News that a genetic algorithm has cracked the Starcraft code prompts Troy to ask the panel about the place of build orders and in-stone strategies in the genre. Julian talks about chess and Hoard, Rob talks about the beauty of a game that opens up and Troy sucks at everything.
Also news on the upcoming DC area fall meetup. November 20th, btw.
Not to take away from a very interesting discussion, but in this particular case you guys had a couple things wrong:
The genetic algorithm doesn’t play against itself, it just creates a build order, then tweaks certain parts that don’t work out until it obtains an optimal time for the parameters for which it is fed. In this case: seven roaches.
The early roach warren and spawning pool can be scouted; you don’t have to automatically start the counter at the beginning of the game. At this early point in the game, every race pretty much has the same few moves: Make workers. I’m not sure what the particular counter for each race is, but I counter it with the simplest early game tactic: piles of zerglings.
I like the build orders for particularly competitive games, because it teaches a newbie about the game in a way. Instead of coming up with general theories (or excuses) why you lost, you can look at a build order and analyze why it particularly works, given that there are few parameters to analyze and even fewer outside influences from other players. This leads to understanding mechanics, which is as important if not necessary for understanding grand strategy. I think the algorithm’s result is particularly interesting because it flies in the face of several tenets: Don’t waste larva, and don’t overproduce food.
“I think the algorithm’s result is particularly interesting because it flies in the face of several tenets: Don’t waste larva, and don’t overproduce food.”
This is an important point. The reason why this algorithm is so significant is that it has created a build order that is so far outside of the box and so heavily separated from conventional Starcraft (and general RTS) wisdom that it would take a highly eccentric player to even conceive of this order, let alone actually perfect it. That is why the algorithm’s result is so significant. It is capable of examining raw data and casting aside our preconceived notions of what makes an effective build order and an efficient playstyle, and is capable of producing the outlandish order that prompted this whole conversation in the first place, and has fundamentally altered the way in which people view the early game in SC2.
Just to echo the above comments — yeah, this system is basically just an “economic” simulator that is given a set of rules (the building costs and production rates and pre-requisites) and a goal condition (produce this many roaches in the minimum time) and then produces a set of commands that it repetitively tweaks, keeping those that tend to produce faster results and pruning those that tend to be further from the goal condition.
So in this case, a human has decided that getting as many roaches as possible as quickly as possible is the linchpin of their strategy. The algorithm is a brute force method of finding a very good pathway to whatever condition the user decides, but it’s still very much up to the player to determine what position he wants to use that power to be in, and how to capitalize on it once he’s there. In a way, you could very well argue that it’s no different than having automatic path generation for move orders in Civilization 5.
I was pretty disappointed by the Starcraft discussion. It was pretty clear all along that neither of the panelists really understood what the build in question was, how it had affected the metagame, how the algorithm arrived to it, or even what was involved in doing it (at one point, one of you guys called the Extractor trick, which has been part of Starcraft 1 since 1998, an ‘exploit’ that would obviously be patched out).
I really like your podcast, and the level of discussion is usually really great (and really, more in depth than any other gaming podcast I can think of now that Idle Thumbs is gone), but this episode mostly ended up being misleading to anybody who isn’t a Starcraft II player, and kinda cringe-inducing to people who are. Obviously, it’s impossible for you guys to stay on top of every nuance of every strategy game, but I thought you were extrapolating from imperfect initial premises here, and I know you can do better.
ZZZZZZZZZZZZZZZZZZzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz…
yeah, this one was really painful.
This is why I love Memoir ’44 so much: no matter how great an “order” you have in a scenario, the cards and dice will screw it up. The luck in the game frustrates the type of people who want an orderly match of “I know this, so you lose.” Memoir’s card system actually does a great job of showing the frustrations of a commander who lacks perfect control of the battlefield and his units. Next week the open beta for Memoir ’44 online is opening, so you guys might be able to do a show on it in the future.
The whole build order thing is why I stopped playing RTS’s. I can’t stand that someone can beat me just because he has knowledge I can’t have until I have played as much as he does.
Visit the site
Also a bit disappointed in the level of discussion here. Overall, there was a profound lack of skepticism as to what this “genetic algorithm” means. Usually, there is more insight on TmA.
This is not a dominant strategy. The best evidence for that is simple: In professional tournaments, all the Zerg players (with large sums of cash on the line) do not follow this strategy. Not even a substantial minority does. That´s pretty much all you need to know.
I’ve complained about FoS’s coverage of Starcraft before. I haven’t had the opportunity to listen to this episode yet, so I can’t honestly judge it, but perhaps I shouldn’t bother.
I’m not a regular listener, but this weeks episode was relevant to the 2 activities I’ll be participating in today: a Bughouse tournament and a Starcraft 2 LAN party.
I think I would have quit playing SC out of frustration by now if I also had not been losing at Bughouse for the same reason: a poor series of moves in response to an opponents known opening, which can be quite varied by the nature of the game.
The point I’m walking away with is that a build order/opening demonstrates the application of general strategies. As they mentioned on the podcast, in a good strategy game a memorized build order will only get you so competitive unless the player can adapt it.
From that perspective, is “build order tyranny” a reasonable price to pay for getting into even the lowest level of play? I could go either way, depending on if I’m winning or losing.
In the end I decided to listen.
I don’t think very much of Troy’s fog of war point. It underestimates the value of scouting, and the fact that good players scout. In fact, if anything the fog of war point undermines the theory that a given build order is an “I win” button. The build order is a complete template to victory, but only of you execute it correctly, out scout the other player, etc., and they don’t counter it effectively.
