It was 20 years ago when a stupid machine won against the Strongest Grandmaster in chess. Nobody believed it could happen. The computer was calculating way ahead and far many positions but still it was just a strong calculator and thats it. But it won. Same is happening now in game starcraft Few years ago even strongest starcraft bots could easily lose to amateurs and even noobs. Even today bots are still rule-based, which means they are extremely stupid once they encounter something they are not coded to respond. But... that started to change little bit by now. And thanks to improved makro, micro and other aspect of the game... bots started to win games against humans. And not just noobs. They have started to win against decent amateurs with pretty good understanding and experience of the game. Still human will remain the smartest and better for upcoming couple years at least.. but that does not mean bots are bad. They are coded by smart humans, dont forget that. Today top bots have decent opening, very good makro skills and good micro skills. They have thousands of APM. Reaction time is pretty much instant. Bot does not get tired or slow. He is everywhere at any time and can control many things at once. So human vs human tactics like make distraction with this unit so you attack with other unit over there does not really work. If the bot is coded to do this thing if tha happens it will do it. Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games... And i can tell you that... even tho bots have critical holes which can cost them the game pretty much instantly... they are getting stronger and stronger. If you are D- D D+ C- Iccup rank player or Under 1400 at europe server you will have heck a lot of hard time winning against the bot. I can say that because i have seen better players lose to bots. That does not mean bot skill is equal to human skills.. no no.. Bots could stil lose to D+ rank.. and beat say B- rank player. The gap is big and this topic could help us see where exactly today bots stand. If you are interested playing them and stream the games post a comment and we will do it.
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
well.. but AI is cheating, it needs no control device. its just controlling the game with its "brain". I like googles take on this way more, but its not convertible to the average user. At least i think it was google who want to build robot arms and let the machine only see whats on the screen. but then again, they chose SC2 if i'm correct.
Perfect micro and macro is irrelevant, that's hardly an "AI" thing. Of course you can expect that from a computer. What is interesting is of course if AI can start to match humans when it comes to strategy, tactics, deception etc.
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
I thought a lot of bots are doing apm caps no? Latency isn't exactly easy to quantify, since some human reflexes vs cognition/judgement happen on very different latencies. I would be more than happy to have a bot that can "instantly" storm dodge as long as it's apm limited.
The only context in which a bot could beat a [decent] human in broodwar is if it breaks the physical limitations of the game and micros' each unit a million times a second or something else absurd. Which, really, is not impressive from an artificial intelligence stand point, nor would it be fun to play against. So in short, yawn. Come at me when you can actually make an ai that engages in reasonable strategy.
To the bot, what's the difference between pressing F3 and clicking on the minimap at (345, 188)? For humans, using camera hotkeys is better than clicking on the minimap since it's faster and centers the camera perfectly while clicking on the minimaps is imprecise and is slower. But to the bot, it has perfect precision. It can click on the minimap wherever it wants. It effectively has infinite camera hotkeys.
It doesn't matter if you cap the bot's APM or force the bot to play on a physical keyboard and mouse. The bot will always have perfect execution and mechanics, and that alone will beat the top pros. The world's best BW bot could consistently beat top pros by just worker rushing with the starting 4 workers.
Here's another example. Tastosis mentioned that Flash counts drones in one of the ASL casts. Obviously, Flash doesn't count each individual drone in the mineral line, he just eyeballs it. Are there a small number of drones or a lot of drones? From there, Flash can determine what the zerg is doing. How many frames does Flash need to count the drones? 20? 30? 100?
How many frames does a bot need to count drones? Just 1. And it will know the exact number of drones. Capping APM or hardware won't stop the bot from being better at scouting than Flash.
What's more interesting in regards to AI is how the bot learns the strategies. What are the best machine learning models to teach the bot micro and macro? How does training the bot work? How will the bot respond to weird strategies and playstyles? The programming questions are more interesting to me at least than wondering whether or not a bot can perfectly dodge storms or use lockdown.
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
Currently bots are done to compete against other bots. The field is pretty difficult already so doing that will increase the difficulty for coding even more for lonely hobbyists. That will be done I'm sure by Deepmind in few years from now. But I don't remember deep blue being limited to only 100 positions per second vs Gary Kasparov or AlphaGo being limited vs the Go champion. Why should they do that in starcraft and how is not fair? If you are playing vs human and he has 6 finger then is not fair... What's next.. Oh I lost to human who plays the game 5 years me only 2 that's not fair.. I heard a comment where bot should have play with mouse and see only part of screen just like human... Really.. Does the plane fly like a bird or does submarine really swim?
On December 11 2018 23:55 odeSSa wrote: Perfect micro and macro is irrelevant, that's hardly an "AI" thing. Of course you can expect that from a computer. What is interesting is of course if AI can start to match humans when it comes to strategy, tactics, deception etc.
Well no much inteligence from rule based bot, but still... On the clip above Saida did tactical tank push in the intro, also cherry pi a Facebook bot has pretty good zerglings surrounding enemy units. Bots has some advanced a. I technics is not just do that if this.. More like do analyse of the situation, do calculation, take decision based on this and that etc... Deep stuff. But what you are looking for is a some kind of Machine learning, reinforcement learning/deep learning..neural nets and etc.. Self learning.. That is still to come, but Saida I think will have some Machine learning stuff inside next year, also cherry pi has been trained on human Replays for building placement I heard. And not sure what do you mean... Even 2 years ago bot can do reaver drop. The shuffle needs to find safe path to enemy base, find mineral line, drop reaver if there is no turrets in range or sieged tanks or any enemy unit that can kill it instantly.. To avoid wraiths etc.. Then that reaver has to find the best target for splash damage.. Should be pack of scv or pack of Marines.. Etc.. Then needs to pick reaver again, do shuffle micro avoid obstacles, find new target, and so on so on... Isn't that Impressive???
On December 12 2018 00:27 Dazed. wrote: The only context in which a bot could beat a [decent] human in broodwar is if it breaks the physical limitations of the game and micros' each unit a million times a second or something else absurd. Which, really, is not impressive from an artificial intelligence stand point, nor would it be fun to play against. So in short, yawn. Come at me when you can actually make an ai that engages in reasonable strategy.
What do you mean by reasonable strategy? Correct decision making in strategy game by bot will be very impressive. A bot does not have memory yet like human, which knows things by past experience and with one look to instantly know is it good or bad time to engage. Aka patterns. That's why bot needs to calculate every second. But to calculate correctly on a game with fog of war, unit change, position change, angle change of units and who knows what more it's very difficult task for computer. On top of that to know should it attack this expo or that expo or retreat units to defend his base? Or should he sac his units to slow down opponent because bot have better economy? Very difficult, even for super computers of Google. That's why when we see such things done correctly by luck or not by bot is very impressive. Anyway... What should we call artificial inteligence? They call Google image recognition smart and intelligent... But is it really? Just recently the system recognises a dog instead of a perfect looking wolf.. Because knows its unlikely to be wolf if there is no snow in picture... Extremely stupid reason..
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
Currently bots are done to compete against other bots. The field is pretty difficult already so doing that will increase the difficulty for coding even more for lonely hobbyists. That will be done I'm sure by Deepmind in few years from now. But I don't remember deep blue being limited to only 100 positions per second vs Gary Kasparov or AlphaGo being limited vs the Go champion. Why should they do that in starcraft and how is not fair? If you are playing vs human and he has 6 finger then is not fair... What's next.. Oh I lost to human who plays the game 5 years me only 2 that's not fair.. I heard a comment where bot should have play with mouse and see only part of screen just like human... Really.. Does the plane fly like a bird or does submarine really swim?
