Interview with Pranas Pauliukonis

This is the continuation of our interviews series of the folks at Big Huge Games, which was conducted during our visit to their office back in March 2004. We put our focus this round on Pranas “Pancake” Pauliukonis, programmer in charge of AI (artificial intelligence) at Big Huge Games.

PP: I’m being recorded! Oh boy, I’m being grilled!

Soc: To start things off, could you give us a bio? How you came to work at Big Huge Games?

PP: I went to computer engineering college and part of that was working on robotics; helping the robots find ways to get places. I was interested in that. At the time I had a friend who suggested I get a job working on Madden Football at Electronic Arts. I got that job and worked on AI for them. I was there for about a year and a half, then that same friend told me about BHG. I applied and started working here. Here I work on AI, pathfinding stuff. I’ve expanded from there into a lot of game side systems; keeping things knit together.

Soc: How do you like things here [at Big Huge Games]?

PP: It’s pretty sweet! The people here are awesome to work with; everyone knows what they’re doing. All work great together. I’ve made great friends. The offices are really nice. Also, I grew up in this area, nearby, in D.C. It’s nice being in this area. It’s really nice just as a programmer…we get to express our creativity, ah…it’s not like being given a list of eight things you’ve got to get done; it’s more a general “this is what you need to do.” I’ve been in a job like that, that other kind where it’s not as fun an environment to be in.

Soc: What kinds of things do you do here in your job?

PP: Umm…I do, let’s see, everything from…My main job is to make all the game rules, and let’s see…game rule implementation, low-level AI like in pathfinding and combat; unit-level AI, strategic AI for end-game, armies, construction, except for some of the early build stuff done by the designers. For a while I was doing some multiplayer stuff. Kind of making sure the end game runs from a game standpoint, not graphics or actual numbers [unit/building stats etc] in the design.

ODA: In terms of the AI, it’s one of the best ones I’ve seen out there for RTS. What element made the AI stand out?

PP: I think the thing that helped the most was I actually spent a lot of time watching the multiplayer games being played everyday [by the staff], and was watching them and see what they do and really just try to write the AI to do the same thing they were doing, such as keeping armies as cohesive as possible, try to actually have goals. In some ways the structure of the game helped me, because you have to have clearly defined goals – “I have a city I want to capture,” thus a matrix: “get to that city, capture the city.” In other games that would be a little too hard where buildings are strewn across the map. Try to make it look as much like you’re playing a human as possible. Yes, the AI grew out of the game structure.

ODA: You worked on sports games. How did that influence development of the AI in RoN?

PP: I don’t know other than just building up my skills as a programmer. In sports games like football you’ve got eleven characters each with something specific they need to do. There are some AI choices made as to where to have coverage. It was nothing as complex as what we do in RoN.

ODA: What about the formations in [football] plays, where guys would be doing these flanking runs?

PP: We had stuff like that, but in Madden a lot of that stuff was done through scripting that was done before I showed up.

ODA: In RoN was that handled similarly, or did you use scripts too, to compose some of the builds?

PP: There’s scripting to handle how the initial building is done, and in the Xpack scenarios campaigns where it’ll try to coordinate what the AI is doing. It’s nice to put things in script because it’s quicker to change things and lets more people look at it. If it’s in code, then it’s harder; if there are problems it takes longer to make changes. It’s easier to do things with scripts; the designers can look at it and even the playtesters can look at it. I think probably we’ll start moving things out into scripts more in future, as a general principal.

ODA: What about scripting the low-level [AI] stuff so people can tweak the AI?

PP: Probably the deepest low-level stuff we wouldn’t open to let them do it. For instance, if I’m a guy running around trying to decide who to attack. That kind of AI stuff has to run really fast…putting that into script slows that down somewhat, and in putting it into script you may not be able to tell you’re slowing it down. So we’ll script more for strategic-level stuff, like which city to capture, military stuff, economy stuff.

ODA: So you’ve built scripts by watching players play. Did you look at strategies people always try and then try to counter those strategies in your AI?

