Artificial intelligence (its weak form) is gradually becoming more and more skillful. The computer successfully solves even those problems that just a few years ago were considered understandable only to humans. One example is the game of Go, where only those who have well-developed intuition and logical thinking can become a champion. Go for a machine was considered an unattainable “ceiling”. Now we see that there are no longer people on Earth who can beat the machine.
It is clear that AI does not only play Go; the range of tasks is much wider. However, specialists from various fields of IT are trying to train a computer to fight on equal terms with a person in computer games in order to test the capabilities of their software platforms. In this case, the person and the bot are in the same conditions. Usually, opponents are pitted against each other on a strategy map with the “fog of war,” so that both opponents are not aware of what objects are located and where on this map (under normal conditions, the game’s “AI” knows where everything is). It was under these conditions that a person defeated a machine in StarCraft.
This is about a battle between professional StarCraft player Song Byung-gu and four different StarCraft bots. One of them, CherryPi, was developed by Facebook. Other bots are created by companies in Australia, Norway and Korea.
The game took place at Sejong University, Seoul, Korea. Since 2010, StarCraft competitions involving people and machines have been held here. The event is supported by various organizations (both commercial, research and public). This particular competition was supported, in particular, by the Institute of Electrical and Electronics Engineers (IEEE).
It is clear that Go and StarCraft matches cannot be compared. In the first case, the player sees the entire board and needs to find a winning strategy. In the case of StarCraft, you also need a winning strategy, but here you have to keep many objects on the map in mind, have good reactions and be able to act suddenly, unpredictably. Understanding all this, the AI developers decided to use the StarCraft environment to train the machine.
Famous StarCraft players have previously stated that they would really like to fight a computer opponent. AI companies such as DeepMind have welcomed the move, saying they intend to hold similar competitions in the future.
And this future has come - people were able to fight AI. The computer didn't work out - the man defeated all four bots in 27 minutes in the overall standings. The maximum match duration was 10 minutes and 30 seconds. The shortest match lasted 4 and a half minutes. And this despite the fact that bots performed much more actions per unit of time than humans. For example, a system created by Norwegian specialists is capable of performing 19,000 actions per minute. Cyber champions from the human world are, at best, capable of performing several hundred moves per minute.
Son stated that bots act differently from humans and their play style is noticeably different. “We professional players only start a battle when we have a chance to win with our army and other elements,” he said. People start fights when they are confident in their abilities. But for some reason the bots that Son fought tried to keep their units alive, without making particularly bold decisions. According to the e-sportsman, the only thing that distinguished bots from people in the game was organization. In particular, the bots controlled the crowd of their units very skillfully, trying to protect themselves from Son's attacks.
According to Kim Kyung-Jung, the organizer of the competition, the bots that competed with humans simply did not have enough experience. The matches they played were too few. For example, AlphaGo played tens of thousands of matches, both against human opponents and against copies of itself, before it learned to play at a high level. But bots playing StarCraft did not have this opportunity.
However, this situation will soon change. Back in August, DeepMind and Blizzard Entertainment announced AI development tools compatible with StarCraft II. At the moment, this version of the game is the most popular among esports players. Some experts believe that AI will soon “take out” humans in the same way that happened in Go. “Once bots are equipped with high-level decision-making systems like AlphaGo, people will no longer be able to win,” said Jung Han-min of the University of Science and Technology, Korea.
A representative from Facebook, which is developing CherryPi, said that the company has long wanted to see how AI would fight against an eSports player. After the match, Facebook received a huge amount of useful information that will be used to improve the bot.
A bot created on the basis of a neural network defeated a famous Dota 2 player on the stage of the main tournament of The International (TI) 2017. The main competitions in Dota take place between teams of 5 vs 5 people, but there is also an individual competition. Open AI created an AI that trained itself by playing against itself, 1 on 1. On stage, the bot beat a popular player 2-0 Daniil Dendi Ishutin from NaVi.
