Now Reading
Evolution Of AI In Video Games: Transition From Cheating To ‘Smarts’

Evolution Of AI In Video Games: Transition From Cheating To ‘Smarts’

Anirudh VK

The field of video game artificial intelligence, which borrows heavily from concepts in machine learning today such as control theory, procedural content-generation, data mining, and on-the-go learning techniques, has recently risen in popularity. These technologies, however, have been in use for over 40 years. Only, they were not referred to as ‘AI’ in the truest sense but includes various techniques that are used today by ML programmers.

However, even today, new frontiers are being made in ‘game AI’ that find application in real-world situations. Various companies have also used complex games to test their AI models. Some companies have also entered into the field with AI research, such as NVIDIA.

The field has grown from something driven by the need for more challenge to a science functioning at the juncture of psychology and AI.

The Genesis Of Playing Against The Computer

At the beginning of video games in the early 1970s, there were a considerable amount of offerings that offered the opportunity to play against the PC. This included titles such as Speed Race, Qwak, and Pursuit. While all of these games functioned in different genres, they used computing methods from the 50s to give the semblance of intelligence to on-screen sprites.

However, one of the most memorable uses of AI in video games came with the release of Space Invaders 1978. The game had multiple never-before-seen features such as a progressively difficult player curve and discrete movement patterns between various types of enemies. This was followed up with the release of Pacman, which introduced pathfinding AI to the video game community.


The game was also popular for assigning ‘personalities’ for the 4 different ghosts in the game. While these were no more than simple routine changes, they created the illusion of playing against something smart.

In the 90s, games began evolving as a whole, as seen by the expansion of diversity in genres. Newer styles like fighting games and sports games necessitated the creation of a discrete type of AI.

The Building Of Game AI’s Foundation

Sports games, such as the widely popular Madden Football series and various baseball games, often collaborated with the coach the game was based with. This allowed them to code specific subroutines that emulated the coaching styles of the managers represented in the game.

The expansion of the market in general also enabled developers to begin using more complex processes to model the AI. These include the use of finite state machines and modelling game reactions on emergent behaviour. Player actions were also taken into consideration while influencing actions in the game, increasing the amount of immersion.

The early ’00s were a time when game AI began finding its feet, with commercial level game AI tools developed by companies such as Unreal. More AI concepts, such Monte Carlo Tree Search and the Travelling Salesman Problem began entering the video game domain. These were employed in conversation trees and pathfinding respectively.

Games such as Creatures and Black & White also introduced ‘smart’ pets that the users can take care of. Even though they picked from a limited pool of reactions, these were dictated by the player’s actions. The result was an illusion of choice that made the creatures seem smarter than they were.

A Discrete Sub-Discipline Within AI

While most early applications were nothing but simple algorithms, they were classified as AI due to the ‘intelligence’ they exhibited. Moreover, the intelligence factor itself was influenced by what the player thought was effective. In an alternate case, a game with ‘dumb’ AI would not be considered fun to play alone.

It is in this area that video game AI functions. the programs are written to give the player a sense of accomplishment. Hence, they must function as a foil to a single-player experience. Owing to this, it is acceptable for game AI to do things that would be considered ‘cheating’ normally.

See Also
Top 6 Baselines For Reinforcement Learning Algorithms On Games

For example, an AI that is required to find an object in a 3D space will have multiple subroutines such as image recognition, auditory perception etc. However, a game AI does not require all of this, and can simply query the game engine for the location of the object.

Progression From Cheating To Smarts

Other cheating methods give the computer an unfair advantage over the player, added to its superhuman speed and accuracy in player actions.

However, it still represents the cutting-edge of other AI applications such as procedural-content generation, where AI can produce game levels by using GANs. Indeed, many games have also implemented difficult AI without resorting to cheats of any kind.

For example, the game Far Cry 2 introduced a novel concept in video game AI known as action selection. Upon encountering the player, the AI picks from a set of behaviours which can be executed simultaneously. This includes calling for help when hurt, taking cover when hearing shots, and investigating suspicious noises.

Stealth horror games also make clever use of the game environment, such as those seen in the game Alien:Isolation. In the game, the eponymous alien stalks the player as they hide in closets and under tables. Upon the player making an inkling of a sound, they are discovered by the alien.

These kinds of cases prove that it is indeed possible to create immersive game AI without ruining the player experience through excessive cheats. Game AI is and continues to be one of the fastest evolving fields under AI.

Provide your comments below


Copyright Analytics India Magazine Pvt Ltd

Scroll To Top