PP: We tried to do some of that with the openings. The openings are really tricky, because it’s so important that you do something in a certain order. Trying to get it…it didn’t make sense to do something that adapted too much because it would make the AI too generic and not be able to handle a well-coordinated rush. Even now it has trouble handling a really well-coordinated rush. That’s a kind of case where we’re looking at people playing the early stages of the game and try to match that with the scripts that we had. Not just be able to counter the player but also that helps later on to get a base for the AI – seek more resources to work with, improve its strategic position.

ODA: If someone wanted to clone a strategy using the AI script to practice against it, is that possible?

PP: To some extent; there’s only so much we can do. There is only so much power it has. There are some ways you can change the build orders; well, hopefully, you know, so so.

ODA: In terms of some of the more traditional AI methodologies, fuzzy logic or case based, what methodologies are you trying to do at various levels of the AI?

PP: Not so much the classical. I Did some work with neural nets in college, some fuzzy logic. RoN in its current state is a little too complex to use algorithms like that. We have some standard things, like the standard A* pathing algorithm. But as for other methods, there’s some fuzzy logic. The AI waits and sees different game parameters and then decides what the current best plan is. That’s the closest we came to a traditional AI.

Soc: What was your biggest challenge in designing the AI in the game?

PP: Probably the biggest challenge was keeping up with changes in the design. Because I don’t know if you know, the design changes happen fast around here. We would be working off one set of game rules, and then Brian Reynolds would go into the game, since he’s a programmer himself and go “now I want the game to work this way” and the all of a sudden the AI is completely broken, 2 minutes in. So part of the challenge is to just keep up with that. We needed the AI work pretty consistently throughout the whole project. Cause you can’t just play multiplayer games. People need to play against the AI The designers might be playing the entire day, so they need an AI to work.

ODA: Have you tried matching AIs vs. AIs?

PP: On the design side it’s not done so much, but I do that sometimes to watch for failures in the AI. The quickest way to find out if something’s wrong in the AI – I’ll set up four AIs playing against one another and increase the speed of the game beyond what a person would play at and see if one side gets a huge advantage or a breakdown develops where someone isn’t attacking.

Soc: So, are you making changes to the national and unit-level AI for the [Thrones & Patriots] expansion?

PP: There aren’t too many changes if I remember. There are some new kinds of units that are handled differently, like Elephants. Elephants have a guy on top who shoots arrows or cannons later on while the elephant walks up to people to gore the guy. It took writing new AI to account for the new mechanics involved in his deciding whom to shoot. Other than that, the nations didn’t need much. There’s some like the Lakota, like the Lakota are less restrictive on where they can place things.

ODA: The way the Dutch banking bonus works is kind of counterintuitive. What did you do there? Are you planning changes?

PP: I didn’t want to change the AI much for the Dutch. Because I was never too happy with the way the AI uses resource. They could accidently end up stockpiling resources. It may sell it at the market if it feels it should, but sometime they just end up having a lot of wood and doesn’t know what to do with it. But then with the Dutch, and that doesn’t turn out to be such a bad thing. But if they have a lot of wood but just end up with more wood if that happens they can just start abusing the market. So with the Dutch that’s already in there, same kind of thing applies with the Indians and their ramping cost advantage. The AI takes account of the ramping costs to decide when to build so it just kind of naturally worked out for the Indians; “Now I’ll build more buildings.” now that its cheaper. Not too much specifically done on the AI, just that the national powers themselves changed.

ODA: In terms of the CtW campaign, on the strategic map did you have to make consideration in terms of how the AI would work?

PP: That’s one part of the AI I didn’t really work on, so I can’t speak too much to that. There wasn’t much to add to the Alexander campaign in terms of the AI in that campaign.

ODA: In terms of the AI personalities did you do anything to match the historical nations in the Conquer the World campaigns?

PP: I don’t know the kind of things they do in there; so I can’t really help you. Sorry.

ODA: An issue some people found with the AI in very long games. The one thing it’s susceptible to is rushes, and in longer games you don’t necessarily need to rush the AI but out-ramp it. It wants to dawdle around and not want to attack you, or even when its an ally, it doesn’t seem to attack its enemies.