Before this, behind the scenes of the competition, the bot fought with SumaiL - the best 1v1 player - and Arteezy - the best player in the world, his MMR is more than 10,000. Their bot also won “dry” - 2-0.
Fun fact: after the tournament, everyone was invited to try their hand at the winning bot for rewards. By the evening of the same day, the bot was defeated by players from the Internet with an MMR of about 8000. It took several attempts to identify the weak point. However, the bot won on the tournament stage.
Why is this an important event?
The tournament is organized by Valve, the game's developer, with a prize pool of almost $25 million. This largest eSports competition sets a new record every year for the number of views and prize pool. International is watched online by half a million viewers on Twitch, several hundred thousand more through the game client, and millions more in recordings.
Level 2. How does it work?
Let's go down a level deeper (or higher, depending on how your imagination works). Let's find out!
The developers decided to approach the problem from scratch. They realized that if they tried to describe the number of rules in an open game with thousands of variables, it would take several months. Instead, they set up a bot, brought it into the game and gave it complete freedom of action.
What evolution looks like:
- The bot played against itself, the same bot. They exchanged experiences.
- For the first few games, the bot stood still and did not move, after which it began to perform random actions or movements.
- After thousands of attempts, the AI came to the conclusion that you need to go to the center of the map and fight the enemy in order to win.
- And after several weeks of continuous training, the bot began to play at the level of first-class players.
During training, the AI learned:
- make the final blow - this brings bonuses in Dota;
- provoke small enemy creatures to attack;
- slow down your creatures to better group them;
- keep the enemy at a distance from your creatures, depriving the enemy of money and experience;
- cancel ability animations to avoid taking damage;
- bluff canceling ability animations, which leads to a defensive reaction from the enemy;
- adapt to new items that the enemy buys;
- pursue the enemy by predicting his possible location in the fog of war.
To visualize the process of training a neural network, look at the video:
How is a neural network different from playing against a computer?
Game against the computer
The main difference between the Open AI bot and the enemies built into the game is the technology on which they are based. The game's built-in enemies are based on a decision tree. Solutions are found by searching through combinations of input parameters and solving optimization problems. Such decision trees include a fairly large number of parameters, so the player will feel like he is playing against a living or adapting opponent.
Nevertheless, all possible variations in the behavior of such opponents are predetermined by the creators of the game. Built-in bots skillfully imitate intelligence, but do not possess it. Creating such artificial opponents is comparatively easier and faster for developers.
Game against a neural network
Neural networks take a different approach. The network that Open AI used is based on a principle that is similar to evolution. By performing random actions, the network creates various solutions. Solutions are evaluated by a fitness function.
Successful solutions connect with each other, mutate and create new generations of solutions. New generations are compared with their “parents” to understand whether progress has appeared. If there is no progress, the “parent” solutions are crossed in new proportions. The process is similar to the evolution of organisms in nature. Such an algorithm does not always lead to an ideal solution, but it will come very close after going through generations of errors.
In the video, the chief engineer of the project talks about the process of training a neural network when preparing the bot for the tournament:
A neural network-based bot requires significantly more resources than a classic decision tree. Such neural networks are used to design jet engines and find cures for Alzheimer's disease. Open AI used the computing power of the Microsoft Azure cloud server, since Microsoft is one of the founders of Open AI.
Musk's tweet after the bot's victory in the tournament:
The entire process of training neural networks regarding the fitness function can be seen in a video about the game Mario. Using the method of neuroevolution of complementary topologies, or simply evolution by trial and error, the author of the video created a neural network that passes a level in Mario.
At the beginning of training, such an AI does not even know how to walk, after 34 generations and several days of training, the AI was able to complete the level without dying:
Is winning in Dota really that important? What do the experts say?
I went through programming forums, reddit, and communities of neural network enthusiasts. The opinions of the participants were divided: there was both enthusiasm and strong criticism.
Admiration
Translation: The bot wins by slowing down its creatures with the character's body. This is awesome.
