Deep Learning. What is it for?

Glossary terms:

Algorithm: A set of rules or conditions that information is passed through in order to lead to a definite conclusion. This is usually done in a series of code.

Director A.I.: An all seeing A.I. that constantly tracks the player in Alien Isolation as they progress through the map. It feeds information to the Alien A.I.

Keeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeenan’s Part:

How does deep learning work?

Well, let’s say the input was an image of a dog or a cat, and the algorithm had to indicate which species it was; First, the images would be separated into numerical values, such as one per pixel. Then, they would be parsed through “biases,” which would alter the values based on other inputs and a set of rules. Finally, the numerical values would be put into an output algorithm that would compare them all together and guess whether the image was a cat or a dog based on the numerical outputs.

In order to train the algorithm, the right and wrong guesses would be fed backwards through the algorithm to fine tune the biases, creating a more accurate neural network with every input.

Neural Network algorithms are not only used to guess species or nationalities, but can also predict movements. By taking in your location data from any given moment, the algorithm can eventually predict where you are or where you will be in the future. The video game Alien: Isolation is a good example of this type of algorithm:

Oliver’s Part:

Alien: Isolation is a 2014 survival/horror game that takes place in the Alien film universe. The game takes place on an abandoned spaceship in which you are relentlessly hunted down by a killer alien, and you must hide for your life as you progress through the story. The game utilizes two sets of artificial intelligence: The Director AI and the Alien AI. These two systems work together to create a challenging and scary atmosphere for the player. The Director AI is all seeing, and constantly tracks the player as they progress through the map. It feeds information bit by bit to the Alien AI, giving it clues as to the player’s location but never outright telling it. This forces the Alien to think for itself and learn from the player’s actions. The Alien has many behaviour trees that dictate how it will act. Based on how the player progresses, different behaviour nodes will unlock to adapt to the player. This keeps the game difficult as the player gets more skilled, and forces outside-the-box thinking.

Beeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeen’s Part:

The Neural State machine for Character-Scene Interaction is an AI made to allow a player controlled character to move in its environment fluidly and more naturally. It takes a sample of movements recorded from motion capture and predicts based on what action you are doing, what the next one will be. With this, it generates a smooth transition between each action the player does. Along with this, the AI sets points on itself and the objects and environment around it in order to properly navigate the area it is placed in. It is important for the AI to have these points as it creates a better exploration of the environment. 

Connnnnnnnnnnnnnnnnnnnnnnnnnnnnnor’s Paart:

In the 2019 remake of Resident Evil 2, the player is stalked for certain portions of the game by the hulking Tyrant. The Tyrant is drawn to even the slightest sound made by the player, and roams while offscreen, even in unloaded areas, forcing the player to adapt a more cautious playstyle. It also becomes more or less aggressive to suit the player’s skill and said playstyle.