For some background info check out this video: • Progamer vs the HIGHES... Patreon: / harstem My second youtube channel: / @harstemcasts Twitter: / harstemsc2 Twitch: / harstem Instagram : / harstemsc #harstem #SC2 #StarCraft2
AI Observer: *clicks away from a battle, zooms on natural* AI caster: OMG did you see that? Micromachine derped and his SCV left the gas bunker at a slightly wrong angle, his gas mining efficiency just dropped by 0.05%! AI crowd: Ohhhhh!!!!
The thing I noticed about this particular replay is that the move of all units reminds me what a "real" futuristic fight would look like, because it gives an illusion that each and every unit is independent from each other when it comes to movement and decisionmaking.
i would like to see a game that each created unit were controlled by a single human imagine a single game were serral controlled just a single roach, or a single worker, another one by reynor, etc, against the same from the other side. When the unit die, they get the next one created you would need 400 players in line, just in case you reach max pop, but i bet it would be like this
@@BlindMidget I think it's more complicated than that. You can clearly see the units acting in groups, coordinating together, which is really difficult to get done if every unit thinks just for itself. On the other hand the drones moving a bit back from time to time could mean they are indeed controlled individually. The Marines covering each other fighting the lings or the groups of mutas/lings walking together. I believe they are more or less moved like a group, with every single one moving in the same direction. Obviously figuring out what they should do is extremely difficult in a game of starcraft as the options are endless.
The abriviature ,,AI" means ,,Artifical Inteligence". For every avarage inteligence human being on Earth, it is obvious that ,,AI" is fake news, because every normal inteligent person on Earth, knows, that a machine cannot have human inteligence, nor inteligence can be artificial. That literally means that inteligence is lacking there. A computer CANNOT have inteligence, THATS AN ITEM.
@@ch4osaeternum74 That is because we are essentially insects to AI I mean what human can do what AI can and comprehend. The only thing is that AI are limited currently and aren't ahh self learning I suppose is the right word, it's bound to a continual script so to speak so it only holds enormous power over it's field and that's it. But inside the AI's field we are insects really.
it's a pretty smart idea for the bot maker to make the bot chatty because then he gets feedback on why the bot does what it does and can use that information in adapting it's decision making tree
@@Diogenes2077 myself and my friend who also worked on MicroMachine thought about adding a conversation AI to plug in our bot but it seemed like way too much work for no improvement on its gameplay skills
To me it seemed a lot like the battlecruisers were being used because it was exploiting a flaw in the other bot than because they were inherently good - Eris was just really really bad at dealing with them (every time it used neural parasite and teleported to the other side of the map it should've instead gotten a kill on the battlecruiser by either just attacking it while it was under control or just teleport it into a stack of hydralisks to kill it instantly if no hydralisks were nearby).
@@Ole_Rasmussen I'm not saying it was deliberately designed to counter it, i'm just saying that it happened to play a lot against Eris and found out that battlecruisers were effective against zerg because Eris constantly misplayed against them. The AI can't tell whether the opponent is misplaying or not, all it sees is 'battlecruisers increase my win rate', but they were only performing as well as they were because of flaws in the opponents it was facing.
@@Ole_Rasmussen i think Micro was just better at being terran than eris was at being zerg. i'd like tosee stats but i feel like zerg bot would've been the weakest one since units have generally low range and range seems to be winning factor for bots, otherwise thhey just derp around in attack range
Huh, that bunker mining trick is really cool, even if it's also incredibly annoying to hear XD The chat scripting was hilarious, too XD (I think the relaunch/cancel offensive messages were just to help the bot-maker understand the decisions being made since they were in allies-only chat)
In Age of Empires some bots does provide allies with info for when they attack to help players to coordinate with them,or use taunt commands to cancel orders etc.
@@Steellmor I personally think it was the first one - so that you know where he's going in and when he is retreating so that the player knows whether it is time to retreat because hes going to fight 1v2 or when is a good time to attack because your bot attacks as well etc. (Kinda good to not have to be like "Well, lets attack and hole the bot will join me." Here you know exactly)
Terran bots have a lot of difficulty to deal with Tempests, because they can kite extremely well with infinite APM. Maybe I'll try to find a replay and fill a request for Is It Imba Or Do I Suck?