Translation: As I understand it, the test was to see how the bot learns to play on its own, without given rules and frameworks. He learned and beat the best. Don't get me wrong, this is great progress.
Criticism
Translation: A bot that wins against a professional in Dota is comparable to a robot that shoots balls into a basketball basket from a standing position. This is not real basketball and not real Dota.
Translation: It sounds much less impressive when you realize that the bot is, by definition, equipped with tools that allow it to have superhuman reactions and accurate knowledge of damage in numbers. This allows you to accurately calculate the damage and deliver the final blow. It will be really interesting if they can teach the bot to play macro strategies that are not available in a 1v1 game. If they can win evenly against a team of 5 people, that will be an achievement.
There were also some who were completely doubtful:
What is the conspiracy theory?
In connection with the release of the API from Blizzard, a conspiracy theory arose on the network. There's a reason tech giants suddenly focused on computer games.
Seconds after the victory of the Open AI bot, Elon Musk tweeted that the bot organization, initiated by Elon himself, won a much more complex game than chess or Go:
The tweet is likely a boast to IBM and Google, who have mastered "simpler" games. Don't forget that Musk and his colleagues have invested more than a billion dollars in the non-profit research organization Open AI.
At a minimum, they will be able to attract the best specialists, who will subsequently be offered jobs at Tesla or other sponsoring companies. On forums and on Twitter, Musk was accused of clickbait and exaggerating the merits of Open AI.
Developers of bots for Dota 2 were dissatisfied
Valve did not invite programmers to the tournament who worked on creating third-party solutions for Dota 2. There are several programs based on decision trees and other technologies that are capable of playing even in 5v5 mode. However, an early prototype for the game 1v5 was demonstrated on stage. 1 by Open AI. Valve did not comment on this claim.
In conclusion, I want to say: “I, for one, welcome our new robot overlords.” As for me, I welcome our future robot guards. And I regularly write about them in our Telegram channel Robotics Channel. Join us!
On October 31, 2017, a duel took place between professional StarCraft player Song Byung-gu and four different AIs. The man won each of the battles. One of the bots was developed by the Facebook FAIR laboratory, the rest were developed by programmers from Australia, Norway and South Korea.
Human vs AI
The competition took place at Sejong University in Seoul, South Korea, where annual StarCraft AI competitions have been held since 2010. In previous years, matches took place only between bots and were organized by the Institute of Electrical and Electronics Engineers (IEEE).
Although the matches in Seoul did not attract as much attention as those between AlphaGo and Go professionals, they are still of great importance in the development of AI. The research community considers StarCraft to be a particularly difficult game for bots. Following AlphaGo's success in Go and some AI advances in chess, scientists' attention turned to the idea of teaching AI real-time strategy games such as StarCraft.
Unlike the same Go, in which constant control of the entire playing field is possible, due to which players can develop a game strategy at the first step, StarCraft requires using memory and developing tactics on the go within a limited and virtual game world. As a result, StarCraft becomes an effective tool for helping AI take a step forward in its development.
Past battles are just the beginning
Numerous professional StarCraft players have said they are willing to play live matches against bots, similar to AlphaGo matches. The leaders of the DeepMind project have already agreed to organize similar matches in the future.
Returning to the past matches, it is worth noting that the total time spent by Song for 4 matches was only 27 minutes. The longest match lasted 10.5 minutes, the shortest was only 4.5 minutes. This was despite the fact that bots could move their units much faster than humans, as well as manage multiple tasks at the same time.
The Norwegian bot, for example, at one point in time reached a speed of 19 thousand actions per minute. This is despite the fact that professional players cannot perform more than several hundred actions per minute.
The winning man noted in the post-match interview:
Professional players begin battle only when they have an army and unit management skills sufficient to win. The bots, on the other hand, tried to save their units without making any bold decisions. However, the way they fought off my attacks was sometimes simply stunning.
