(This work was part of a final exam and presented during the annual students’ conference titled Journeys Across Worlds: Deconstructing Identities Through Language and Literature)
When Alien: Isolation was published, its AI was marketed as a “learning entity” that adapts to the player’s gaming style, presenting an inevitable threat that needed to be avoided at all costs since it could not be defeated. Usually, the player uses weaponry to defend themselves, but even the alien’s initial weakness, fire, proves useless as the alien seems to develop a higher resistance. Additionally, it even searches lockers and other hiding spots that the player regularly utilizes for their defense. Therefore, the player cannot run or hide: The alien will find them.
But is that really the case? Years after the game’s release, the publishers de-constructed the alien’s AI and explained how it works. Until then, it operated on patterns unbeknownst to the player. They could see that the alien reacted toward sounds or movements but when it suddenly crawled underneath the table and pulled the player out of it for their inevitable death, even though the alien had no visual clues for the player’s current position, they ultimately realized that they were powerless against their enemy.
When the input and output are known, but the mechanisms in between are not, it is called a “black box.” A black box describes processes we cannot understand, and often, when we try to de-blackbox, so to unveil the mechanisms behind a system, we quickly realize that more black boxes are revealed. A scholar named Latour used the example of a projector: It is used daily, no questions arise behind its mechanisms or workings, but as soon as one part malfunctions and needs repair, we are exposed to our lack of knowledge – the black box.
In the case of the alien, or any NPC (non-playable character), the AI’s black box becomes visible when the player faces an obstacle they cannot overcome, or if the behavioral pattern does not add up to the player’s expectations. Consider this: the player provokes a sound and expects the alien to investigate – but it does not. The player repeats the process, and still, the alien does not show up. They turn around to leave, but they are greeted by the alien turning the corner and bodyslamming them into another death screen. The player asks themselves if this was an error on the AI’s part or if it was programmed unfairly (There is an interesting article on that subject by Jaroslav Švelch: “Should the Monster Play Fair?: Reception of Artificial Intelligence in Alien: Isolation”).
The process just described is “Step 2: interest (interruption, detour, enlistment)” in Latour’s sequence of “Reversible Blackboxing:” It is the moment we realize a process is not working in the way we are used to, leading to an interruption or detour and creating interest in its inner workings. This interest more often than not is forced upon since mechanisms are seldom questioned.
Let’s deconstruct the alien’s black box now by using the information the developers published. There are extensive video essays on how the AI works on YouTube, so I will cover the basic concept. The AI consists of two parts: 1) the Director, and 2) the alien itself. The Director AI overviews the alien’s and player’s movements and current locations and guides the alien into the general area of the player. The alien, then, needs to utilize the reactions in its behavioral tree to either search an area for visual or audio clues or wait until the player reveals their whereabouts. Additionally, the Director overviews the “Menace Gauge,” a tension meter that determines whether to send the alien away from the player for them to progress or provoke an encounter.
As mentioned, the alien bases its reactions on a behavioral tree, which can easily be described as branches containing reactions when something is triggered. The alien can choose from different sub-behaviors and unlock them as the game progresses. So when, at the beginning of the game, the Director sends the alien in the player’s direction, its behavior tree might only include roaming the area, whereas as the game progresses, the alien might search in lockers additionally since it could not locate the player through other clues/behaviors alone. Therefore, the alien is not a learning entity in terms of adjusting to the player’s gaming style: Even if the player does not hide in the vents regularly, the alien might still search them. Even if the player does not defend themselves with fire, the alien might still develop a higher resistance. Even if the player runs from one part of the map to the other, the alien will still find it because of the Director.
Therefore, through deconstructing the alien’s AI, we understand that the player’s choices have no impact on the alien’s learning or behavior – it simply utilizes the sub-behaviors implemented in its behavioral tree. However, deconstructing this black box uncovers another phenomenon briefly mentioned before: more black boxes. Even though we now know how the alien operates, the behavior tree or when it unlocks specific sub-reactions is still an opaque process to us and the player.
It is also crucial to mention that calling strange/unknown processes “black boxes” might lead to mystifying otherwise simple mechanisms, and therefore making the term redundant. Even though this example is very much centered around video game AI, the general public demands more transparency regarding artificial intelligence and its inner workings. A recurring example would be the TikTok algorithm or Spotify suggestions.