@@pepi560 I did submit a replay of MicroMachine losing against Tempests with a very funny balance complain text, but I haven't got any news from Harstem or his team yet :(
This is why AlphaStar (AI) is just outright superior to these bots. Its APM was capped at 300 and it could only see 1 part of the screen at a time and it still outplayed Serral.
What I noticed is these bots are really scared of units dying. If engaging would mean guaranteed death for the units, they simply don't engage, even if the trade would be worth it. I think that's why the siege tank was left alive, if the lings stayed they would have died so instead they ran away, still taking heavy losses in the process.
@@cloaker7139 I would argue it isn't. In many fights the bots could have been ahead by trading away units but didn't consider it BC they viewed it as their units dying = always bad so any trade that costs money = bad.
@@nandornagyilles3290yep you are completely correct. The Zerg could have won (well really either of them could have won) numerous times but they were too concerned with minimizing individual unit losses over securing important victories that would put you net ahead if you traded out.
Watching the micro is both frustrating and amazing. I love how perfectly every unit gets individually micro’d, but I hate how the AI seem to lack the ability to rationalize “trading”. It seems like it will not attack into favorable trades, and wouldn’t even allow the first reaper to take damage despite the fact that it can just heal whenever.
Wow, out of all the AI games I watched, I felt like that bunker trick to mine minerals faster was the first, real AI-only tactic I've seen. It's a trick that, without the constant superspeed of the computer, it wouldn't be feasible for a human. That was so cool.
@@Rasmorak Of course, which is literally a terrible move, the bot was saving low HP BCs and letting them get repaired. The BCs should have been teleported into the hydralisk to die, not into safety.
I reflected on a couple things that might be going through the AI’s minds. 1) perseveration of resources. Above all else, every enemy unit must at some point die. In order to accomplish this task, you yourself must have units. The goal then is to maximize how much damage you do per engage until a certain threshold is met and then disengage. You must continue this attrition style until you’ve slowly mined out your opponent. Versus an AI, they won’t move like a typical human and can’t be as easily pinned down. They will out micro you, retreat and whittle you down efficiently-or try to. A human on the other hand will accomplish a different goal, trading a whole section of resources at one time in order to achieve a macro goal. I’ll not sure if other AI are capable of this. This type of goal is different since you’ll make a crippling play in order to buy time out delay timings and prepare another attack or create a distraction for one. The overall goal is the same, kill every unit effectively but it is done through a much different a full committed mechanism. Human players will still make small trades, but are aiming for larger “abstract” plays to disable technology or weaken parts of the enemies base for a future siege. The AI only cares about unit trades, all units must die first at the optimal level and that’s why they split like they do. When you saw the hydras prioritize the cyclone, it was due to the cyclone’s high kill efficiency at that time. That unit had a high threat level even as a single unit. It was very mobile and had a high trading ratio. That unit would in the long run create a winning state for the enemy. As you saw, Eris had not yet understood the new presence of the battle cruisers as a higher threat over those couple cyclones at that time. The battle cruisers had not yet built up they kill efficiency compared to the cyclones. 2) the chat read out from MicroMachine might be for debugging or viewer appreciation. Any one watching will be able to see what the AI is deciding to do. Eris is likely running a similar internal decision tree without posting it, but it works be great to compare both in chat. MicroMachines chat highlights the concept described in “1)” where the goal is to eliminate every unit through pure efficiency. You must kill every unit but you must preserve your own at the highest level of certainty. Once a certain threshold is calculated, you must disengage. A human failing to disengage happens often and results in dramatic loss of resources. However, those lost resources are sometimes still made in order to accomplish the larger goal of a macro trade that an AI wouldn’t commit to. However, when a AI faces an AI, it cannot easily make a macro trade effectively as the opponent will fight with the same unit superiority. The AIs exclusively focus their efforts to run through mining out each others units. A great comparison of this would be Medivac widow mine drops on mineral lines in order to stem income and boost your own offensives. Even if your attack isn’t favorable, you have time to rebuild and attack again. However, if your primary offensive has been successful and the opponent also didn’t succeed in fighting off the widow line drop they are now behind in micro and macro states of play. A human player would also take on disadvantaged plays but attempt to outplay the situation through they micro control. They would make effective trades with their composition, timing or grand strategy to improve their future game state. On the other circuit board, the player only will engage again at points on the map where they are already favorable for an outcome. The disadvantage of this is that once the game state has reached certain thresholds across the map, one AI has an incredibly higher chance of remaining at optimal engagement. That AI has more favorable positions to mine out enemy units efficiently and will win quickly, their enemy doesn’t have a have macro play at this point either. So much map control has been lost, the opportunity for a larger and abstract trade in order to gain time or prepare a future attack will result in total defeat. That commitment would result in a absolute impermissible amount of resources lost. 3) every unit is moved individually and the camera pans onto every friendly unit and every visible enemy unit concurrently-impossibly high APM. The nature of enemy movement increases diffusion of both players’ forces, favoring the race with higher mobility and range (e.g., cyclones). I’ve seen AI play against humans and they don’t move as sporadically due to their opponent’s nature to clump and move in unison. However the AI will still individually micro units in combat to optimize damage and may still perform multiple offensives. Each AI also spread out to protect their bases and collapsed at certain times during more heated engagements-before dispersing again. The AI players are like a buzzing swarm that will gather for a strike, then return to their fleeting and multitudinous positions and optimal trading states. Again, the AIs have the supreme goal of farming each others’ units more efficiently. They will leave a target on 1hp if that their own unit alive.
