The history of game AI evolution, the only thing that can stop AlphaGo may be Werewolf
"Stimulate the battlefield below the golden level, 80 AI in one game", this is a topic that cannot be avoided recently about mobile PUBG.
Regardless of whether the chicken-eating mobile game is really ***ting cold, after all, it is not uncommon for such a hundred-person online mobile game to use robots to make up numbers to speed up the matching rhythm. But why, almost everyone who has played the exciting battlefield will invariably question – "These are all robots, right?" The answer is obvious, the robots used to make up numbers are really ******.
Without collecting equipment, without looking for cover, with poor marksmanship, and slow response… In a game where everyone has a lot of tricks, these special players are like couriers delivering equipment, which is very abrupt. Even if it is to take care of the gaming experience of novice players, but in high-end games, they will also encounter such "silly and sweet", which makes people wonder: Can’t these AIs be smarter?
Today, we will discuss the past, present, and future of AI players in the history of gaming.
Evolution history of game AI
Since the birth of video games, there has been a shadow of game AI, whether it is the earliest oscilloscope tennis or tic-tac-toe, especially in the past when multiplayer online games were still trapped in the difficulty of popularizing network technology. AI in stand-alone games has become the most reliable companion for players and the absolute overlord of the game world at that time.
In a sense, the prototype of video game AI may have already appeared in the classic arcade game "Pac-Man". At that time, the AI in the game was beginning to rise around the world, and "Pac-Man" at that time made a great attempt in this regard. There are a total of 4 different colors of monsters in the game, and each monster has a different pursuit algorithm, and all monsters have a unified goal of encircling and chasing. Therefore, in the game at that time, those monsters did not line up to chase the player, but "chased and intercepted" the player from different paths and methods, and there may be different choices at each intersection. This mechanism has made Pac-Man a significant attempt for modern video game AI, and has also made Pac-Man a classic in the hearts of gamers.
Since then, more classic games have begun to revolve around opposing AI characters, such as Super Mario Brothers, The Legend of Zelda, Prince of Persia, and Mega Man 2. In these popular games, AI has become the key to ensuring the game experience and controlling the difficulty of the game. At that time, the core of the game was actually the confrontation between the player and the AI, but in different ways.
At that time, the ultimate purpose of AI was to let players defeat and then gain a sense of achievement. However, AI at that time was still in the "pseudo-intelligence" stage, and the actions of the relevant virtual characters in the game were based on a series of scripts set in advance based on the conditional judgment language.
However, as game types and gameplay have evolved, the roles and functions of AI have also changed. A typical example of this is 1987’s Metal Gear, in which AI first appeared in the game as a player’s equivalent, was given the role of patrol, and for the first time added the hostile reaction mechanism to the player. This mechanism gradually formed the enemy alarm mode of the core NPC in the Metal Gear series. The appearance of this AI mode also further expanded the mission and role scope of AI in the game.
With the rapid development of hardware, algorithms, and computing power in the video game industry over the past 30 years, the pace of evolution of AI in games has gradually widened compared to that of human players. Especially at that time, most game mechanics were testing the player’s reaction and operation. In these aspects, AI has quickly destroyed humans. First-person shooter games are a good example.
Starting with Counter Strike in 1999 and then Road to Survival, Valve laid the foundation for AI in FPS games, but in the early years, robots in CS have always been one of the places where players complain. At that time, robots were responsible for filling vacancies in the game. They had no other role than being cannon fodder and meat shields, because they basically acted according to easy-to-master rules, and most non-novice players could easily kill them.
However, it is not difficult for those familiar with FPS games to find that in recent years, the AI in many FPS games has become more and more "brutal", and the difficulty level has also increased. In many cases, the highest difficulty AI combat power has far exceeded that of most players. In a recent online elite game of "CS: GO", there was an amazing scene – a robot named BOT Connor single-handedly killed the members of the opposing team.
Today, AI is beginning to surpass humans in all aspects of different games, and even the various chess games that humans are proud of have long been beaten by IBM, Google and other tech giants. Not long ago, Go was completely lost without accident.
Humans are responsible for inventing games, while AI is responsible for conquering games and players, which seems to be an irreversible trend. But in fact, there is always a battlefield in the field of games, where AI cannot completely defeat humans.