Well, if were going to compete against one another, and if there is supposed to be some basis of comparison at all, starcraft is a physical sport, so expect demands on physical limitations, or simple disinterest from the community. Why would it be any other way? It's a bit like creating an A.I runner who competes in marathons against humans, but its powered by a car like engine and on wheels. Wtf is going on?
On December 12 2018 00:27 Dazed. wrote: The only context in which a bot could beat a [decent] human in broodwar is if it breaks the physical limitations of the game and micros' each unit a million times a second or something else absurd. Which, really, is not impressive from an artificial intelligence stand point, nor would it be fun to play against. So in short, yawn. Come at me when you can actually make an ai that engages in reasonable strategy.
What do you mean by reasonable strategy? Correct decision making in strategy game by bot will be very impressive. .
Correct choices based on the information presented. I.e something that would actually work if it were not for the fact that the computer is using 400k apm to offset its otherwise idiotic choices. And your right, it would be impressive if the ai made the correct decisions: Just like it isnt impressive if an ai plays well by abusing the mechanics of bw absent strategically correct decisions. Which, atm, is what bots currently are in bw.
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
It will be cool when they can build a bot that can win with these conditions.
On December 12 2018 02:01 Dazed. wrote: Well, if were going to compete against one another, and if there is supposed to be some basis of comparison at all, starcraft is a physical sport, so expect demands on physical limitations, or simple disinterest from the community. Why would it be any other way? It's a bit like creating an A.I runner who competes in marathons against humans, but its powered by a car like engine and on wheels. Wtf is going on?
Starcraft isn't physical, it works on function calls. That humans deliver them mechanically through keyboard/mouse combinations which lead to physical fatigue doesn't matter to the engine. So its not really anything like your robot car/runner analogy.
Here's the basis of comparison: who wins the game? When the AI starts winning then it's better at the game.
It's really great to see how far these bots have come! Honestly, if you think it's trivial to make a bot play this well, try programming one yourself. This is incredible progress.
To those who are asking why an AI should be limited in APM or lag, a bad analogy would be tennis. In a competition against a human, a tennis ball launched at 200 m/s by an air gun with a camera to aim anywhere that wouldn't injure Roger Federer on contact with said tennis ball, that machine will always take the point overr Roger Federer (or whoever the human is) every time. Is that exciting? Does that truly fulfil the strategic sport aspect of tennis? It does not.
On December 12 2018 03:33 Dangermousecatdog wrote: To those who are asking why an AI should be limited in APM or lag, a bad analogy would be tennis. In a competition against a human, a tennis ball launched at 200 m/s by an air gun with a camera to aim anywhere that wouldn't injure Roger Federer on contact with said tennis ball, that machine will always take the point overr Roger Federer (or whoever the human is) every time. Is that exciting? Does that truly fulfil the strategic sport aspect of tennis? It does not.
Watching a humanoid robot that can serve a tennis ball at 200 m/s against a human would be boring indeed; such a robot would be much better at tennis than the best human. However, I would be interested in watching a tournament of such robots.
On December 12 2018 03:28 L3gendary wrote: How does one go about playing against a good bot? Is there a repo or something I can download to play one offline?
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
Currently bots are done to compete against other bots. The field is pretty difficult already so doing that will increase the difficulty for coding even more for lonely hobbyists. That will be done I'm sure by Deepmind in few years from now. But I don't remember deep blue being limited to only 100 positions per second vs Gary Kasparov or AlphaGo being limited vs the Go champion. Why should they do that in starcraft and how is not fair? If you are playing vs human and he has 6 finger then is not fair... What's next.. Oh I lost to human who plays the game 5 years me only 2 that's not fair.. I heard a comment where bot should have play with mouse and see only part of screen just like human... Really.. Does the plane fly like a bird or does submarine really swim?
Well, if were going to compete against one another, and if there is supposed to be some basis of comparison at all, starcraft is a physical sport, so expect demands on physical limitations, or simple disinterest from the community. Why would it be any other way? It's a bit like creating an A.I runner who competes in marathons against humans, but its powered by a car like engine and on wheels. Wtf is going on?
On December 12 2018 00:27 Dazed. wrote: The only context in which a bot could beat a [decent] human in broodwar is if it breaks the physical limitations of the game and micros' each unit a million times a second or something else absurd. Which, really, is not impressive from an artificial intelligence stand point, nor would it be fun to play against. So in short, yawn. Come at me when you can actually make an ai that engages in reasonable strategy.
What do you mean by reasonable strategy? Correct decision making in strategy game by bot will be very impressive. .
Correct choices based on the information presented. I.e something that would actually work if it were not for the fact that the computer is using 400k apm to offset its otherwise idiotic choices. And your right, it would be impressive if the ai made the correct decisions: Just like it isnt impressive if an ai plays well by abusing the mechanics of bw absent strategically correct decisions. Which, atm, is what bots currently are in bw.
Wrong comparison. This is strategy game. Apm/speed won't matter as much as how smart you play. This is the goal after all. If the human is smarter than a. I. It will outsmart it and win. Besides pro level humans are fast enough so they can play perfect timing attack if they want. Is not like they need more apm to compete vs bot. Just like chess, but sc even better chance to show your brilliance. And wrong again. Apm does not save the bots. They don't really compensate bad play with huge apm. Let the bot have his 20k apm. It won't save him if his decisions are bad. And wrong again. Bots does not do anything out of normal. They use strategy and they adapt during the game and even for next game. You have zero clue about the bots and very wrong assumptions. Download and try one bot bellow. They are noting close to cheating build in ai in custom maps created by some suspicious hackers.
On December 12 2018 03:28 L3gendary wrote: How does one go about playing against a good bot? Is there a repo or something I can download to play one offline?
1. Download this and extract it https://github.com/adakitesystems/DropLauncher 2. Download a bot from sscait website, I'll send you one of best protoss https://sscaitournament.com/bot_binary.php?bot=BananaBrain 3. Download his Bwapi file https://sscaitournament.com/bot_binary.php?bot=BananaBrain&bwapi_dll=true 4. Launch the launcher and load the bot and his Bwapi file. Choose protoss as his race. 5 start broodwar in LAN UDP mode and choose mellee game mode on FS map or python or benzene etc.. Circuit breaker... 6. Press join on DropLauncher and the bot shout join and press start. 7. Don't forget to record your game and stream it or upload on YouTube and type link here Don't rush bot like 5pool as they are not very good at defending it. Don't judge bot by only one or two games, they have different openings and Sleightly adapt on your play. This is rule based bots, so half the things he does might look weird or stupid, since it does not have mind of his own but it plays absolutely fair and tries to follow decent game plan and strategy. Bots are made with competition against other bots so tricks by humans most definetely will work.
Here's another example. Tastosis mentioned that Flash counts drones in one of the ASL casts. Obviously, Flash doesn't count each individual drone in the mineral line, he just eyeballs it. Are there a small number of drones or a lot of drones? From there, Flash can determine what the zerg is doing. How many frames does Flash need to count the drones? 20? 30? 100?
Just a detail, but I think Flash and other pros actually do count (almost) every worker in certain situations, like versus a 2hatch-muta build or something to determine how all-in an opponent is. Or during early scouting, to sniff out hidden proxies in TvP for example, a lack of 2-3 probes can give that away. Obviously not in the mid- to lategame when the workers are overlapping or when you don't really need the exact number but, as you wrote, eyeballing is enough.
In ZvZ it's very common to count every drone, admittingly there aren't that many in a lot of situations though.