Then there was the point about the youtube video made by, I think Rob? Not sure. Anyway, I watch a lot of SC games, and seeing someone do something and doing it yourself is TOTALLY different. Seeing that a strategy works is nothing like having a macro playing for you.
Then the conversation gets into the idea of exploits. Philosophically, when does something become an exploit? (Subpoint about testing, the game is organic, of course the developer does a lot of testing, but there’s no reason not to rely on fans to also break things. That’s the entire concept of beta testing. Ah, you guys get to that later.) I’d have liked to see the conversation zoom out to that broader question. For example, one of the major points of EVE’s design philosophy is that there’s no such thing AS an exploit. If you can do it it’s fair. Obviously that’s an extreme example, but is there a sweet spot that games should strive for.
In the case of Starcraft, a lot of people are upset about effective early game strategies that will get you up the the platinum level. Four-gate, roach rush, whatever. But – and TMA has touched on this in an earlier episode – Starcraft 2 is are really two different, and at the pro level it’s ENTIRELY different. None of those strategies work any more. So, are those easy builds exploits are are they just difficult to counter. Perhaps the definition of an exploit is a strategy that is dramatically easier to execute than counter?
Troy catches the point that small changes in in-game values make big differences, but says that it’s because people are invested in current cost-benefit analyzes.
I don’t really agree with that – the genius of Starcraft is that early game units never lose their relevance. Starcraft 1, that is. So small balance changes are big changes because of their ripples throughout the game. Starcraft was actively patched for balance for over three years, and Starcraft 2 has been out for three MONTHS. In my mind, the big question right now is if Starcraft 2 will live up to the legacy of Starcraft over a period of years. I don’t think we know the answer yet.
I love the point that there are build orders in games like Civ as well as Starcraft. I think that undermines some previous FoS coverage of Starcraft that criticized it for being built on build orders. If you think of a build order as a step by step guide to get a player from the null state to any specific initial goal then EVERY game has a build order. Even Counterstrike – which weapon do you load out with and which route do you take out of your starting area?
Now, how deep does a build order go? I think very shallow in Counterstrike and very deep in Starcraft 2, and maybe that’s one issue. The guys point out that you can follow a Civ build order for just six turns or for fifty. In Counterstrike, the “build order” I described takes seconds and leads directly to very rapid adjustments. Because Starcraft is a little slower, you can follow a build order for longer.
Ironically, among RTSs, Starcraft is one of the fastest. In Age of Empires II, I could follow a build order deep into the Castle Age before worrying about anything other than scouting and defending from worker harass. I think that Blizzard has tried to make Starcraft 2 more competitive at the top levels by making it less forgiving, but I don’t think that means it’s more dependent on build orders. There are many viable build orders but you have to execute them – as long as you don’t have idle production facilities for the first however many minutes, you’re probably doing OK. Troy has previously observed that Starcraft is more like an athletic competition and not like a sandbox game at all. That’s fine, but it is different – I think it’s a much more important point than anything about build orders.
This is rambly, but one last word: a response to Troy’s remark about being too far behind to bother with Starcraft anymore. I played two games last night for the first time in three or four weeks, and lost them both miserably. But in neither was I rolled over. I just focused on my base and keeping my troop production steady, and both times I won the first few encounters before eventually succumbing to superior multitasking.
I have no intention of getting better – I have many other things I enjoy doing more, but picking up a game here and there can still be fun.
Here’s a thought – FoS could assemble a list of people at low-medium skill level who want to play together. By playing with friends instead of strangers we can crit each other and even agree to play a slower game and experiment with later game units. Would there be interest in that?
And let me add, I’m really glad I listened. Snide comments aside, this was really thought-provoking. As evidenced by my ramble above.
I wonder if it would be cool if an RTS had a “setup” phase where one of two things would unfold, in either case:
A) There is a 2-5 minute grace period where players can basically play out their build orders without having to wait; building a unit costs resources and “time”, and you have to keep within certain limits before setup ends. Building workers spends time but nets more resources, buildings spend quite a bit of time, etc..
B) Each player is given a deck of cards, perhaps ones they have built before the match started, perhaps it’s a static deck that everyone uses. Each player picks a card in secret and when everyone is finished, all cards are presented. These cards represent opening builds and/or strategies which modify the starting environment to suit their choices. If a player draws a Boom card, they start with an expansion and plenty of workers and supply. A Rush card starts you with several powerful units but a weak economy, and Turtle gives you solid defenses but mediocre economy and no real offensive ability. Perhaps there are various flavors of boom which provide different sets of starting buildings / units. You could also make it competitive; you can play a Rush card on a player you think will Boom in order to nullify their bonus. However, if they play a Turtle card, it will nullify your Rush, and thus an RPS system emerges to “fast forward” the game to the interesting parts and improve pacing.
These might be stupid suggestions, and would probably not work well for Starcraft or a similar hardcore game, but I think it might be interesting to try.
ערימה של לינקים | המשחקיה // Nov 17, 2010 at 4:03 pm
[…] לזרג בטארקראפט 2. חשבתי להקילט על זה פרק, אבל ב" שלושה מהלכים קדימה" עשו את זה יותר […]
Uh yeah. Must echo the comments above. Really should read up on what this is before reporting on it. The build order in question is just a computer optimized order where a human player chose to build the most roaches in under 5 minutes. Any unit for any race could have been chosen. The extractor trick has been in Starcraft since 1998. The reason it does not always get used is that it trades lots of valuable early minerals and a slight gas delay for slightly delayed economy. It works because this build uses roaches. This build is easily scouted, does not scout itself, and has NO defense before the roaches pop.
The reason this build is noteworthy at all is not really because it was optimized, but rather the result that came up. It breaks the assumed rules of “always be building” and “never waste supply”. By not following assumptions this system found a more optimal solution.