Artificial Intelligence 

& how it works 

Students: Ikonomou Clio , Nikolaos Myritzis 

Class: Α3 - Grade: A - School: 1st Arsakeio Senior High School of Psychico

Course: IT Applications

School year 2022-2023

CONTENTS

HOW DOES ARTIFICIAL INTELIGENCE WORK

IMAGE 1: Basic Ai function
IMAGE 2 : Keyboard (Keyboard, mouse, microphone and controller are all forms of input devices.)

Input

 

 Before creating an image, the Ai needs information to process: Information such as text or speech which is otherwise called "Input". If for example we want to take the word pineapple and give it to the Ai to create an image of a pineapple, the word pineapple given to the Ai would be considered the input. In other words input is the amount of information given to the Ai program. However, how does the Ai know what a pineapple is? Our human brains are thanks to years of evolutionary progress capable of not only understanding the concept of a pineapple but they are also capable of recalling memories and information learned in the past to further illustrate the image of pineapple. The ai on the other hand is in need of time consuming training with deferent methods to understand the input whose form varies.

IMAGE 3 : Flow Chart of a basic algoritm.

The AI Algorithm

After enquiring the input depending on the form of the input, the Ai proceeds different functions in order to create an image. However, before the Ai is able to utilize the input, first it needs to understand what the input means. When given text the majority of the Ai algorithms use an advance model of the natural language processing (NLP). To begin with, by finding spaces in the sentence, the Ai splits the input in small different strings (string = piece of text). For example "I love you" would be split in "I" "love" "you". In that way it is easier to understand what the meaning of the sentence is because now the Ai understands that it has to deal with two nouns and a verb. After that it will examine its huge library of words to identify what each string means. If the Ai cannot identify the word given it will either report an error or provided that the developers have thought about it, the Ai will create data about the new word it learned. The way data will be created is by enabling a teaching Ai.

Computers and Natural Selection

The previous paragraph described the way an ai learns, which is basically by "random mutation s". To uderstand better what that means, we will have to understand the concept of Darwins' theory called "natural selection". Acording to Darwin, as animals have children, random changes may appear in their DNA. Theses changes are known as mutations. To evolve as a speace, the new generations' mutations have to be benefitital for their survival. For example. lets asume that there are two sister fish living in the crystal clear waters of indonesia. Despite the fishes' Breed being red, one of them has a random mutation which changed its collor to blue. The fish with the mutation will be able to camuflage better in the water and therefore pass its genes to the new generation changing the collor of the breed and its
DNA. In the same way,  random wyring is happening in the code of the ai in order to create multiple versions of the same ai with different sensibility to certain features. For example, two ai with the same structur may look for diferent aspects to sepatrate two images. One ai may look at collor while the other my look for text or the form of the subject. Out of all the different versions and after a lot of testig. one ai will emerge that has just the correct amount of sensibility on all subjects just as we humans do. 

Teaching Bot

IMAGE 4 :  Future androit

When it comes to an Ai learning new words there are two different bots needed. The first one is the teaching bot and it is provided with two galleries. The first gallery is constituted by a collection of images. The second gallery is a collection of words that describe what 

the images of the first gallery depict. For example if there is the image of a tomato in the first gallery, the second gallery will have the word "tomato"in it. The teaching bot will show this images and words to the Ai in order to create a database of words and what images they are equal to. However,  sometimes mistakes happen and the Ai may not recognize images that look similar. For example it may confuse a sword with a fork. In order to make it clearer for the Ai what really separates the two images, there is a second bot whose whole purpose is to test the bot. The testing bot will show the Ai similar images. If the Ai does not recognize the difference between the two images, random "mutations" will automatically occur in the Ai in order to help it identify the difference. If the new version of the Ai can separate the two images, the test bot will show it more similar images and the cycle will continue until the Ai is a master at distinguishing images and comprehending words.


DALL-E


DALL-E 2 is an AI-powered image generator created by OpenAI, the makers of ChatGPT. The original DALL-E was released in 2021, and DALL-E 2, the updated version, was released in November 2022. Users enter text descriptions into the system, and the software spits out realistic, original images.