On December 12 2018 03:33 Dangermousecatdog wrote: To those who are asking why an AI should be limited in APM or lag, a bad analogy would be tennis. In a competition against a human, a tennis ball launched at 200 m/s by an air gun with a camera to aim anywhere that wouldn't injure Roger Federer on contact with said tennis ball, that machine will always take the point overr Roger Federer (or whoever the human is) every time. Is that exciting? Does that truly fulfil the strategic sport aspect of tennis? It does not.
Again this is strategy game. In Zvz an skilled amateur with 180 apm can beat a pro player with 400 apm by simply making 4 lings more in simular build or hiding them and do surprise attack. Anything can happen. APM is pointless. When the bots wins a game vs pro human most cases will be human fails to outsmart the bot of bot launches tricky attack, it won't be because one player was so slow he had 20 units less in the battle that cost him the game because low apm. . In order this to happen the bot should demonstrate outstanding skills in other areas other than makro and micro and to survive first several minutes which are critical. . In near future pro vs bot won't be long games like 20 30 minutes where human gets tired and bot is taking whole map and controls many units etc. No. All games will end in first 5-13 minutes. Where is most critical part and humans can demo nstrate brain.
On December 12 2018 03:33 Dangermousecatdog wrote: To those who are asking why an AI should be limited in APM or lag, a bad analogy would be tennis. In a competition against a human, a tennis ball launched at 200 m/s by an air gun with a camera to aim anywhere that wouldn't injure Roger Federer on contact with said tennis ball, that machine will always take the point overr Roger Federer (or whoever the human is) every time. Is that exciting? Does that truly fulfil the strategic sport aspect of tennis? It does not.
Again this is strategy game. In Zvz an skilled amateur with 180 apm can beat a pro player with 400 apm by simply making 4 lings more in simular build or hiding them and do surprise attack. APM is pointless. When the bots wins a game vs pro human most cases will be human fails to outsmart the bot of bot launches tricky attack, it won't be because one player was so slow he had 20 units less in the battle that cost him the game. In order this to happen the bot should demonstrate outstanding skills in other areas other than makro and micro.
On December 12 2018 03:00 Jealous wrote: I will play your bots on stream. I played LetaBot a few years ago (as well as a few others) and found it to be fairly interesting, sometimes.
Also, JFL @ opening the discussion by comparing a Grandmaster chess player to a sub 1400 MMR player in Europe.
On December 12 2018 03:00 Jealous wrote: I will play your bots on stream. I played LetaBot a few years ago (as well as a few others) and found it to be fairly interesting, sometimes.
Also, JFL @ opening the discussion by comparing a Grandmaster chess player to a sub 1400 MMR player in Europe.
On December 12 2018 03:33 Dangermousecatdog wrote: To those who are asking why an AI should be limited in APM or lag, a bad analogy would be tennis. In a competition against a human, a tennis ball launched at 200 m/s by an air gun with a camera to aim anywhere that wouldn't injure Roger Federer on contact with said tennis ball, that machine will always take the point overr Roger Federer (or whoever the human is) every time. Is that exciting? Does that truly fulfil the strategic sport aspect of tennis? It does not.
Again this is strategy game. In Zvz an skilled amateur with 180 apm can beat a pro player with 400 apm by simply making 4 lings more in simular build or hiding them and do surprise attack. Anything can happen. APM is pointless. When the bots wins a game vs pro human most cases will be human fails to outsmart the bot of bot launches tricky attack, it won't be because one player was so slow he had 20 units less in the battle that cost him the game because low apm. . In order this to happen the bot should demonstrate outstanding skills in other areas other than makro and micro and to survive first several minutes which are critical. . In near future pro vs bot won't be long games like 20 30 minutes where human gets tired and bot is taking whole map and controls many units etc. No. All games will end in first 5-13 minutes. Where is most critical part and humans can demo nstrate brain.
Modesty I love you man but I don't think I agree. Have you seen the micro these bots can do? Like just the muta micro alone....a human player would be decimated...decimated utterly by a 2 port wraith or 2 hatch muta with each muta individually microed. Same in battled. There are some vids out there of bot going vs 12 goons with 12 goons and it takes zero losses. Not to mention the multitask. Think Bisu PvZ but unfathomably faster.
That's a crazy, crazy advantage. I mean if the bot is dumb as rocks it won't matter...but they will get better. And I'm not sure why all games end 5-13 minutes...the bots will get smarter than the players. Bots will eventually get better there as well.
Here's another example. Tastosis mentioned that Flash counts drones in one of the ASL casts. Obviously, Flash doesn't count each individual drone in the mineral line, he just eyeballs it. Are there a small number of drones or a lot of drones? From there, Flash can determine what the zerg is doing. How many frames does Flash need to count the drones? 20? 30? 100?
Just a detail, but I think Flash and other pros actually do count (almost) every worker in certain situations, like versus a 2hatch-muta build or something to determine how all-in an opponent is. Or during early scouting, to sniff out hidden proxies in TvP for example, a lack of 2-3 probes can give that away. Obviously not in the mid- to lategame when the workers are overlapping or when you don't really need the exact number but, as you wrote, eyeballing is enough.
In ZvZ it's very common to count every drone, admittingly there aren't that many in a lot of situations though.
TIL. I thought Tastosis meant eyeballing because counting each individual drone is beyond the capacity of this D rank Terran. Time to fall down the card/worker counting hole.
Can i pls get a link to the best Z (Microwave? looks very weak tho) and T (krasi0?/SAIDA already dled) bots? thx!
BTW my antivirus won't let me extract DropLauncher-0.5.24.a.zip t_t. Would be nice if we could get a new SCR tutorial on how to play vs the bots, i'm not tech savvy at all :D. I'd love to practice builds against them.
well your AV probably intervening bc those launcher do process injection, a technique that is also used by malicious software. If you have it from official source add a exception to your AV.
I’m currently at work with no access to YT (L), but I strongly suggest you go check some AI Micro bots on StarCraft 2. Some people made some amazing tank/dropship micro bots that can shred a full army in no time while microing 10+ dropships/tanks simultaneously. I think it’s pretty much self-explanatory after that that you need to implement APM limitation in AI VS Human games, else the impact is just too great.
Using that kind of micro bot, you can win if you don’t let the bot get one tank/dropship. If it reach 1+ of that combo, you just CANNOT win anymore, that is just not possible. A dumb bot with no macro script, no intelligence, nothing basically, can outclass any pro player just with 1 dropship/tank, let alone a full army of that fully microed…
To remind some details on the Chess/Go bots : the fundamental difference between Chess and Go is the size of the board. It is precisely what made AlphaGo’s achievement incredible. In chess, a computer can just “brute force” calculate the entirety of the position and then extrapolate to other combination and then find the best one on a given situation.
After the win over Kasparov, the next game decided to improve bots was Go, for a simple reason: the board is waay larger than chess. Which makes these “brute force” calculation impossible, because even a top computer have not enough calculation power to do what was done in chess. So, in order to win against human, the bot needed to be more than just a calculating machine, which is where machine learning intervene.
The tennis analogy was a good one I think, let’s put it this way. There is a humanoid bot playing against a human. No super service, no super anything. Just a superior calculation power. The ball gets to the human player, who hits it. From the time the ball left the human’s racket, the bot has already calculated, given the angle of the racket, the position of the ball on the racket, the energy put into the hit, … WHERE the ball is gonna end on his side. So the ball has not even left the racket of the human player, the bot has already started to move where the ball is gonna be. So basically, the bot cannot ever be misplaced, so he cannot ever miss a ball. So he basically cannot lose a point.
