Maybe... the thing is, the hard part of AI isn't the reasoner, it's all the logic that goes into making the decisions by using the reasoner - and that logic is game-specific. This is also why I'm not hugely hopeful for AI standards or middleware to save the day. * Shortage of Skilled Developers This is a big one for me (I even voted for it :)). To my mind, really good AI programmers are a special breed. That's not to say we're better - but we have a different set of skills than, say, a graphics programmer. We attack problems in a different way. I can't quite define how to recognize this, or even what I'm talking about, but I definitely know it when I see it. We revel in uncertainty and fuzziness. We think about the world - everything in the world - in terms of how it could have been made more intelligent. There are far too few people who think about things that way. I don't know how to create more or how to find the ones and exist that hire them (Although *shameless plug* if you're looking for a job at a stellar game company in the Boston area, drop me a line...) * Middleware Products Not Suitable I'm a skeptic regarding the utility of middleware. As I said above, the hard problem isn't creating the reasoner, it's the game-specific stuff. No middleware can be game-specific (cue John Funge ;)). * Shortcomings of Academic Research (14%, 13 Votes) This definitely exists. With a few glorious exceptions almost all of academia doesn't "get it." With that said - so what? Who needs those guys anyway? We're just as smart as them (*ahem*, smarter! :P), and we know what the real problems are. If we're not solving them then let's address the reasons why *we* are not solving them, not look for somebody to ride in on an ivory horse and solve them for us. * No Affordable/Accessible Conferences (8%, 7 Votes) Yes, but the problem isn't that the conferences don't exist. I see two problems here: 1) We're not taking the ones that exist and making them our own. AIIDE is comparatively inexpensive, happens every year, and is a great chance to pull ideas from both academia and industry. Academia has shown up and is trying to share - and trying to find out what we need. To date, we have gotten almost zero industry participation. I honestly don't understand why - we typically have really fantastic speakers (better than GDC, if you want AI-specific talks) and the opportunity to hang out and talk to people like Will Wright and Peter Molyneux. But regardless, industry support has not materialized and shows no signs of doing it this year (we got a handful of industry papers, but not very many). Worse, I think those of us from the industry side who have been supporting this and helping to make it happen are starting to get fatigued - and I don't see a fresh crop stepping in to replace us. At the rate things are going, my prediction is that this will be a purely academic conference within the next 2 or 3 years. 2) Management is giving little if any support. They certainly aren't pushing to send people to conferences. Even if you get out and find out about a conference and ask to go, you may very well not be allowed, and you will almost certainly not get any monetary support. This short sighted view on the part of management seems to result from the fact that (a) they are usually hard pressed just to get their current game out the door, so worrying about the future beyond that is pretty low priority, and (b) employees tend to move around alot anyway, so even if they spend money on developing you, odds are you'll just take that expertise with you when you leave. I don't agree with either of those points, by the way, but that seems like the mindset that we need to counter.

alexjc on September 12th, 2008

[quote=Kevin Dill;4775]I don't feel like it needs to be split, Alex - both of those things are contributing factors to the overall problem. [/quote] To me there's a huge difference between AI taking low priority on purpose because of design decisions, vs. AI being neglected indirectly because the rest is so time consuming and challenging. I personally think that the second is more of a problem than the first. If game logic was easier to build, it'd improve the quality and focus on AI tremendously... Alex

RyanP on September 12th, 2008

Alex, While I do agree with some of your points I do not feel that Game AI has been solved. In certain cases such as animation it may be as we have all seen the animations that blend into each other and new ones formed. However even in something as simple as path finding I do not believe it has been solved. You can give objects, units and waypoints, attributes until the cows come home but if they cannot find the way logically through a series of objects we have not reached the pinnacle of even path finding technology. I have seen many of the AAA developers as you’ve mentioned create enemies that walk into walls or don’t follow the logical path. Now this may be due to other factors as well as I did not develop their path finding algorithms however from the visual interpretation of those algorithms they are not working as well as they could be. Your mentioning of the different factors that could be holding Game AI back I do believe some of them are correct and others may or may not be. A) Shortcomings of Academic Research I am not sure in this area however I do not believe this could be a holding Game AI back, unless they are just not researching in areas that would be beneficial to Game AI programmers. B) Shortage of Skilled Developers I have not worked in the games industry so I would not be able to comment on how skilled the current developers are however I do know when I apply to jobs for Game AI programmers they never get back to me!  (Short shameless plug back to Kevin Dill) C) General Game Development Challenges Having not worked in the games industry I can’t really comment on this although you’re synopsis sounds good. I’m going to mix the order up a bit here. E) Middleware Products Unsuitable This is only my opinion on what I have seen and witnessed however this may be holding Game AI back a bit, because from what I have seen it appears all Middleware companies do is create a library for something that already exists and that there is plenty of information that companies already have people doing on their own. I am going to combine D, F, G, and H D) Too Little Practical Knowledge Exchanged F) Insufficient Code Sharing & Reuse G) Not Affordable / Accessible Conferences H) Missing Game AI Standards I do agree with all of these to a point. From what I see online everyone is very tight lipped about the who/what/how of what their company does for AI in regards to a game. This not only leads everyone else to do the same thing, it also has everyone else re-treading the same territory as opposed to creating something new. Now I understand that they are all in competition with each other, but if all of these companies are using the “best” algorithm set as you are saying for certain things shouldn’t they just be sharing the code as its widely available? As for not reusing the code, isn’t that what object oriented programming is all about? Reusing code over and over, it seems rather silly if they aren’t reusing code? I cannot stress to you how much I agree with you about the conferences. The conferences of AIIDE or GDC or the one you had in Paris, all of them were or are expensive for myself to go to at the moment. I was looking into going to GDC next year for 2009, $2000 for the 5 day ticket? Not including air fare or hotels or food or anything. So quite a bit of money and that puts a damper on visiting conferences to either share my ideas or learn new ones. The reason I added the “to a point” after saying I agreed with all of your statements was because of the Missing Game AI Standards. I agree it would help lead the AI forward however having a standard is because there is no other efficient way to do something. So far there is nothing in Game AI that I would say is the absolute pinnacle in that area and that everyone should adopt it. We may be close in certain areas but nothing from what I have seen is at its pinnacle yet. I) Lack of Processing Power Unlike most people I am not quite a firm believer in this as I mentioned in other posts, most notably with Paul Tozour in regards to an AI processing hardware card. I think that all the power is nice, but until we can max the power used on a CPU in regards to AI and be fighting with the game engine for every little drop of power that the processing power question is moot. We do not do that now; most of the games use a fraction of the CPU for AI and all the rest for the game engine. Now in the future I am sure this may come to be a problem however at the moment I do not feel it is. Thanks for reading, Ryan P

FBMachine on September 12th, 2008

It's all about deadlines. It's much harder to take risks with AI than with graphics techniques - scrapping your spherical harmonic uber lighting pipeline has a lot less impact than an AI foundation that just isn't working adequately. So, we go with the simpler methods that we know are going to have predictably shippable results. That's why I would think we're only seeing minor progressions in AI, and nothing groundbreaking. It's just too risky. :) Daniel

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!