AI Foundations for Absolute Beginners

This introduction to AI literacy course covers topics such as what AI is, machine learning, how machine learning works, and responsible AI. This course was developed by Henry Lee and the learn AI anywhere.org team. This organization focuses on creating accessible offline first AI literacy lessons that educators can use in schools across all types of environments. Their lessons have been used in over 70 countries from Low-resource settings to fully connected classrooms. Hi, and welcome to this offline first course introduction to artificial intelligence from learnaianywhere.org, where a global community of teachers and students, and we're super excited to

have you here [music] as we work to make AI literacy accessible to everyone everywhere. Now, the story we hear about AI is that it's humans [music] versus these AI machines, but the truth is AI Is really just a product of human choices. And unlike machines, we humans choose what we care about, like our values and our communities. >> [music] >> Our course goal is to give our learners the knowledge and judgment to become responsible creators and critical users of these AI tools. You're going to be learning on what exactly is artificial intelligence, [music] the key

parts of machine learning, how machine learning Works, and the ideas behind responsible AI. You're going to learn all this by designing your own AI classifier with Nearpod, which builds a strong foundation for understanding how many modern AI tools work. [music] We'll guide you step-by-step through our four lessons using our AI design activity sheet. And by the end, you'll design your own AI classifier and develop a critical understanding of how AI tools work, [music] ready to present Your project and your reasoning. We're super happy to have you here and shaping how AI is created and used

in our world from anywhere. We'll see you in class. Okay, welcome to the class. So, there's going to be two things you're going to need. The first thing is you're going to need to have a device that contains a webcam. And so, this device can be a laptop, it could be a tablet, it could be a phone. The important thing is it has a webcam Because we're going to be using it to take pictures. And so, the app that we're going to be using is something called Nearpod, and this is what we're going to be

using to design our image classifier. So, you can simply access this online, or if you want to access it offline, there is apps available for Windows and Android devices. So, if you're using an iPhone, you would have to access it through the website instead. The other thing is we have an optional project worksheet. And so, what this project worksheet does is that it's going to guide us to design our AI classifier as we go through each of the lessons. And of course, if you don't want to print it out, you could just use a simple

sheet of paper, and it will do the same job because the questions aren't very long. Okay, so that's all we need for the prerequisites. So, I'll see you guys in The next one. Okay, welcome to this introduction to artificial intelligence from learnaianywhere.org. So, before we go through all the lessons, you're going to notice some symbols on the bottom left corner. If you see a book, that's simply a symbol that we're going to be teaching something. If you see somebody running, it means we're going to do something interactive, like a game or hands-on Practice. And if

you see a cloud, that means we're going to have the opportunity to reflect on what we have learned. All right, I'll see you all in the next one. Okay, welcome to lesson one on what exactly is artificial intelligence. So, in lesson one's objective, you're going to explain the key qualities that make something artificial intelligent. And then you're going to identify the Examples of AI and not AI and explain why. So, before we even get started, think about this question. What do you think makes humans intelligent? Now, pause the video and just think about it for

a few seconds. Okay, now that you have some time, I want to first say that human intelligent is extremely complex, but we can look at two key traits. Some of the key traits that make human intelligent is its Autonomy and adaptivity. So, when I say autonomy, it's a human that acts and makes decisions on their own. And when I say adaptivity, it's a human's ability to learn from experience and improve. And so, a bunch of curious people wondered, how can we build these aspects of intelligence in machines? And that's where the field of artificial intelligence

came to be, a field of study about building technology that imitates intelligent behavior. And so, some of the key things that make AI seem intelligent is also its autonomy and adaptivity. So, when I say autonomy, I'm really just saying a technology's capability to assist or operate independently. And when I say adaptivity, it's a technology's capability to improve as it analyzes more data. And so, some examples we could think of is like our search engine. If you have ever used Google, you have engaged with Some sort of AI. It's autonomous because it can assist you by