I fail to see the fun in watching this, let alone the fairness of that kind of “competition”.
Soooooo, to sum this up, this is why you can’t overlook APM regulation for Human VS AI competition. I don’t want to minimize the work you or other bots designers are putting into it, or the achievements made. I am just expressing why I wouldn’t be interested in watching AI VS Human on StarCraft if there are no clear “rules” implemented which limit the calculation and sheer mechanical advantage the bot has over the human player.
Are there any kind of apm limitations placed on the bots? If not it would be pretty unfair as the bots would be capable of actions that humans can't physically replicate.
On December 12 2018 15:11 TT1 wrote: Can i pls get a link to the best Z (Microwave? looks very weak tho) and T (krasi0?/SAIDA already dled) bots? thx!
BTW my antivirus won't let me extract DropLauncher-0.5.24.a.zip t_t. Would be nice if we could get a new SCR tutorial on how to play vs the bots, i'm not tech savvy at all :D. I'd love to practice builds against them.
If you click on the name of the bots in this page, you will go to a page where you can download the DLL/EXE of that bot:
I played these over a year ago and I guess they've improved a lot, I'm a D player but I could easily detect exploitable patterns after 4 games. If they could make it that the routines adapted more then it would be great fun for ppl like myself
On December 12 2018 03:33 Dangermousecatdog wrote: To those who are asking why an AI should be limited in APM or lag, a bad analogy would be tennis. In a competition against a human, a tennis ball launched at 200 m/s by an air gun with a camera to aim anywhere that wouldn't injure Roger Federer on contact with said tennis ball, that machine will always take the point overr Roger Federer (or whoever the human is) every time. Is that exciting? Does that truly fulfil the strategic sport aspect of tennis? It does not.
Again this is strategy game. In Zvz an skilled amateur with 180 apm can beat a pro player with 400 apm by simply making 4 lings more in simular build or hiding them and do surprise attack. Anything can happen. APM is pointless. When the bots wins a game vs pro human most cases will be human fails to outsmart the bot of bot launches tricky attack, it won't be because one player was so slow he had 20 units less in the battle that cost him the game because low apm. . In order this to happen the bot should demonstrate outstanding skills in other areas other than makro and micro and to survive first several minutes which are critical. . In near future pro vs bot won't be long games like 20 30 minutes where human gets tired and bot is taking whole map and controls many units etc. No. All games will end in first 5-13 minutes. Where is most critical part and humans can demo nstrate brain.
Modesty I love you man but I don't think I agree. Have you seen the micro these bots can do? Like just the muta micro alone....a human player would be decimated...decimated utterly by a 2 port wraith or 2 hatch muta with each muta individually microed. Same in battled. There are some vids out there of bot going vs 12 goons with 12 goons and it takes zero losses. Not to mention the multitask. Think Bisu PvZ but unfathomably faster.
That's a crazy, crazy advantage. I mean if the bot is dumb as rocks it won't matter...but they will get better. And I'm not sure why all games end 5-13 minutes...the bots will get smarter than the players. Bots will eventually get better there as well.
A. I. Mutalisk micro is still terrible. There is a. I. bot called Marian Devechka that does human like mutalisk micro, stacking with overlord which is much improvement of muta micro all other bots do which is horrible separate control. So that micro of the bot is visually looking correct but decision making where to attack, which unit, when to attack, when to retreat, how much to retreat, should I go to enemy base or expansion, what to target there and when to leave, should I attack enemy Marines on center of map or should I wait for regroup or should I protect my natural or my 3rd expo.. Should I be making more muta or lings or lurkers or taking another expo, or placing makro hatch or going for hive... And so on and so on.. All that bot HAS zero to little Idea... All the apm in the world won't save the bot unless opponent does very weak micro or makro or really stupid decision. By the time bot improves on all that which might take 5 years.. Then lowering the apm really won't matter at all bot will be already super human. Now with all that benefit bot is still struggles but is fun to test him and play vs humans. Because you don't need 300 apm or 3000 to be equal to bot. Now if 2 stupid bots play and one has 15 000 apm other has 150, that's unfair. Human vs bot whole goal and idea is to see how "smart" bot is.. don't worry how fast makro micro he does. There is locutus called protoss bot which does fantastic goon micro. It beats every other bot in goon vs goons even if other bot has reavers and wins battle, that locutus has fantastic makro as well and pimps goons from 16 gates, becouse it's faster than other bots taking expo and stuff. But you know what. All that is pointless vs human. A human on 2 bases terran turtles and has 2 000 minerals because is slow and bad on makro while the bot is already on 5 bases and upgrades on with bigger army and close to 200/200.. Teran goes out.. The bot constantly loses couple units while trying to check and see if he can win the battle which he can't so he retreats.. The bot does not understand that enemy was sitting on 2 bases whole time so he need to sacrifice army even if he loses it all, because it has superior economy and will create same army in 30 seconds... Instead he retreats to his main bases and loses units without fighting... Then the whole game.
On December 12 2018 15:11 TT1 wrote: Can i pls get a link to the best Z (Microwave? looks very weak tho) and T (krasi0?/SAIDA already dled) bots? thx!
BTW my antivirus won't let me extract DropLauncher-0.5.24.a.zip t_t. Would be nice if we could get a new SCR tutorial on how to play vs the bots, i'm not tech savvy at all :D. I'd love to practice builds against them.
Wait for 20 December. Then the best zerg will be updated Facebook bot Cherry Pi. It has improved a lot since 1 year. And other zerg bots will be improved as well. Now they play really weak if you are experienced sc player. No high rank, just experienced. Currently best zerg is Marian Devechka but has 0.. Again zero overlord safety and suicides like first 3 overlords and gets really behind. then has major flaws like he goes for muta but as soon sees 1 valkyrue or corsairs switches to unupgraded hydra and abandons his whole powerful mutas. Instead of making scourge... Hopefully all that will be fixed on 20 December as well improvements on his muta micro.
On December 13 2018 12:01 Broodwar4lyf wrote: I played these over a year ago and I guess they've improved a lot, I'm a D player but I could easily detect exploitable patterns after 4 games. If they could make it that the routines adapted more then it would be great fun for ppl like myself
You will be up for a candy. In last 10 months was massive improvements and new bots showed up and they all be updated and improved on 20 December. But If your rank is max D D plus, then I'm afraid you will have serious difficulties l vs top bots, but do try and record video.
On December 13 2018 04:28 Dante08 wrote: Are there any kind of apm limitations placed on the bots? If not it would be pretty unfair as the bots would be capable of actions that humans can't physically replicate.
Not really sure on that but Well if the bot has 300-500 apm max it won even beat D minus human... Because it needs to constantly check everything.. Human can see whole picture and plan ahead. A bot does a action for couple seconds... Then he starts from zero and need to do that same action again.. You know what I mean? Other example human is zerg and does 9 drones and overlord. By the time overlord finishes human does not need to make single click with mouse or hitting any key on keyboard. The bot on other hand I think will spam that apm to 5 000 because its how is made to work. Not sure completely on that but you get the idea. Don't be afraid of apm. It's like a sprint marathon. You are slower, bot is faster, but you take clean short cut while the bot takes longer path with many obsticles. Show that you are smarter and better so coders to know what to fix and improve. Now if you have 50-100 apm.. I'm afraid it's too late for you and top bots might already passed you by 10 miles. Your only chance is 4pool rush or bunker rush something like that and might win as this is the weakest side of bots, defending super early attack. But they have improved on that also.