Real-time strategy, humanity’s last "high ground"?
After AlphaGo’s 3-0 victory over Ke Jie in May last year, many people lamented that humanity had become a complete catch-up on the road of "Game". However, AlphaGo did not think so, because it has another goal – "StarCraft".
As early as 2016, after AlphaGo defeated Lee Sedol, Deepmind announced that it would challenge StarCraft players with AI in the future. But a year later, when AlphaGo was already alone in the Go world, AI was still "crushed" by top human players in the game of StarCraft. Google’s DeepMind practiced for a year and couldn’t even beat simple NPCs. In a game at the end of 2017, CherryPi from Facebook AI Research Institute was also swept by South Korean former professional player Song Bingju 0:4.
The momentum of AI’s destruction in other fields seemed to stop abruptly in the face of StarCraft, all because of the characteristic mechanism of games like StarCraft, the fog of war.
"StarCraft" is a game like Go, but the addition of the "Fog of War" mechanism has become a gap that AI cannot cross.
In the game of Go, both sides could see each other’s current situation at any time, which meant that the player had no hidden information, and all the information in the past was clear at a glance. The so-called knowing oneself and knowing one’s enemy depended on the player’s thorough understanding and imagination of the rules of the game.
In real-time strategy games such as StarCraft, the fog of war makes it impossible for both sides to know all the information in the game. In this case, based on the limited information collected to make judgments and decisions about the dynamic situation, when to do what, many times need to rely on inference and guesswork. This is the "incomplete information game" that makes AI headaches, and games like Warcraft, Dota, etc. belong to this category.
There is a common phenomenon between top AI and humans in such games – winning the battle and losing the war. If you analyze the gap between AI and humans, intuitively, it mainly lies in the "big picture" and "rhythm" aspects.
As Geek Park mentioned in the previous article "Win Go first, then win the tower, and how can artificial intelligence catch up with humans in the game?", in this type of game, the change of strength is more dynamic, no one is absolutely strong and weak, the real winner may be at a certain point in time to seize the weakness of the other party, and the judgment of this point in time is a complex decision made through investigation and empirical judgment. Both sides of the game can change the direction of the battlefield by inducing and feigning attacks. Even if they are weak, they can achieve the transformation of the situation through some psychological games.
When to attack, when to retreat, where to attack, divide and outflank, sneak attack and harassment, feint and lure… In the incomplete information, it is difficult for the AI to tease out absolutely valid logic to support the decision, so it is difficult to make key judgments at the right time.
Therefore, in the past, many real-time strategy games could only improve their AI level by increasing their AI resource income, building speed, and fully opening the map. Under the same restrictions as human players, it was difficult to simply rely on operations to win the entire game.
Where is the future of gaming AI?
Perhaps AI will soon be able to defeat humans in the field of real-time strategy games, but that is not the future of gaming AI.
With the investment and guidance of tech giants such as Google, Facebook, and IBM in this regard, it is not surprising that AI is likely to defeat humans in real-time strategy games in the future. For humans, it is more important to make better use of the development of AI to provide players with a better gaming experience. Just like when AlphaGo defeated Ke Jie, AlphaGo did not destroy the 3,000-year-old game, but opened the door to a new world of Go for humans through its own advantages.
Regarding the exploration of the future of game AI, the famous game masterpiece "GTA5" is a perfect case. This open game with high realism and complexity has huge randomness, providing players with an unprecedented gaming experience. What’s more interesting is that this game has also become an experimental field for many autonomous driving AI. Intel Labs, Darmstadt University of Technology in Germany and Princeton University have all chosen to use the "GTA5" game to train artificial intelligence for autonomous driving.
You know, smarter AI in the game can do far more than just defeat opponents, more importantly, according to the development of the plot and player characteristics, to provide a more perfect game experience, which can be dynamically increased difficulty, of course, can also be reduced, everything is to make the game more interesting to play.
There is no need to worry too much about whether AI will eventually defeat humans, after all, we still have Werewolves. If one day AI can really read words, read expressions, and tone, and then deceive all humans to win, that will be the end of AI…
Sina Technology official account
"Palm" grip technology fresh news (WeChat search techsina or scan the QR code on the left to follow)
Loading