PP: That’s a tricky problem. The identification of how much I [as the AI] should sit back and do vs. how much I should attack. Unfortunately what it ends up doing is trying to do both. It never does a true boom and doesn’t build units if the human player doesn’t; it’s not like a human. But we didn’t want it to cheat like giving it the ability to check on your army and say, “You haven’t built any units so I won’t either” and then do a true boom. No, I didn’t want to write an AI that cheated like that.

ODA: Is there anything to address the problem? When playing against the AI, if you do out-boom it, you can go in and trash it.

PP: I don’t think we have much time is left to do that kind of thing.

Soc: If you had a chance to be involved in writing the AI for another kind of game, what would it be?

PP: I wasn’t too into sports games myself, so I can’t say I’d really want to go back there. I like seeing the AI done in FPS’s. It would be cool; it would be a really different challenge from now where you’re trying to manage hordes of guys. It would be cool dealing with the specific actions of one unit to the finest detail. That’d be different. Any strategy games are interesting to me. I just like games. I like to figure out how games work and play the game as best I can. Any kind of strategy game.

Soc: Regarding RoN, what’s your favorite national or unit-level AI aspect?

PP: That’s a thinker. Hmmm. Let’s see. I’ve never really thought of that. One thing that was really an interesting thing was working with supply wagons and generals – how they really have to change their behavior based on things going on around them. I don’t know how much people notice that. With like a group “Attack” order or “Attack to” orders they have to stay behind the line yet stay with the troops. Maybe players don’t notice when you’re playing but it’s important in the AI. Before that [part of the AI] was done they would charge straight into combat, but we wanted Generals and Supply Wagons to stay where they’re needed and not have to rely on players’ micro.

Soc: What’s your favorite unit graphic?

PP: I’m really partial to the Treb [Trebuchet]! I love watching the guys take it down and, well, you see in other RTS’s the siege moving around by themselves. I was really impressed by the movement of the cart guys pulling on ropes and pulling on the lever to fire a shot.

Soc: Who would you say to be your guess to be the best and worst RoN players on staff?

PP: Here? Let’s see…I can tell you among the programmers it’s Jhon Restrepo – he was a recent hire. When he got hired, he started playing Rise online like crazy and can trounce any of the programmers here. Last I heard it was Johnand Mark who were the two best, John Hawkins and Mark Sobota; they’re two of our designers. I think the balance of power may have shifted while I wasn’t paying attention.

ODA: What are their Gamespy ID, so I know who to avoid! Haha!

PP: I know Restrepo’s is the Columbian…

GS: Oh he uses that as his smurf name.

PP: Oh does he? I didn’t know that. Haha!

ODA: Well at least we know one now!

PP: Oops, well he’ll have to get a new one. I forget what John and Mark play as…

GS: Hawkins is Pikachu’s Party Pal…

PP: The three of them they get into some good battles…You always want them on your team. Haha!

Soc: Do you have a favorite in-game taunt?

PP: “Purple” was the one for a while. Purple is good. I actually like making the icons in the chat box.

Soc: What are your future plans in the gaming industry and in particular BHG?

PP: I’ll say this: I definitely want to stay at BHG and keep making great AI. I’m happy here. As a company we came out of the gates strong. I’m glad the AI was up to par. I still love AI and want to keep working on AI, keep making it better. As for the game industry in general, I’ll be here for a while. It’s a really neat place. I’m having a great time here; there’s more for me to do on the AI.

ODA: There’s a question I’d like to go back to. Regarding scaling the toughness of the AI, are there any other ways to scale the AI abilities vs. the human’s so that there’s a range of strategies or something like that?

PP: We do, As far as a range of strategies, usually it’s handled by pulling the AI’s teeth on the lowest levels, like make it try to capture fewer cities. Now here’s where we sort of let it cheat and let the AI see how many units the player has made and then send in a tiny army as if to kick the user and tell him it’s time to make units. The other way, we can scale how much the AI cares about having and maximizing resources. If you’re playing against a tough AI, it’s always maxing out and use it’s econ cap and increasing it’s econ cap. Always makes sure it has econ going. Whereas on moderate/easy the AI doesn’t care if it’s 50 below the cap, but “I’ll won’t make anymore citizens”. That naturally gives it less money to do things, and slows the game down to a beginner’s pace.