Someone asked me. That does not make any sense. A bot should already know by long ago how to stop such early attack since there are so little units and nothing smart really. Simple answer will be analogy to chess. Endgame at chess has very few pieces on board. The less pieces/units the less to calculate right? Well.. Wrong. The room for the pieces to maneuver is much greater. Also Thing is the cbance of mistake is actually greater and a single mistake is valued huuge. You don't care if you lose 2 lings/workers at 10 minute mark. But at 2min mark might be game over you see..
On December 11 2018 22:14 Modesty00 wrote: Example storm dodging. Even pro humans lose units to storm in PvZ. But they can predict when other human could storm. But a bot can storm dodge instantly. Bot can seperate irradieted mutalisk from the stack instantly. They are better than pro humans in that scenario. You see where im going. Little by little bot do things better than even pro humans. And now is pretty much the time to test and see how much those bot things can help the bot to win games...
If you make a "fair" bot, restrict it with human limitations like response time of ~150ms and max 400 apm, it is not going to do that well
It is a difference in perception of what starcraft is.
If you see it as a strategy game that is about strategy, it makes no sense.
If you see it as a sport that is at least partially based on physical performance, and the physical control of things is the interesting thing, than it makes sense.
I tend to be on the first side of the argument, i think strategy is more interesting than execution.
Yeah, I think if get to the point where AI can become really strong, the interesting part will be how clever it can get rather than how perfectly i can micro it's units by optimat targeting, distance feeling etc.. We know that it will be easiest for AI to beat humans on the mechanical side, as SC is just so overcomplex in that matter.
i would actually be interested to play vs AI with insane micro lol would be a fun challenge to try to out-cheese and out-macro the metal fucker knowing that your units are x times less efficient
On December 12 2018 03:28 L3gendary wrote: How does one go about playing against a good bot? Is there a repo or something I can download to play one offline?
1. Download this and extract it https://github.com/adakitesystems/DropLauncher 2. Download a bot from sscait website, I'll send you one of best protoss https://sscaitournament.com/bot_binary.php?bot=BananaBrain 3. Download his Bwapi file https://sscaitournament.com/bot_binary.php?bot=BananaBrain&bwapi_dll=true 4. Launch the launcher and load the bot and his Bwapi file. Choose protoss as his race. 5 start broodwar in LAN UDP mode and choose mellee game mode on FS map or python or benzene etc.. Circuit breaker... 6. Press join on DropLauncher and the bot shout join and press start. 7. Don't forget to record your game and stream it or upload on YouTube and type link here Don't rush bot like 5pool as they are not very good at defending it. Don't judge bot by only one or two games, they have different openings and Sleightly adapt on your play. This is rule based bots, so half the things he does might look weird or stupid, since it does not have mind of his own but it plays absolutely fair and tries to follow decent game plan and strategy. Bots are made with competition against other bots so tricks by humans most definetely will work.
Thanks. I don't see an executable file for the launcher so how do I launch it? edit: nvm I had downloaded the repo instead of the version 0.5.24a below in the link
any easy way to install these new bots now? It used to be a pain in the ass and not every bot out there worked for me no matter what I tried (i dont like using the old API)
Yes it's very easy just follow my steps quoted above. Nothing to install. The war is heating up. Less than 36h before the Anual tournament starts. Coders are updating their bots every couple hours with latest fixes and improvements. Some have been waiting and hiding their work for more than months and now will shock everybody with their powerful bot (locutus/cherrypi/saida) . You can watch at twitch or YouTube channel of Sscait. Top updated bots are playing better than ever. Still some weird things but huge improvement over year ago. I will post here few bots that are excellent for human bots play, after their final update tomorrow. You will witness huge makro zerg bots with great build opening, never supply blocked (atlough overlord safety needs work) tons of units, good mutalisk Pro style micro (stacked) while makroing non stop etc, Bots with excellent goon micro, proxy gate zeilot rushes, Teran bot that will stop almost any kind of cheese (maybe) and then will make timing push and beat you.. and so on so on..
There was a tournament of starcraft enthusiasts and progamers (I remember Stork participating) who faced different AIs. It was held about an year ago (in Korea obv). I believe the AIs had APM and something else restricted to a given value.
On December 19 2018 23:10 BlueStar wrote: There was a tournament of starcraft enthusiasts and progamers (I remember Stork participating) who faced different AIs. It was held about an year ago (in Korea obv). I believe the AIs had APM and something else restricted to a given value.
AI revolutionized how chess is played by chestmasters today, and it'll revolutionize Boor War as soon as bots become stronger than the top gamers, this will be fascinating.
By learning by playing against itself millions of times, with no reference to how humans play the game, it'll come up with all sorts of crazy builds and strategies that no one has ever came up with. Maybe M&M is actually the best unit composition in TvP when you have 10 000 APM. Maybe scouting at 6 then double FE is an undiscovered good build in ZvT. And pro gamers will start to adopt those strategies into their own plays.
On December 20 2018 01:24 lepape wrote: AI revolutionized how chess is played by chestmasters today, and it'll revolutionize Boor War as soon as bots become stronger than the top gamers, this will be fascinating.
By learning by playing against itself millions of times, with no reference to how humans play the game, it'll come up with all sorts of crazy builds and strategies that no one has ever came up with. Maybe M&M is actually the best unit composition in TvP when you have 10 000 APM. Maybe scouting at 6 then double FE is an undiscovered good build in ZvT. And pro gamers will start to adopt those strategies into their own plays.
I'm pretty sure that's not how these AI work. They don't learn by playing themselves and they absolutely do have reference from how humans play the game, as far as I know.
i dont care what anyone says, as long as bot capabilities are restricted to match humans (apm etc), they arent going to beat pros for a long time. starcraft isnt just about micro and macro executions. you have to make decisions with imperfect information; most of these decisions being calculated risks and gambles based on what you THINK your opponent might do/be doing. this is where ai will fail. you cannot teach a computer to understand the tendencies and traits of your opponent. what flash is able to do comes from not just having superior understanding of the game, but understanding who his opponent is. what jd might do in any given situation will be different from what larva might want to do in the exact same scenario.
the above is the same reason why bots will not beat pros in dota also. not for the forseeable future at least.
On December 20 2018 01:24 lepape wrote: AI revolutionized how chess is played by chestmasters today, and it'll revolutionize Boor War as soon as bots become stronger than the top gamers, this will be fascinating.
By learning by playing against itself millions of times, with no reference to how humans play the game, it'll come up with all sorts of crazy builds and strategies that no one has ever came up with. Maybe M&M is actually the best unit composition in TvP when you have 10 000 APM. Maybe scouting at 6 then double FE is an undiscovered good build in ZvT. And pro gamers will start to adopt those strategies into their own plays.
I'm pretty sure that's not how these AI work. They don't learn by playing themselves and they absolutely do have reference from how humans play the game, as far as I know.
Sc1 bots no, but deepmind by Google already made progress and came up with its own strategies like probe rushing the build in a. I. And winning. And perhaps after they concur sc2, they will drop a bot for sc1 as well. But current rule based sc1 bots do have some learning implemented like dynamic opening build, some have over 10 openings vs single race. Also cherrypi has Machine learning things inside like building placement that was trained on human Replays. Next year saida might have reinforcement learning so yeah.. We will wait and see, meanwhile why don't you guys proof humans still better?
On December 20 2018 01:24 lepape wrote: AI revolutionized how chess is played by chestmasters today, and it'll revolutionize Boor War as soon as bots become stronger than the top gamers, this will be fascinating.
