When playing modern video games, it's hard not to be surprised at how smart computer dummies have become. These pixelated creatures, animated by thousands of lines of computer code, seem truly dangerous, calculate their actions in advance and are always ready to riddle the player's back with a clip of lead. But is the artificial intelligence of computer characters really that smart, or is it just skillful tricks of the developers?
For more details, see our article on the evolution of AI in games.
The worst ghosts in human history
Nominally, the day of conception of artificial intelligence in games, which is able to think and model its behavior based on the actions of the player, is considered May 22, 1980. It was on this day that Pac-Man came out, and the ghosts, tirelessly chasing poor Pacman, became a real nightmare for gamers of the 80s.
It seems that there is no way to hide from these flying multicolored beasts, they not only chased the player on the heels, but also took the shortest routes in order to catch Pacman by the sirloin places. However, the AI of the ghosts was trivial, which is generally expected for a 1980 game. Pacman's opponents had two states: calm and pursuit. In the first case, the ghosts worked according to the simplest algorithm: they moved along a straight path and made decisions only at the forks. In fact, a virtual lot was thrown, which randomly determined the further route of the ghosts.
Ghosts acted much more interestingly in pursuit mode. Their intellect still could not boast of foresight, but at the same time they saw the player in a straight line. That is, the algorithm of actions did not take into account the multiple forks of the maze and the ghosts chose the route to the player at random. It was this imperfection of the ghosts' intelligence that made their behavior most often erroneous, but thereby more realistic and unpredictable.
By the way, intentional AI errors in games are not a bug, but a feature and one of the most common tricks of developers. The most recent example is Wolfenstein 2, in which Uber Soldiers, upon detecting a player, fired past Blaskovich for the first few shots, allowing him to take cover and reload.
When we see how AI-controlled opponents act in concert, how they assess the current situation and prepare attacks from the rear, it seems as if computer characters, if given free rein, would take over all of humanity like Skynet. In such cases, the desire to sing endless praises to programmers and their difficult task of endowing inanimate beings with intelligence.
The work of programmers is really difficult, but there is no special magic here, because all actions of artificial intelligence are based on a simple algorithm, where the options for actions are calculated in advance and the most suitable is selected. The simplest analogy is the game of chess. You can be sure when you play against a computer opponent in a strategy or, say, a shooter, in fact, he is playing a sophisticated version of chess against you. But the possible algorithm of his actions is limited to the given behavior models, patterns, which ultimately destroys the illusion of intelligent AI.
It's time to recall the genre of strategy games and its ancestor - Dune 2, released in 1992. At the dawn of the 90s, the concept of artificial intelligence, which operates independently of the player and independently develops its base, seemed like something incredible. But attentive players easily discerned the patterns by which computer rivals operate, such as the algorithm for constructing the same buildings in the same order. And this problem affects almost all AI in games, whose essence of existence is confrontation with the player.
House 2 in Video Games
The next milestone in the development of artificial intelligence in games was the immortal simulator of human life - The Sims, released in 2000. Will Wright's creation was doomed to success, because The Sims gave a real opportunity to feel like a god and take control of human lives, albeit for fun. Sims lived their lives, communicated with each other, relieved themselves, did their leisure time, found problems on their point, and all this without the participation of the player.
To create sims with the most authentic human behavior, Will Wright decided to take, yes, people as a prototype. Using Maslow's pyramid of needs as a basis, he set priority points for each separate category of needs: physiological, security, social, etc. Thus, it was extremely interesting to follow the behavior of the sims, they were essentially a reflection of people.
Of course, there were some funny incidents of artificial intelligence. For example, a Sim could relieve themselves in the wrong place or leave meat on the stove, thereby causing a fire in the house. But these miscalculations were planned, because, again, sims are a copy of you and me, and in real life, representatives of the species Homo Sapiens often do strange things and get into funny situations.
In attempts to endow computer characters with intelligence, most often, role-playing game developers are seen sending the player to explore a huge virtual world filled with hundreds of NPCs. And, of course, the characters standing day and night, waiting for the opportunity to entrust the gamer with an important mission to kill rats in the basement, do not contribute to creating a believable atmosphere. The first attempt to endow NPCs with a daily routine and create the illusion of a living RPG game world was made by the Germans in 2001 after the release of the first part of Gothic.
Today, characters' daily routines have become almost ubiquitous in Open-World RPGs. You can remember the last "Witcher" or even Kingdom Come Deliverance. True, the implementation is still lame on both legs and the Elder Scrolls IV: Oblivion with the vaunted Radiant AI system was noted in a negative light most of all. No matter how hard the inhabitants of Cyrodiil tried to imitate real life, they often spent the whole day standing in one place. Although, on the other hand, they can be understood, because the Oblivion gates that open everywhere can shock anyone. Maybe this is the reason?
In the shooter genre, the artificial intelligence of opponents is one of the most important components that can make every encounter intense, spectacular and unique. And if the demons from DOOM did not have any requirements in terms of intelligence, then after Half-Life, released in 1998, the gaming industry has changed forever. Gabe Newwell's company managed to endow virtual rivals with an active AI, which confidently flanked from the flank, threw grenades and committed other dirty tricks.
Moreover, Valve went even further and did not limit themselves to the development of the AI of the special forces, but created unique algorithms of behavior for all characters, including friendly ones. For example, cockroaches actively scattered by the light of a lantern, and bulskids reacted to meat, which could be used as a distraction.
Bungie was the next AI innovator in shooters with the release of Halo in 2001. The gameplay concept of the game was noticeably different from Half-life and was an open sandbox, where adversaries, focusing on vast locations, had to actively interact with each other and use a wide range of vehicles to destroy the player. In general, the developers managed to create an AI ahead of its time in shooters, which acted logically, to a certain extent predictably and, nevertheless, could set the heat even on low difficulty.
Of course, we could not ignore our editorial favorite - the shooter FEAR 2005. Fighters from the army of clones of Paxton Fettil still amaze with their ingenuity. Blind fire, flip cabinets, crawl under obstacles and hand out luli in close combat - these are just a few of what the artificial intelligence of opponents is capable of in FEAR
Next Generation AI
No matter how videogame developers try to endow computer characters with intelligence and ingenuity, electronic dummies are in any case infinitely far from real people. Artificial intelligence operates in a predetermined algorithm and repeats the same mistakes over and over again. Sometimes I just want to quote Stanislavsky and shout: “I don’t believe!”.
The only salvation in this situation is the self-learning AI, and there are some recent examples in the big games. For example, the Director AI from Left 4 Dead 2, which is able to analyze the player's actions, on the basis of which he decides which opponents to throw at the four survivors, which weapons to place on locations and what kind of weather to create on the level at the moment. You can also remember Metal Gear Solid V, in which the AI adapted to the tactics of the gamer and could provide the military with night vision goggles if he decides to arrange sabotage at night.
But in these games, the AI only regulates the game rules and does not come into direct collision with the gamer. Now, if the same opponents from FEAR could adjust their actions by observing the player - this was already really interesting. However, judging by the active development of neural networks, waiting for adversaries with self-learning artificial intelligence in games is only a matter of time.
For more information on several of the games covered in this article, read our selection of the best shooters and best RPGs.
The Topic of Article: The Evolution of Artificial Intelligence in Games - From Pac-Man to Metal Gear Solid V.