Artificial Intelligence for Kids

Key Takeaways

hello world and Suraj and hold on hello Suraj my name is north and I really need your help what do you need I don't have time to explain they're coming after me can you please just press the number 4 wait who are you hello hello number 4 okay whoa what where am I what is this place freeze put your hands where we can see them hey over here get in the car wait you who are you just get in before they perform a brain drain no way okay I definitely don't want to bring forward oh my god oh my god okay that was close wait a second you're just a kid how do you know how to drive around here kids are better drivers than adults I can take video games alright well what is this place and why do you need my help this world is called Cola it's similar to earth but not really I asked a friend to teleport you here because I need your help finding a legendary device called the luma wait so I'm on an alien planet yeah don't all right so what's the luma it's a device that compared the entire planet with unlimited free clean energy it would save our entire planet from the destruction caused by pollution its location is on this USB stick but I need your help finding the relevant information inside of the stick the king of Cola King Nana wants the stick because he wants to use its power for himself to create a giant burger that only he can use instead of powering them too don't let that happen we gotta find it before he does okay I definitely don't want your planet to be destroyed so let's do this all right so I'll just plug this USB stick in my laptop and there we go wait this is just a bunch of pictures of your planet like mountains and rocks and lakes exactly I got this pictures from a satellite one of these pictures has the luma on it it's bright pink we got to find which picture has the luma but it would take me years to find it there is five million pictures on this USB stick and that's why I need your help you understand computers so maybe you could show me how to do this all right I can help let me see hmm do you know what programming is I've heard about it but I don't know much about it all right check this out so a computer is a machine and we can use it to do literally anything we can dream up and the way we can do this is by a process called programming you can also call it coding or software engineering basically it's a set of instructions that we give a computer to do some tasks that task can be creating a game world like Super Mario or playing a list of songs or even creating a website that sounds awesome so how do I program something well you need to choose a programming language first to code in okay which one well computers read instructions as a set of ones and zeros these are instructions that are easy for a computer to read but they're kind of hard for us to read so we can use programming languages like C or Java or Python that are made up of words instead of numbers that way we can read them and understand how they work if we take a language like Python we can write a simple program that just says hello world all I do is type the command in print then the phrase hello world into this Python text editor then I hit the Run button and it will say hello world and the output window that's the output of the program seat to us the program just says print hello world but under the hood all this text is broken down into ones and zeros so a computer can read it and process it that's awesome so I guess Viking is the best programming language well it's definitely my favorite one I thought is a really easy language to start learning programming with but in the end programming is an art form it's something you can get better at with practice it's the art of problem solving I really like Python because you want to think too much about what the right commands to use them and instead you can just focus on solving your problem that's for equal I will do you spite enough so how do we use fighting to find the new line all this pictures well we're gonna have to do some machine learning using Python what's machine learning well the traditional way to do programming was to tell a computer what to do like given exact instructions let's say we want to tell it to detect the luma in a picture well we can write code that can detect all the different features of what the luma looks like but it's color and shape and size but the luma could be in many different kinds of positions in one of these photos there's no guarantee of what exactly it will look like you do have some pictures of a luma right definitely there on the computer as well okay so instead of us telling the computer of what a little looks like we can have it learn what the limit looks like that's called the machine learn it's when we give a computer not a set of instructions to complete a task but instead a set of data and a task and have it learned from that data to complete the task so we can use Python to create a machine learning model then we'll feed our model a bunch of pictures of the luma our model will learn exactly what the luma looks like then want it to learn what a luma looks like we can give it all the satellite pictures you have and it will tell us when it detects so luma awesome so how do we make well there are lots of different machine learning models we could use but a really cool one I like to use is called a neural network a neural network can learn anything if we give it enough data and computing power here's how it works if we have some data with two columns the first column is a set of pictures of the looma the second column is a label that says luma or not luma we conveniently drove the data to the neural network one at a time there is some pattern here some kind of relationship between the two columns and a neural network can figure it out wanted to learn this pattern we can give it a new picture and it will automatically be able to tell us if there is a luma or there isn't how does it do that well the neural network is actually just a few math operations our input is going to be an image but if you think about it an image is really just a set of numbers it's a set of pixels and each pixel has a value that corresponds to a color a computer reads these pixels so it can display the colors of the image so it's a table of numbers and our neural network is a series of math operations it consists of a set of layers let's just use a simple network with three layers an input layer a hymn layer and an output layer in a neural network data goes from the ankle layer through the hidden layer to the output layer when it reaches the output layer it outputs a prediction of what it thinks it sees what connects each of these layers are a set of ways each weight is a matrix and since we have three layers we have two weights between each layer what's the matrix please don't tell me it involves that movie with slow motion what you watch Hollywood movies on this planet all right never mind a matrix is a math word for a Google numbers that has a set of rows and columns just like our pictures we will multiply our input by our first weight matrix and that'll give us a new set of numbers then we take those numbers and feed them to what's called an activation function the activation function will output a new set of numbers and that's our hidden layer we'll take those numbers and repeat the process again using the new numbers as inputs input times weight activate just like that and it produces our output a prediction value that says what percent the neural network thinks it's a luma or not where are we what's an activation function it's a math formula that make sure our neural network can learn any kind of relationship there are two kinds of relationships linear ones and non linear ones that means if we were to graph the relationship between any two things the line would be either straight or squiggly an activation function make sure our neural network can learn both kinds of relationships but yeah we can add as many layers as we want to our neural network we can also try to modify our neural network to do all sorts of mathematical operations wait wait wait that's a lot that they can let just use a simple neural network for now agreed we can code this neural network in Python really easily in just a few lines of code each line represents a different type of layer that the neural network is using each of these keywords is a different math operation that the network is using the process of learning that network uses is called training we create our neural network as just two matrices and the numbers in each of these matrices starts off as randoms we're going to input our luma images into the network one at a time we'll do this 60,000 times it's called training we do our input times waiting operation then activate the result we'll do that twice to get our prediction then we calculate how far off our prediction was from reality and use that change to update both of our neural network weights ever so slightly we just keep doing that over and over again and eventually those random numbers we started off with in our neural network start to get really useful so when we give our network on new Illume oh it will know exactly whether or not it's a luma or not but monkeys they're here we're going to find the Duma how long till he's been training did you just say blasted monkeys yeah alright it just finished okay let me give it the satellite pictures they found it it's in this picture awesome I've got the coordinates let's teleport there there it is I've got it yes thanks Ron anytime good luck we need one of those on earth - speaking of Earth can you teleport me back now sure thing bye see ya alright it's good to be back there are three things to remember from this video was it's the traditional way of programming is to give a computer some instructions on what to do machine learning is a form of programming where instead of instructions we give a computer some data and a goal then the computer learns how to do the goal using the data and a neural network is a machine learning model that can learn from data and it consists of matrix operations input times weight activate and there are two kinds of relationships two data points can have linear and nonlinear ones that means a straight line or squiggly line neural networks can learn both types please subscribe for more programming videos and I hope you have a happy new year thanks for watching