By learning by playing against itself millions of times, with no reference to how humans play the game, it'll come up with all sorts of crazy builds and strategies that no one has ever came up with. Maybe M&M is actually the best unit composition in TvP when you have 10 000 APM. Maybe scouting at 6 then double FE is an undiscovered good build in ZvT. And pro gamers will start to adopt those strategies into their own plays.
I'm pretty sure that's not how these AI work. They don't learn by playing themselves and they absolutely do have reference from how humans play the game, as far as I know.
Sc1 bots no, but deepmind by Google already made progress and came up with its own strategies like probe rushing the build in a. I. And winning. And perhaps after they concur sc2, they will drop a bot for sc1 as well. But current rule based sc1 bots do have some learning implemented like dynamic opening build, some have over 10 openings vs single race. Also cherrypi has Machine learning things inside like building placement that was trained on human Replays. Next year saida might have reinforcement learning so yeah.. We will wait and see, meanwhile why don't you guys proof humans still better?
Like I said before, I was expecting you to be hosting this bot on ICCup. I imagine it can't be that hard to have the bot autohost a game on ICCup, start the game when an opponent is inside, play, then host again after game is done. Maybe not inside the bot itself, but as a tertiary script or something. Gameranger does it quite easily and that is ancient technology.
The more barriers of entry you place between players and AI, the fewer people will care enough to make the leap. Playing on 1.16 is already practically unheard of for the majority of the foreign Brood War population. A few of us still have it installed from well over a year ago, which is why I would be willing to log on to play a game or two on it. However, having to download a bunch of stuff to run it on my own is just too much of a hassle for too little benefit.
I know you've been saying that AI have been improving, yet you still put the challenge out to D and C players. D players (and a few games at C-/C level) were already losing to AI over a year ago:
So if you're still putting out the challenge to the same level players, it sort of sends the message that not nearly that much has changed.
For example, in one of the games I played against a bot, I simply placed a Pylon in his natural to prevent his CC from landing. The creator commented on it, saying "oh, this guy must know how this bot works, and he is smartly abusing its weakness!" No. I never played that bot before in my life. I just know that bots are dumb and all you have to do is prevent them from doing their build and they fumble like idiots. It seems from your comments about the bots and how they still have readily exploitable weaknesses means, on top of the lack of current full machine-learning, that players will not be as motivated to jump through the hurdles I mentioned above to play AI when they can just log onto Battle.net and play against a human. It certainly sounds like the revolution you promise in your OP is very far away.
I am not sure of current level of skills on iccup. I have played 5 years ago and I imagine B plus back there to be C now, since now there are less people.. not sure.. But then I started to watch streams and I see C+ playing like D+ back then or worse. Maybe points are changed don't know. Also the thing with bots now is they can do pro level play/ tricks on certain aspects of the game, . But lack of brain is still importhant so maybe there is a experienced inactive player with low apm who is stuck on d level but plays all races, , but really knows what is he doing and knows build orders timing push etc. Which might work very well against the bot but not vs human.. It has to be tested and to see. Also new bots are arriving and if one protoss bot has wins vs human, and that bot loses to new better bot, it does not mean the new bot will beat that same human... Sc is really complicated, in bots world D+ 60rank can score win vs #1 A- bot.. Which is almost impossible in human vs human I also want to host a game on iccup in obs mode, bot joins and plays and obs watching.. But no bot coder has done this yet. Looks like they have so much to code and improve that they see this is waste of time, I imagine would be whole week or two work..or they don't know.. And I don't know how to do it since I'm not coder.... Revolutions has started. Bots can beat humans. Even pro human as you can see from the video. Now I started watch bots only by year ago.. But I can see how bots from 2013 have played since some are not updated since then. And today top bots wipe the flor with them. If back then they have won vs humans.. Imagine now.. But still bots do not know every single trick in the book and since are rule based and not NN, and if they are not coded to certain things, which author thinks is not importhant since no other bot is doing it.. it will bug the bot or slow him a lot. Good thing about this is a top rule based bot like Saida is coded very well and recognizes simular tricks and if scout or sees it imidiately change its plans, etc.. Tries to adapt.. If it sees you with 3 gate goon range it understand you want to rush him.. That's smart or it mimic being smart. That's why I created this topic to see is it enough to beat humans and how cleaver are humans etc.. Already tested with one friend and bot scored 3 wins and human is like C so.. Other bot won vs pro Soulkey but might lose to d plus iccup player, need to be tested.
Based on what they have planned, it will be a lot easier to play vs AI than the way it is done currently. I mainly switched to SC2 AI until BWAPI for Remastered is released.
Also, the SSCAI tournament will start soon. Submissions will be closed 1 hour from now, so there is that to look forward to.
i dont care what anyone says, as long as bot capabilities are restricted to match humans (apm etc), they arent going to beat pros for a long time. starcraft isnt just about micro and macro executions. you have to make decisions with imperfect information; most of these decisions being calculated risks and gambles based on what you THINK your opponent might do/be doing. this is where ai will fail. you cannot teach a computer to understand the tendencies and traits of your opponent. what flash is able to do comes from not just having superior understanding of the game, but understanding who his opponent is. what jd might do in any given situation will be different from what larva might want to do in the exact same scenario.
the above is the same reason why bots will not beat pros in dota also. not for the forseeable future at least.
I'm not sure you're correct about this. To use a simple example, some BW bots execute a 3-hatch muta strat and swarm with perfect muta micro. Although it's good to scout your opponent with builds like this, it isn't always absolutely necessary if you build safety sunks at the natural.
In the realm of deep learning AI, I doubt that they will have much trouble obtaining information they need to make good decisions. If there's anything I notice about the deep learning AI's that have played chess and DOTA, it's that they are hyper-aggressive. I would predict a BW deep learning AI would similarly attack early and often, meaning it will almost always have a good read on the human's army.
I also doubt a deep learning AI wouldn't learn to spread across the map with e.g. overlords to detect drops. It would not take many lost games for it to learn that it fares better when it sees a drop coming ahead of time. Once it learns about how to scout and maintain vision I doubt it will have trouble with surprise attacks or tech.
On December 20 2018 16:26 evilfatsh1t wrote: i dont care what anyone says, as long as bot capabilities are restricted to match humans (apm etc), they arent going to beat pros for a long time. starcraft isnt just about micro and macro executions. you have to make decisions with imperfect information; most of these decisions being calculated risks and gambles based on what you THINK your opponent might do/be doing. this is where ai will fail. you cannot teach a computer to understand the tendencies and traits of your opponent. what flash is able to do comes from not just having superior understanding of the game, but understanding who his opponent is. what jd might do in any given situation will be different from what larva might want to do in the exact same scenario.
the above is the same reason why bots will not beat pros in dota also. not for the forseeable future at least.
A. I. Just succeeded in exactly that. Cherrypi bot uses "machine learning moder for high-level strategy selection" basically the bot switches his plan and strategy few times during the game, based on what the bot thinks in that moment which strategy will have highest win rate taking in consideration what enemy is doing. Not as close as Trained Neural net like from deepmind, or anything close than human knowledge and decissions, but better I guess than automated rule based responses, sees air unit, biids turret but does not know how much or when to stop or to continue. https://torchcraft.github.io/TorchCraftAI/blog/
On December 20 2018 16:26 evilfatsh1t wrote: i dont care what anyone says, as long as bot capabilities are restricted to match humans (apm etc), they arent going to beat pros for a long time. starcraft isnt just about micro and macro executions. you have to make decisions with imperfect information; most of these decisions being calculated risks and gambles based on what you THINK your opponent might do/be doing. this is where ai will fail. you cannot teach a computer to understand the tendencies and traits of your opponent. what flash is able to do comes from not just having superior understanding of the game, but understanding who his opponent is. what jd might do in any given situation will be different from what larva might want to do in the exact same scenario.
the above is the same reason why bots will not beat pros in dota also. not for the forseeable future at least.
i dont care what anyone says, as long as bot capabilities are restricted to match humans (apm etc), they arent going to beat pros for a long time. starcraft isnt just about micro and macro executions. you have to make decisions with imperfect information; most of these decisions being calculated risks and gambles based on what you THINK your opponent might do/be doing. this is where ai will fail. you cannot teach a computer to understand the tendencies and traits of your opponent. what flash is able to do comes from not just having superior understanding of the game, but understanding who his opponent is. what jd might do in any given situation will be different from what larva might want to do in the exact same scenario.
the above is the same reason why bots will not beat pros in dota also. not for the forseeable future at least.