suggesting words independently, as you can see as I'm typing, you can see I'm getting all these various suggestions. And it's also being adaptive because it can improve its suggestions by analyzing people's past searches and more. Now, a good example of not AI is a calculator. It's not autonomous because it only does what you tell it. So, you have to manually press the buttons for Everything you want to get done. And it's especially not adaptive because it can't analyze new data, and it always works the same way. Okay, now it's your turn. Let's play AI or

not. I'm going to describe a situation. You're going to observe whether it involves autonomy and adaptivity, you're going to decide if it's AI or not, and you're going to share your reasoning. Okay, so after playing the AI or not Game, here's a question for you. How were these AI examples helping people? So, feel free to pause the video and think about it for a few seconds. Okay, so some of the ways AI technologies can help us is it can improve the quality of our work. It can also do parts of our work so we can

focus on other tasks, or it can also make our experience simpler. So, some examples of people using AI tools to help them solve real problems could be Farmers. You know, farmers are using these AI tools like PlantVillage Nuru, where they can take a picture of a plant leaf and see whether or not it's sick or not and how to treat it. Some of these farmers also use something called Farmer Chat, which is a way for them to engage with a chatbot to get these advice on farming. So, it makes them have better decisions on how

they can take care of their crops and soil. If we look in the context of doctors, a Lot of these doctors are using AI imagings like X-rays to help them scan for those very difficult tiny signs of sicknesses, and it just helps them make better decisions. And then Suki is one of those things where the doctor and patient can have conversation and the AI will essentially take notes for them, and this allows the doctor to focus more on the patients. So, one of the big things we're going to keep emphasizing is that AI technology Is

a tool. And I say that not as a way to undermine the capabilities of AI, but more as a reminder that we are responsible for it. Because the thing is, people decide how to use them for the things they care about. Alrighty, project time now. So, if you remember PlantVillage Nuru, farmers used this AI classifier tool to take pictures of plant leaves to see whether or not their leaves were sick or not. And now it's your turn to design your own AI Classifier. In the next few lessons, you're going to design your own AI classifier

on whatever theme you choose. That could be plant leaves, that could be school supplies, it could be whatever you can think of. And so, some examples that students have done is that, you know, some of them went around their school garden to take pictures of plant leaves to distinguish between the various plant leaves in their school garden. And um some have Taken different pictures of materials like glass cups and plastic cups. And even some have gone creative where they did sign language, where they took pictures of various hand positions to see what letters they were.

So, after each lesson, you're going to write your response on your worksheet. If you do not have the worksheet, just simply use a piece of paper to record your responses. And you can also work together in groups if needed. So, this Is the worksheet, and by the end of this, you should have it all completed after you finish the lessons. Okay, so after all lessons, you should have completed your worksheet and design your AI classifier for your chosen theme. And so, your teacher may set up an optional presentation day where you present your AI classifier

and your worksheet responses. Okay, so let's start answering the first question on the worksheet. So, you're going to start with only answering lesson number one, which is choose your classification theme. So, choose a theme that you would like to explore for your AI classification and explain why you chose it. So, this is an example of what a student has wrote. I want to classify fruits because I love fruits and they are easy to find. Many people around me grow fruits, so I can bring different kinds to use for my project. Okay, welcome to lesson two,

where we're going to talk about the key parts of machine learning. So, lesson two's objective is we're going to be able to explain what machine learning is. We're going to identify the key parts of machine learning. We're going to describe why or what each part does and how they work together, and you're going to explain how data affects how well AI work. Okay, so how exactly do you prepare for an exam? Think about This just for a few seconds. Okay, so when a human is preparing for exam, the first thing you're going to do is

you're going to need to collect some information, and that information could come into the form of notes in class. So, once you have your notes, you're going to study those notes, and what happens is your brain is going to connect those new information to your lived experience to better make sense in the world. And so, what happens over Time is you're going to become better prepared to taking that exam as you connect new information to your uh previous knowledge. So, the four parts of prepare for exam is brains, that brain basically take notes, you study