My favorite part of all this is looking back afterwards and remembering that marine/cyclone/tank/banshee/viking/BC is apparently this bot's idea of "standard" TvZ XD
The name STANDARD for the strategy was actually chosen because it was the original build of MicroMachine. I did not put much thought into it and did not realize it could be interpreted as standard pro build order. The EARLY_EXPAND strategy looks a lot more like what pro players consider good. But MM is a bit stupid so when it chooses that strategy and gets all-ined by its opponent, it dies painfully. STANDARD strat is the most defensive strategy of MM.
This reminds me of Brood War's AI- every single unit in the AI's army behaves as if it has a mind of its own. You'll often find individual units like HTs, Science Vessels, Queens, Scouts, and Wraiths wandering around the map, taking shots at you. Only now, the AI knows how to micro, and they have a build order that's both really good and flexible when it needs to be.
My God, seeing this bots was hillarious, it seems so unreal to watch all those units fighting with so perfect microing....it just looks awesome, specially those sick marines...i think in AI terms, terran looks virtually invensible.
The vikings are very much worth it. Just look at the unit lost tab. The vikings killed more resources than their own worth in overseers alone, not to mention they waste hydras time, provide vision, and kill other air units.
yeah but its not about vikings being good, its just how bad bot deals with longer range. I think in bot vs bot of tvz terran would always have adventage considering how their micro looks. mutas would fly towards viking and then run away after ALMOST hitting once while 2 vikings were just shooting at them all this time. there were also lots of skrimshes of zerling vs marine that were at a huge adventage to zerg but they lost cause they just wouldnt attack, just ran back and forth being shoot out like ducks
It really sucks that the only way to gain mileage off vikings is to have literal inhuman APM computational ability that isn't even possible for non-machines.
Yeah, the fact is that there was no need of air units at all from the zerg. If he simply played hydra ling bane and A moved he would have won easily and the vikings would have done nothing.
@@manofculture8038 I wonder if it’ll work though, the AI’s certainly can outdo the progamers, but if you turn off their ability to macro(produce+ build) then it might work, since the AI’s certainly can override any commands given by a human due to having 100s of times the apm.
We actually once had a tournament with Archon games. Each author could edit their bot to allow Archon mode and were playing against other bots. However, due to the limitation of the API, we can't do Archon vs Archon games. It was still fun though, when I was playing with MicroMachine, it was letting me control the units that were selected and once the units were not selected anymore, MicroMachine took back the control of them. I did retire my bot from the tournament though because the tournament organizer was plotting against my bot to make sure it doesn't reach the finals (we got aware thanks to a wistleblower that showed us private Discord conversations). And that fucker denied all allegation...
Really glad someone wrote that comment, because as soon as I watched the uber-fast clicks and marine-splits, my thoughts went straight-up to what a Maru or Serral can do with these to x100 their micro. If you and me thought of this, then chances are the botters have as well. This gonna be *fun!*
Whenever the AI says something like "launching offensive" or "canceling attack" its probably only him who can see it so the developer can use it to find bugs/make sure its working properly.
I think we need a Harstem v MicroMachine battle or two (since it has different builds). It looks like it might actually be a bit of a challenge to face off against since Terran has many more units that benefit from extreme micro.