I'm not sure you're correct about this. To use a simple example, some BW bots execute a 3-hatch muta strat and swarm with perfect muta micro. Although it's good to scout your opponent with builds like this, it isn't always absolutely necessary if you build safety sunks at the natural.
In the realm of deep learning AI, I doubt that they will have much trouble obtaining information they need to make good decisions. If there's anything I notice about the deep learning AI's that have played chess and DOTA, it's that they are hyper-aggressive. I would predict a BW deep learning AI would similarly attack early and often, meaning it will almost always have a good read on the human's army.
I also doubt a deep learning AI wouldn't learn to spread across the map with e.g. overlords to detect drops. It would not take many lost games for it to learn that it fares better when it sees a drop coming ahead of time. Once it learns about how to scout and maintain vision I doubt it will have trouble with surprise attacks or tech.
i never said good ai would have difficulty in obtaining information and i do not doubt for one second the ai's ability to use that information. alphago is a perfect example of what the ai is able to do given perfect information and an environment where they can calculate the absolute best play with zero variance. dota/sc are games that do not provide these environments. no matter how good the ai is at collecting information, their information is far from complete and because of the real-time element of the game, there are so many variables in every scenario that it leaves no choice for the ai to have to make a gamble based on probability. my opinion is that the quality of their decisions based on the information they DONT have will be inferior to humans because they cannot understand the nature of their opponent. their understanding of the game is limited to statistics and numbers as far as i know
On December 20 2018 16:26 evilfatsh1t wrote: i dont care what anyone says, as long as bot capabilities are restricted to match humans (apm etc), they arent going to beat pros for a long time. starcraft isnt just about micro and macro executions. you have to make decisions with imperfect information; most of these decisions being calculated risks and gambles based on what you THINK your opponent might do/be doing. this is where ai will fail. you cannot teach a computer to understand the tendencies and traits of your opponent. what flash is able to do comes from not just having superior understanding of the game, but understanding who his opponent is. what jd might do in any given situation will be different from what larva might want to do in the exact same scenario.
the above is the same reason why bots will not beat pros in dota also. not for the forseeable future at least.
this game is not even close to a proper demonstration of what the ai can do against a legitimate pro team in a relatively even playing environment. the human team is comprised of casters that dont play anywhere near the professional level anymore (it doesnt matter that they were ex-pros, being retired for a significant amount of time severely affects playing ability). one of them (capitalist) is pretty much an amateur even. also the teams are given individual couriers since the start of the game which is pretty stupid and the hero pool is limited to begin with. the actual pro team that vsed openai toyed with them.
I just lost to one of top bots... Omg he outsmarted me... I can't believed... He actually outsmarted me...and I'm freaking C/C+ Rank... Jesus... + Show Spoiler +
I was zerg went for standard 3h spire 5h hydra, he went fe.. But he cancelled it and went for proxy 3 gate in middle of the game and rushed me with 20 zeilot just before I mstarted mutas... He eliminated me omg.. Gg.. He even hided all of his pylon so my over in his base saw nothing.. Was too late
Good news everybody. Thanks from help from bot coders, Now you can play vs bot in ICCUP simply create a game in melee mode and message me here first so we set time and date. I am in lobby currently online - name A.I.Bot
So quick update , BananaBrain and Locutus which are top protoss bots lose easily to players who in the past were max B- B B+ A- ranks even if they play offrace as they cheese bot and there is their weakness. I have very interesting replays, like bot goes for proxy, human goes for proxy too ... it was hilarious , bot lost , Other game bot goes proxy again but blocks his ramp, but human does cannon rush him and wins or kills 1 probe enters main , and builds barraks/factory while bot goes DT , human blocks entry and bot is lost etc. But Bot BananaBrain won one game. Other cases is bot encounters play which havent seen by other bots and humans just dominate them with early agressions, mines etc, 2 factory push
On December 22 2018 04:43 Modesty00 wrote: I just lost to one of top bots... Omg he outsmarted me... I can't believed... He actually outsmarted me...and I'm freaking C/C+ Rank... Jesus... + Show Spoiler +
I was zerg went for standard 3h spire 5h hydra, he went fe.. But he cancelled it and went for proxy 3 gate in middle of the game and rushed me with 20 zeilot just before I mstarted mutas... He eliminated me omg.. Gg.. He even hided all of his pylon so my over in his base saw nothing.. Was too late
dude that actually sounds amazing, do you have the rep?
I'm trying to download SAIDA bot (arguably the best one ever) but it seems it's gone. Also those library errors when executing the exe is a pain in the ass.
On November 20 2019 10:36 Broodwar4lyf wrote: I'm trying to download SAIDA bot (arguably the best one ever) but it seems it's gone. Also those library errors when executing the exe is a pain in the ass.
I've been working on an integrated client a while ago. Actually, we got it done, you can request, and play matches over the network, but then I suspended the project for other reasons. It is called StarCraft Human 'N' AI League, or SCHNAIL Here is the first video:
And shortly after, an update:
Now I'm working on it again. I reworked the client to work fully locally, downloading the bots. As of writing this, I can run most bots. I don't have a video on it, but you will hear about this soon enough. As you can see, you don't need to mess around with downloading and configuring anything - just select a bot, a map, and press play. I will publish an article about it in the near future. As you can see, it is 1.16. As soon as BWAPI gets Remastered support, I will make the switch.
On November 20 2019 10:36 Broodwar4lyf wrote: I'm trying to download SAIDA bot (arguably the best one ever) but it seems it's gone. Also those library errors when executing the exe is a pain in the ass.
Most of those bots only works in 1.16.1
I have 1.16.1 but still run into a lot of missing .dll files when running bots with executable files e.g. Krasi0, Adias
This is a topic that pops up time and again – we should modify the Brood War API to do things in a certain way, so bots will behave more like humans! Bots are having unfair advantages!
the goal is to have human-like games, is it not?
Everyone seems to have a different opinion about this topic, but personally, I would say the main challenge for StarCraft AI is rather for AIs to beat humans by outsmarting them - they don't necessarily have to play human-like, unless you want to use them as practice partners while you're preparing/training for playing against humans.