Kim Kyung-jong, a professor of computer technology at Sejong University, said the bots' actions were limited, in part, due to a lack of available training data. He also added that AlphaGo was able to defeat Go champions only by studying a large number of matches played by human professionals.
Development perspective
Note that Google has long wanted to train AI to play StarCraft. Also in this pursuit are Facebook and Alibaba. In August 2017, DeepMind and StarCraft developer Blizzard released a long-awaited set of AI development tools compatible with the version of StarCraft II most popular among professionals.
Experts are already predicting the complete collapse of human dominance in StarCraft. According to Jung Han-min, a professor of computer science at South Korea University of Science and Technology, when AI bots are equipped with high-level decision-making systems such as AlphaGo, humans will no longer be able to defeat them.
Year of manufacture: 2010
Genre:
Developer:
Publisher:
Developer's website: http://us.blizzard.com/en-us/games/sc2/
Interface language: Russian
Platform: PC
System requirements:
Installed game Starcraft 2: Wings Of Liberty version 1.0
Description:
StarCraft II is the continuation of the epic saga about three powerful races: Protoss, Terrans and Zerg.
They will face off again in a new real-time strategy sequel to the legendary StarCraft game.
In this brutal struggle for survival in Space, you will have new combat units at your disposal,
as before, with expanded capabilities.
Agria Valley
"Blistering Sands"
"Burial Grounds"
"Crossfire"
"Desert Oasis"
"Incineration Zone"
"Jungle Basin"
"Junk Yard"
"Novice Blistering Sands"
"Novice Desert Oasis"
"Novice Steps of War"
"Scrap Station"
"Steps of War"
"World Ship"
"Xel Naga Caverns"
"Elysium"
"Arid Wastes"
"Debris Field"
"Delta Quadrant"
"Discord IV"
"High Orbit"
"Kulas Ravine"
"Lost Temple"
"Metalopolis"
"Month Ridge"
"New Antioch"
"Nightmare"
"Novice Discord IV"
"Novice Kulas Ravine"
"Novice Metalopolis"
"Novice Monlyth Ridge"
"Novice Terminus"
"Novice Twilight Fortress"
"Red Stone Gulch"
"Sacred Ground"
"Scorched Haven"
"Shakuras Plateau"
"Tarsonis Assault"
"Terminus"
"Twilight Fortress"
"War Zone"
"Arakan Citadel"
"Colony 426"
"Dig Site"
"Dirt Side"
"Frontier"
"Moonsoon"
"Quicksand"
"Tectonic Rift"
"The Bio Lab"
"Typhon"
"Ulaan Deeps"
"Abyss"
"Extinction"
"Forbidden Planet"
"High Ground"
"Lava Flow"
"Megaton"
"Outpost"
"Primeval"
"Sand Canyon"
"Tempest"
"Toxic Slums"
Copy maps to My Documents folder StarCraft IIMaps
- If the "Maps" folder is missing, then you need to create it.
- Go to the game folder and run the file "StarCraft II Editor" or "SC2Editor"
- In the program, select the map you want to play on, then click on the “Map” tab at the top,
- Select the item “Player Options” and set the starting position to Start location 1,2.
- Set your AI difficulty (set in the map editor - File -
Editor settings - Test run - Game difficulty) and then press test run or Ctrl + F9
- Don't forget to change the control from "User" to "Computer".
Add. information:
updating the popular game StarCraft 2: Wings of Liberty,
making changes and corrections to the gameplay and troubleshooting problems in the game itself.
Designed for the Russian version of the game.
Using this file you can update the game without having to go online,
for example, if you want to install it (the game) on a computer that does not have the ability to connect to the Internet.
List of changes:
update 1.0.3
The Friends tab on the Leagues and Rankings page now works correctly.
Fixed an issue that was preventing some users from accessing single player games without an internet connection.
update 1.0.2
The mechanism for checking victory conditions in campaign missions has been fixed.