ODA: What about surprise strategies, where you see a civ that’s more suited for booming but then suddenly rushes?

PP: Not so much of that, not so much of that. That is actually something it does on the easy levels. On the easy levels the AI cheats and looks at your units and builds the WRONG counters; it tries to make guys that will get slaughtered by the player. We had different scripts based on rushing, and booming, and had the AI roughly choose those strategies based on what civ you’re going to take. But that is an interesting idea; I’ll keep it in mind. Interesting, I’ll have to remember that!

Soc: What’s a rough estimate on the time needed to make the rough draft of an AI?

PP: Ummm…it depends on how much AI you’re looking at. Just talking about the very basics on the strategic level stuff…getting a very rough draft is very simple, just a couple of weeks, getting the structure of things. In that time you will just be able to do all the things you can do in a game. After that it takes just a limited time to make it do things well, do things at the appropriate places. So getting the structure up is pretty simple, but there are lots and lots of details. The more complex the game is, the more complex the rules will be. I think, when we came up with RoN…in RoN we tried to simplify the interface somewhat, but there are some really complex game rules involved, and that increases the time exponentially.

Soc: Besides RoN, what other computer games do you play?

PP: Let’s see. I like FPS’s. We play Battlefield 1942 a lot. Counterstrike’s fun. Halo is fun. Nintento first party titles – I love Zelda. I play a little bit of everything; Gotham Racing for a while. One kind of game that I used to play a lot but don’t play much any more is RPG’s.

ODA: With RoN there’s a big, strong single-player community, and I guess the AI plays big role in that. Did that add responsibility to your plate?

PP: I definitely felt pressure; it was a challenge to try to add Civ-like single-player elements to an RTS. We figured from the start that a lot of players would be interested in just single-player. It felt like a lot of pressure, but we felt we could do it and once we got to the point where people not yelling at the AI for doing stupid things, even if it took a little time to get there, we knew we would get there.

Soc: It seems there’s a good-sized community here at BHG that likes to play board games.

PP: Yes; we’re getting too many now, a lot of games takes six people to play max. We’re a little over the limit on that. When I was a kid I played Monopoly, Risk, games like that. The board Games they’re making now are really interesting. They involve lots of Strategy. [Playing board games] helps designers more but it does also help me to play a lot of games and different kinds of games and see commonalities, what makes a game interesting or the parts that are not interesting. It’s cool to look at that.

Soc: What’s your favorite board game?

PP: I’m a big fan of Settlers of Catan…

ODA: Oh, yeah, I’m into that, too!

PP: That’s a great game; I love that one. There’s a new game that came out, Game of Thrones, kind of like Diplomacy…The big one around the office now has been Amon Ra – people have been playing that every day for three-four months.

ODA: A number of years ago there used to be many superior board wargames. I don’t know if you’ve heard of them, but there used to be Avalon Hill wargames. Did you draw inspiration from them?

PP: We’re constantly getting ideas from board games. The designers do more than I do; they’ll get a game, play it and talk about it, and discuss the mechanics: “This was fun, this not,” and try to apply it to computer games. Though it’s not a straight application, there’s definite inspiration.

ODA: OK, here’s the $64,000 question or the Million Dollar question!

Soc: The most difficult question you’ll ever have!

GS: What’s your favorite superhero? Haha!

Soc: How do you like your coffee?

PP: Lots of cream, lots of sugar, lots of caffeine.

Soc: Lot’s of people seem to like tea around here. What about tea?

PP: I’m not one of the big tea drinkers, not so much anyway, but I like chai. It used to be everyday we’d run out to Starbucks, but that got a little too much, people were getting twitchy! It’s some powerful stuff! Haha! Actually people goes back and forth…

Soc: I think that’s it.

ODA: Ok, let’s get some pictures.

PP: Alright! Cool, thanks!

ODA: Thanks, nice to meet you!

Legend:
PP = Pranas Pauliukonis (Big Huge Games)
GS = Graham Somers (Big Huge Games)
ODA = One_Dead_Angel (Rise of Nations Heaven)
Soc = Socvazius (HeavenGames)

Transcription:
Rohag

Editing:
One_Dead_Angel, Graham Somers