FYI, in the context of SC2, Oriol Vinyals from DeepMind recently gave a presentation on AlphaStar at KHIPU, and in the video at http://tv.vera.com.uy/video/55389 at 1:04:46 he says (paraphrasing):
Why does fairness matter? Why are you limiting your agent? Why not simply treat it as a reinforcement learning problem and just try to win and see what comes out? The problem I think, especially in games like StarCraft, it's a game that's been designed with some of these limitations in mind. So, what we wouldn't do is break the game, so to speak. As I was saying, I don't want "rock" to suddenly be very powerful because then the game of rock-paper-scissors becomes uninteresting. I think the question of why we try to impose as many fairness constraints that seem reasonable as possible is mostly so that we don't break the game that has been carefully designed and actually evolved through many years. The game was not only released - they actually patched the game based on balance issues between races that people find, so it's a really complicated process. The actions per minute is one obvious element of imposing some limits for agents but there are actually quite a few more, like how precise are the actions. People that play, if they get under pressure they might start making mistakes, and so on. It's really hard, even if we had the robotic arm, to really say "this is exactly how people attack or play the game" so we need something that is reasonable in that aspect but there's always going to be more. If you see an image, do you add noise to the image? Certainly, we have more precision in the center of our eyes. It's a very cool problem to start thinking about and one that I think more people might start discussing.
Oriol then also elaborated on the issue of fairness about how the agents play two hundred game years, as opposed to a professional player who plays the game a lot but plays in the order of years not hundreds of years. He then said (paraphrasing):
It would certainly be interesting to limit the amount of experience of agents and there's some papers that do that. We didn't do it in this project but if you limit it, clearly, the ceiling of performance would be quite lower but it would still probably be above average play, I would say, in terms of percentile.
So, the primary reason they add limitations for APM and camera etc were to try to avoid AlphaStar settling on an uninteresting strategy, not because the humans would complain AlphaStar's interface to the SC2 API was unfair compared to a human's interface to the SC2 UI, but I expect that was the secondary reason. Note that I referred to the human's interface to the SC2 UI - not the limitations of humans except for how they relate to the interface to the SC2 UI. It's about the capabilities of the interface, not about the capabilities of the human. Oriol does talk specifically about the limitations of humans, but personally I am more interested in seeing what AIs would capable of using the SC2 UI by running as software on the PC (just using pixels, virtual mouse & keyboard control, and perhaps virtual audio) than also trying to limit their mechanical capabilities to be more human-like (robot arms controlling a physical keyboard and mouse, video camera watching a physical monitor, microphone, etc etc etc).
AlphaStar's APM throttling and camera limitations are just a simple ways to make the capabilities using the SC2 API behave more similarly to the capabilities of the SC2 UI. E.g. in the SC2 UI you need to move the camera to get info and select your units on a different part of the map, which might require many frames to do, but the SC2 API can select and command your units from all over the map all within a single frame, hence the camera limitations and APM throttling. If AlphaStar had just used pixels and virtual mouse & keyboard control (and perhaps virtual audio) like they did for their Atari work (rather than an API that uses a raw data interface containing much more highly structured data as input and output), personally I wouldn't care if they removed the APM throttling logic, because throttling APM would be an unnecessary limitation imposed above and beyond the interface to the SC2 UI.
^^ I can't beat the updated bots without "gaming" the entire bot's strategy. I would like to watch bots fight against other bots and I have tested a lot but Locutus keeps winning at FS. I'm still trying to run ADIAS (SAIDA clone) and Saida bot itself if i knew where to get one for download.
To answer your questions, recent versions of the binaries (but not the source code) of BananaBrain, Locutus, adias (which is currently identical to the version of SAIDA that was used in the SSCAIT 2018/19 tournament, just renamed), krasi0 can be downloaded from https://sscaitournament.com/index.php?action=scores. I won't explain how to install and run them though. SSCAIT streams bot-vs-bot matches 24x7 at https://sscaitournament.com or https://www.twitch.tv/sscait. If you want to see particular bots play against each other, except around the time of SSCAIT's annual tournament, you can vote on which bots will play against each other in the next game, via https://sscaitournament.com/index.php?action=voteForPlayers. SSCAIT and other bot-vs-bot ladders like BASIL (https://basil.bytekeeper.org/) also provide replays.
The binaries for AlphaStar haven't been published, and probably won't be. DeepMind published the pseudo-code and detailed neural network architecture specification, hyperparameters, implementation details etc in a paper, but not the full source code.
On November 22 2019 14:09 Quatari wrote: To answer your questions, recent versions of the binaries (but not the source code) of BananaBrain, Locutus, adias (which is currently identical to the version of SAIDA that was used in the SSCAIT 2018/19 tournament, just renamed), krasi0 can be downloaded from https://sscaitournament.com/index.php?action=scores. I won't explain how to install and run them though. SSCAIT streams bot-vs-bot matches 24x7 at https://sscaitournament.com or https://www.twitch.tv/sscait. If you want to see particular bots play against each other, except around the time of SSCAIT's annual tournament, you can vote on which bots will play against each other in the next game, via https://sscaitournament.com/index.php?action=voteForPlayers. SSCAIT and other bot-vs-bot ladders like BASIL (https://basil.bytekeeper.org/) also provide replays.
The binaries for AlphaStar haven't been published, and probably won't be. DeepMind published the pseudo-code and detailed neural network architecture specification, hyperparameters, implementation details etc in a paper, but not the full source code.
I'm not having success running Adias or Saida because the former keeps telling me the .dll it loads is "nothing" and the Saida one just goes back to the desktop and drops. My computer says something about resolution but I've had that error with some other bots and they go back to play in like 5 seconds. I hope there's a program where you can simply just play with or against bots any time. The "nothing" dll error is incorrect since i've already pointed it out in the droplauncher program
@Broodwar4lyf When a lot of bots run, in addition to depending on BWAPI, they depend on DLLs such as a particular version of the Visual C++ Redistributable(s) or Qt, or BWAPI-related library DLLs (especially Java bots). I suggest you try installing the redistributables at http://www.cs.mun.ca/~dchurchill/starcraftaicomp/all_vcredist_x86.zip and try again. If that doesn't work, perhaps try copying the files from https://github.com/Games-and-Simulations/sc-docker/tree/master/docker/dlls into your Starcraft program folder (in the same folder as Starcraft.exe). Also check you're using the correct version of BWAPI if you haven't already (each bot depends on a particular version of BWAPI). Depending on what the individual bot depends on, it may or may not work. SCHNAIL aims to avoid all these problems and make it easy to play vs bots, and I am looking forward to it.
On December 05 2019 16:58 Peter767 wrote: Artificial Intelligence is the very newest and broadest topic available nowadays. I have learned the basics of AI from Facebook. I got some new and exciting information there.
Awesome initiative! I would love to be able to practice against AIs with selectable (approximate) MMR ranges, maybe 1 for each rank (F, E, D, ...) up to however good they get lol.
I think this could make team games, like a co-op, fun too! Me and my friends used to do like 3v5 comps on large maps, which start out really fun when you're a complete newby, but soon enough after midgame the default comps basically do nothing. (Another aside idea: adaptive difficulty, so the AI constantly changes its difficulty in-game to keep the game going. If it's killing you too fast, it slows down its macro, if the human(s) are winning, it cranks up its micro and multitasking, etc.)
How can people help support your work? (edit - found your Patreon )
I think this could make team games, like a co-op, fun too! Me and my friends used to do like 3v5 comps on large maps, which start out really fun when you're a complete newby, but soon enough after midgame the default comps basically do nothing.
Well, hate to ruin your fun, but most bots only support 1v1 Melee matches That's not to say that this is impossible in the future!
(Another aside idea: adaptive difficulty, so the AI constantly changes its difficulty in-game to keep the game going. If it's killing you too fast, it slows down its macro, if the human(s) are winning, it cranks up its micro and multitasking, etc.)
Some bots have opponent modeling, which basically do this. The more you play against it, the better it gets. And generally, bots that have learning enabled will get tougher. This is one of the great questions for me, how to handle it.
How can people help support your work? (edit - found your Patreon )
Much appreciated! A lot of work went into this, and there is much more to come. Every penny helps.