Fixed an issue that was preventing some users from accessing Single Player mode.
update 1.0.1
Optimized the save process in campaign mode.
Fixed an issue with audio playback on some 7.1 systems.
With the development of the Internet in the world, the gaming industry began to worry more and more about multiplayer. The first matches in DOOM I (of course) were played in a bracket, but as time went on, Counter-Strike gave a powerful impetus to online gaming, promoting Steam at the same time. With the transition to pure online, computer opponents began to lose popularity, and most players did not pay attention to this. I'm in the minority, and I'll talk about why bots are important and where you can find them now.
Why are bots needed?
Many people ask this question when this topic comes up here and there on forums. Like, the days of 56k modems are over, the Internet is in every home, I don’t want to play, people are unpredictable, it’s more interesting with them, you can’t upgrade your weapons with bots, and so on. In fact, they are right - multiplayer with live players will not be replaced by bots. However, the AI is not designed for this.
Bots don’t cheat, don’t spam, they can be customized, they don’t glitch and don’t suffer from rage quit. An MLG player who even does NOSCOPE 720 in the toilet will not log into a server with bots, and will not begin to methodically and with impunity mock others. These are the obvious points. Among the unobvious ones, you can most often pause the game with bots and go about your business; you can use them to practice shooting, flying, dodging and evading techniques.
With bots you can more efficiently explore the map, test weapons, and so on. And if a bot gets into a vehicle, it can most often be kicked out of there. Network code, the scourge of modern shooters (Battlefield 3, don’t close your eyes, shameless one), is also most often irrelevant with bots, because you can play with them without the Internet at all!
Another important point is the so-called power fantasy. Sometimes it’s nice to feel like an unstoppable death machine - the same MLG pro-player who tears everyone left and right without spoiling anyone’s nerves. If the bots are smart enough and are only sometimes capable of shooting crookedly, the scenario is ideal.
Yes, bots are not required in gameplay. Just as optional as good graphics, rich sounds or hitmarks. Without them, the game can be good, but with them it will become even better and certainly not worse. First, let's talk about games that initially have bots, and then we'll move on to various crutches that are worthy of attention.
Star Wars Battlefront
The first part is almost perfect in any plan. Powerful single player mode, good multiplayer, fairly smart bots, although malleable. For example, if you aim at an AI-controlled ship taking off, it will not take off, and you can climb into it.
The franchise's reboot in 2015 was controversial, but a recent update brought something to the game that made me consider buying the game. That's right, bots in private matches! They work in only two modes, yes, but they have three difficulty settings and are capable of almost everything that an ordinary player can do.
Counter-Strike series
This is one of the most stable game series in terms of AI. There were bots in versions up to 1.6 and in Source, and they didn’t disappear even in Global Offensive. Their goal is usually simple to allow the player to explore modes and maps, get a feel for the weapon and learn how to shoot it without compromising online statistics.
At the same time, the bots, although they are not super-savvy, can sit in ambush, throw grenades, talk on the radio, and even listen to the player’s commands. Their complexity determines how quickly the dummy will react to the enemy, how accurately it will shoot and whether it will move while doing so.
Brink
Not as famous as CS, the parkour shooting simulator called Brink treats bots very carefully. The fact is that, like in Battlefield, the single-player campaign consists of standard missions with AI, only the tasks differ. In this case, the bots play the role of extras, and the plot is presented through well-made video inserts.
Since the shooter has a pretty nice class-role-playing system, the campaign works as a preparation for online matches and at the level of leveling. That is, you can enter a fight with live players already at level ten to fifteen, knowing a bunch of tricks with grenades, mines, knowing all the passages on the map and mission details.
Battlefield series
After the release of Bad Company 2, this series of shooters switched to a purely online component. A Battlelog target browser plugin appeared and a list of servers was displayed there. And the upcoming one is unlikely to correct the situation - the declared offline or cooperative component is not there.
