Foreword
If you missed it, Lady Cadenzie took up the challenge, and played a Best of 5 series of games against the winner of last AIST tournament (AIST3), Locutus. The matches were very exciting, it was truly felt like a pitched battle! (She used the beta version of the SCHNAIL client) Warning, spoilers ahead!
(Here is the link to the original AfreecaTV broadcast)
Edit: You can watch the VOD of the matches here.
The interview
Hey! First of all, thanks for the showmatches! It was very interesting, and gave us a lot of things to think about! Did you know anything about AI in Brood War before we contacted you?
I have seen some stuff from ai’s doing weird things like microing individual mutalisks in really funny looking ways, not a while lot though.
What was your first impression? You played a few games vs. AI before the showmatches – what was your experience with them? Not talking about the skill of them.
Just like i saw before with the crazy individual unit control stuff, i sort of expected something like that in terms of a playstyle. It was really quite fun though to play against such a different style of play and its cool to see what AI can do.
And let’s do talk about the skill of them. Obviously they are not beating grandmasters yet, but they have come a long way.
Yeah, they dont quite have the perfect gameplan or ideal build orders, but they have pretty crazy use of max distance attacking on an actual inhuman level^^
How did you find the SCHNAIL client itself?
Except from the downloads that sometimes take a while without really knowing if its even still downloading or maybe its not working… then it was great yeah^^, nice starcraft ui style theme, simple and effective layout, very nice~! the custom hotkeys and just the page where you edit them itself is impressive.
Its really cool that you can change the language in the client(although you might want to make that work on the hotkeys tab:P)
On a scale of kittens to puppies, how cute do you think our mascot is?
I’m not much of a cat/dog person, but i really like reptiles so i’d say its around a baby gecko the size of my fingertip of cute^_^
Is there anything you would add, change, or remove to/from the client?
Maybe if there was some way to choose what kind of playstyle the bots will do as a sort of practice mode against where you can practice against standard play or cheese etc, and maybe some more settings like editing the auto replay saving path. If there was a way to choose what settings you want in-game from the client that would be amazing; sound settings and such.
And finally, let’s talk about the showmatches! I think the 3-2 score surprised everyone, I’m guessing even you. You did play an earlier version of Locutus before. My first question is, how was the whole experience?
It’s really cool to lose some games actually, I would maybe have been a bit bored of it was too easy.. turned out to be pretty cool. I did kinda know how dangerous the bots can be if they start to snowball with their value from perfect control though, so not really that unexpected to me actually.
How the did Locutus match your expectations? Did it surprise you with something (positive and negative surprises both)?
To be honest it was surprising how inefficient its builds were, and how it didn’t seem to really make use of scouting to save resources, also its sim cities weren’t ideal. Yet it can still do an amazing job regardless of how behind it gets itself just from the opening. If it handled the opening as well as a human of a reasonable skill level then I think it would be considerably stronger. Its amazing how much it can get done with 12 dragoons though.
Do you think you underestimated Locutus/took it easy at any time before, or during the matches?
Actually I took the first game a little seriously because I wasn’t sure how strong it was going to be after i underestimated some of the bots while messing around before. But after winning the first game with one of my signature builds quite easily I started to play more freely which may have been a mistake^_^
Do you still hate it? š
(Editor’s note: Previously on another stream, after losing a match)
All is forgiven, we are good friends now.
What is the most dangerous trick/tactic the bots used, or you think they could use vs. humans?
Low economy high micro intensive plays getting maximum value from control of units seems to be the best way for it to win.
It was my impression that during one of the matches, after busting the natural of Locutus, you seemed sure that you would win, and consequently, that cost you the game. Can you talk about that?
Yeah but I often do that in games against humans where I think its just over too and I just get a bit bored and suicide all my units, you might even say that’s my biggest signature move lmao.
All in all, how would you estimate the skill of your opponent? You can say an MMR, rank, amount of good boy points, or however you’d like š
I’d say the standard procedure, reactive scouting based play, builds, building placements and timings from the bots is very low like 1500mmr, but the individual unit control and multi-tasking is sometimes beyond human ability, so lets say at any point during the game its anywhere from 1200mmr to 3500mmr. its like playing a team melee against 4, no, maybe even 10+ players working together at some points. Sometimes though it’s like the 1500mmr player on the team is calling the shots.
Actually I played in a tournament before where there was a team melee relay style with a mix of progamers and beginner level players and they would take turns every 2 minutes, in a way it was most similar to that.
Example: It scouted that I was going hatchery first and still let its 2 cannons before nexus build, putting itself massively behind even with information. It would scout my build a lot with really good probe control but not really seem to react to any of it, so actually im not sure why it scouted^_^.
Related to this: What do you think would be an effective measuring tool for skill level of the bots? We are thinking a lot on this, it is a surprisingly difficult question. Since the humans playing the AI do not play each other, traditional ELO/MMR might not work well.
I don’t know that seems complicated… good luck, lol.
Do you think SCHNAIL could be used as an effective training tool and/or sparring partner for someone who is trying to get better at the game?
I think it has a lot of potential to be a really good training tool, especially for beginner players but actually for everyone. I know even progamers practice their build in an empty game or against the basic ai and just sort of imagine the timings of a normal player and train like that a bit before tournaments. If you could somehow choose what the ai is going to do or have them react to what youre doing in a “standard” way, or choose a bunch of cheeses you need to practice against, then it could be amazing for preparation.
If it just kind of abuses people with ridiculous dragoon control it might not be that great for newer players and might just be kind of frustrating and not really relevant for playing against humans who cant ever do that.
Will you use it as a training tool, or just to have fun? Maybe a warmup match?
Yeah it seems pretty useful, so I think I will. If there would be more options in future to set for what you want to practice it could be incredibly useful, the terran bot I played against at the end did fairly standard mech style just with kind of poor decision making and not ideal vulture control (kind of surprisingly), but actually I could feel that would be useful for practicing standard anti mech builds, especially if it was set to sometimes to 8min timing, sometimes 11min timing, sometimes fast 3rd, etc.
What kind of bot would you like to see developed in the future? (I know you haven’t played them all yet, but still)
I’d really like to see a bot that relies almost purely on scouting, adapting, and standard procedure knowledge based play at the highest level with a more human style micro^_^
Thanks for answering my questions, and for the showmatches! Any closing thoughts you’d like to add?
Thank you it was a really cool experience, and I just wish you good luck with it in future!
Closing thoughts
I really liked the matches – I rooted for Locutus, of course, but I – and the bot authors – learned a lot from the whole experience. A lot of work is ahead of us, that’s for sure, but I’m really happy about Cadenzie’s responses. We will take her feedback seriously, and continue to improve – I’m really hoping we can move to open beta as soon as possible.
Until then, stay tuned for more updates – and consider supporting the project on Patreon! Also, you can follow and stuff on Facebook, Twitter, YouTube, or Twitch!