On the grand scale this looked like 2 colonies of ants fighting each other, and I mean that as a compliment. This was not at all what I was expecting but it was very fun to watch.
Its funny how it looks similar to human tvz. Zerg is eating the map and terran is very cost efficient and slowly push with expansions. Would like to see all matchups.
The fights where the AI spread their units out makes the game seem more interesting than it actually is. if SC2 was like that normally it'd be pretty cool imo.
Interesting watch, seeing all these units moving around more independently due to the AI micro/macro abilities. Also the workers hopping into and out of bunkers was crazy
@@tedobg4778 Actions Per Minute. It's a measure of how fast you're issuing commands. The AI is generating absolutely ridiculous API compared even to the best players because even the best players use control groups (a squad of units gets an attack order as a single action) and maybe a handful of waypoints while the AI issues commands to each unit individually and slices the pathing into hundreds of sections rather than just a few. Also, the AI can constantly have eyes everywhere and never gets human problems like tunnel vision so even while managing every unit individually on an offensive or defense, it's still keeping perfect track of macro at the same time, at all times.
I really enjoy the SC2 bots community they are very interesting to watch and really shows you how potentially nuts AI can get in a strategy game when not restricted by apm.
This was so cool n bizarre! The bigger fights felt like ant colonies fighting- or no, more chaotic than even that. I appreciate your trying to keep up with it all as a caster. 😂
I wonder if something like a Bot-Human combo will be the future of RTS. The bot handles moment to moment micro, the human overrides the bot's directives with new orders, issuing simple commands that they set up ahead of time, sort of like macros on a PC. For example, the dozens of times both armies seemed to dance around engagements. Imagine one side had the ability to simply say "Commit," and the bot would then commit all forces in the area to the engagement. The human player can observe anything on the map and make overriding decisions that the AI will then execute with their vastly superior unit control and precision. The human also dictates build orders, that they can, again, override at any time. They essentially become the commander of an auto-functioning army. I'd be interested in seeing a bot that can do that.
This video makes me appreciate AlphaStar even more. When the bot can do 55K APM and move everything on the map at the same time, it seems kind of pointless.
Idea for a bot or bgwss: as a terran, focus on defending your main and building ccs. Then expand one base at a time, drop 20 mules to mine it out in one minute, rinse and repeat. After all bases have been mined out, use you 50k ressources to smash your opponent
This was actually super cool to watch. I hope at some point they get to a point where the AI does most of the work and a player can 'command', as an example, hit that kill switch at the end of the game.
This is pretty cool, but the programming seems to need a ton of work. Eris in particular seems to really struggle with deciding when to commit to an engagement, and instead seems to just bleed off units while microing just within range.
I have programmed some bots myself and it is actually really hard to define which engagements can be taken and which can not. It is almost impossible to find a general function you have to hardcode most of the behavior for specific situations. It is impossible for the bot to always make the right decisions because he just doesn't know everything that's going on.
It wouldn't be AlphaStar without the APM cap. Because of the way AlphaStar learns, the programmers would basically have to retain it how to fight entirely. It'd be OmegaStar at that point.
In the late game, Eris should just program the AI to spam banelings and run them into the Terran's expansions, then it would win every game, like every Zerg-pro...
To kill the base it the workers? Killing the base is not cost effective at all. You need 19 banes to kill a cc. And if you what to kill scv's with banes, you'll need one bane per scv with this ridiculous micro. I guess that's why not even one baneling was produced in this game.
@@mehdimarashi1736 Okay, I meant you run your banelings into the mineral line and kill workers and the base at the same time due to aoe. And Zerg doesn't need to be cost-efficient in pro ZvT, it just needs to deny income to the Terran and mine more... In this game the Zerg had way more bases and the Terran AI spent half the drone time running them from 1 base to another... If the zerg ran banelings into the command centers he would have won even if it is cost inefficient...
Now we need a mod, where humans bring the overall intelligence into the game. When there are no commands from humans, the AI just goes wild. But the humans get all the power to command the units. And the human commands have always highest priority. Then the games would be heavily action packed and probably more interesting ;-)
You might like the Total Annihilation/Supreme Commander series, then. Individual units will move and shoot on their own independent choices if you leave them alone, but you can corral them into cohesive offensives with a few choice orders. Individual unit decisions aren't terribly advanced or strategic, but it's still neat to see, for example, fighter planes performing maneuvers all on their own without external input.