So if you want to engage in boticide, you should pay attention to the classic parts. And personally, my attention is focused on the unforgettable Battlefield 2. Initially, there was only co-op with bots on maps of 16 people. That is, 8 by 8, almost without aviation - planes were on one map, helicopters - on two, it seems. Of eight or nine cards, yes.
Fashion saved the day, as expected. Back in the days when Battlefield 2 was sold in stores, modifications were made to it, such as unlocking all weapons in a single-player game, full-size maps for 64 people in co-op, or the harshest realism. Some of them affected the network mode and led to a ban, while others were ignored.
The dawn of modifications came with the closure of sales of Battltfield 2. People, disillusioned with DICE's approach to BF3, sat down to their own Battlefield with blackjack and aviation. In the purely multiplayer component, the Project Reality mod showed itself brilliantly, and in co-op, in my opinion, the best is.
The fact is that on AIX, bots behave almost exactly like people. Smart, experienced people. Airplanes use all the weapons they have available, including bombs and missiles against other aircraft, tanks do not hesitate to shoot at helicopters, and rocket launchers too. More than once or twice my Notar Littlebird got hit on the bottom by an RPG-7.
Well, little things - the number of bots is not limited to 48, and in the settings you can set up to 255 (!) dummies. Naturally, this will cause colossal brakes, but carnage is carnage, especially in the place of the helicopter gunner.
The situation is even better balanced by , which adds several key elements - for example, additional points are given for killing vehicles, headshots kill to death and remove the possibility of revival by a medic, and the ammunition of vehicles is reduced. The mod is very realistic, rich and entertaining, and differs from AIX 2.0 as much as AIX 2.0 differs from the vanilla version. The only thing I don't like about the fashion is that the planes don't use rockets. At all.
I also recommend replacing the exe in combination with mods. BF2 file to open all weapons - the focus applies to all AIX modifications, and since the of. The servers are closed, no one will ban you.
Call of Duty series
Strong multiplayer, awesome story, high dynamics, compact maps - and no AI support at all. The co-op from the last parts does not count. The sad things about Battlefield are also relevant here - cheaters and pro players often spoil the mood of the less experienced, and sometimes you want to relax.
There is a solution. This is a mod written for Modern Warfare that adds AI to the game for online play. The latest version of the modification allows bots to use grenades, RPGs and perks. After the first round, they also begin to use killing sprees. The crutches in this case are that the bots move in jerks, their names cannot be changed, and the experience in the modification does not transfer to the others, including the vanilla version, although the difference there is only in the presence of bots, in fact.
The second part of Modern Warfare introduced active online elements, binding to the host, and in this regard, the hope for bots went nowhere. However, rather crooked but working modifications have appeared on pirated hosts, which also add AI to the network mode. The best example is Bot Warfare.
Compared to PeZBOT, Bot Warfare from MW2 has gotten a lot smarter. If earlier you could expect them to rush the whole crowd, now they do not even hesitate to sit in ambush. The AI easily uses available weapons and rewards for killing sprees, has learned to use under-barrel weapons, easily climbs ladders and watches killcams! There’s no point in the last big bang, but, as they say, presence is there!
There is one bad thing about the modification, which made my purchase of the MW2 license the most worthless purchase in my entire life - it does not work on new patches, since there is simply no way to open the console. But I bought the game for the sake of the mod! Well, and to salute the awesome multiplayer too... Fortunately, several pirate hosts are still working, and now I can play with all sorts of React and the like with an honest soul. For modification, I advise you to install the penultimate version, the latter does not allow you to open the desired menu.
As it turned out, there are bots in Call of Duty 2, and their name is . The multiplayer there, although devoid of any leveling, is still fun and class-oriented. The bots there are almost as smart as those in Modern Warfare, although they are a little more difficult to launch.
These are the most striking and interesting examples of shooters with bots. If you know something else, or have an opinion regarding the topic or the article itself, do not forget to comment on the material. And remember - we are not technical support, your opinion is REALLY important to us!