those notes, and then as a result you have a more prepared brain ready to take the exam. And so, the brain is the thing that does the learning. The notes is basically the information you're going to study. The Studying is basically the time and process of connecting these new ideas to what you already know, and the prepared brain is the result. So, after you have done the the studying of those notes, you're ready to essentially take the exam. And so, human learning and the key idea of human learning is that you're building understanding by connecting

new information to your lived experience, so you can make sense of the world. So, how Exactly is human learning compared to machine learning? Well, the thing about machine learning, and we'll go to more details of what we mean by this, is that it's essentially using mathematical constructions to find patterns in data to make predictions. And so, some of the parts of machine learning is that there is a model, that model takes in data, and basically training is where you try to find patterns in that data, and then you have A trained model like a plant

disease classifier that could predict plant diseases, or whether it's sick or not. And so, the four key parts of machine learning into detail is the model is the thing that is going to be capturing the pattern. The data is simply the information we give to the model that is going to train on. The training is basically the process where the model uses to analyze the data to find patterns, and then again the Trained model is basically the results of after the model has finished training, and so it's ready to make predictions. So, even though the

machine process information differently, they actually still both share similar parts, right? So, brain to model, notes to data, studying to training, prepared brain to trained model. They're not exactly the same, but they do share that similar stages of how they essentially approach Their learning. Okay, so we talked about the key parts of machine learning like model, data, training, and trained model. But, it's really hard to see how these different parts come together. So, let's see how these key parts come together by designing an AI classifier with Nearpod kit. So, what exactly is Nearpod kit? Nearpod

kit is an AI app where you can add images, train a model, and see how AI recognizes things. And you have already seen some examples in Nearpod kit in the student example. You have seen students were able to distinguish between different plant leaves, they were able to distinguish between different materials, and even sign language as well, too. So, we're going to be learning Nearpod kit together by doing each step together. So, as you go through each step, feel free to pause the video so you can complete the task or complete the step Before you go

on to the next one. And of course, if you're working in groups or in a classroom, if you do finish early, just help others around you so we can move to the next step faster together. Okay, so the first thing I'm going to do is open up my Nearpod kit app, and I'm going to name the project school supplies project. It could be any name you want, it's just the name of your project. And once you click create, it's Going to bring you inside that project. So, I can actually have more than two groups. In

my case, I want to categorize between scissors and pens inside there. Now, when I create at least my two groups, I'm going to now add some photos for each of those groups. So, in my case cuz I have scissors, I'm going to take some photos of some scissors. In this case, I took approximately about six, and then I added photos. You could take more than that if you wanted to. And Then, I'm going to take some photos of some pens now. And in my case, I also took those of six pen pictures. And like I

said, you could have hundreds if you wanted to. In my case, I'm just going quick. Then you're going to add photos, and now you've added data to your groups. But, you can have more than two groups if you wanted to. So, technically I could create a group like erasers, but for simplicity we're going to stick with this. Then you're going to move into Training, and then you're just going to click on the train model button, and this is going to take a few minutes, so we're just going to wait this out. Okay, so let's test

our model now. So, what I'm going to first do is I'm just going to take a photo of my pen in a different position and see if it classifies it correctly, and you can see it says it's a pen. And then now, I'm going to test with my scissors in another different position, classifies It, and you can see indeed it calls it a scissor. All right, so that's your AI classifier, and you're always free to add more data to improve your classifier. Okay, so welcome back. So, after you had some time to play with Nearpod

kit, was your AI classifier always correct? And if not, why do you think it made that mistake? So, feel free to pause the video and think about it for a few seconds. Okay. So, the thing about AI tools is that they can give you unexpected results. What's going to be obvious to you isn't going to be obvious to an AI tool. So, the thing about AI tools is that they don't understand the same way as humans do, so they have to analyze every single detail on the photo to find patterns. And sometimes that means that

they find patterns or use patterns that we haven't considered before. So, for example, you can see over here that we Clearly have a scissor, but because it's in the same position um as the color pencil here, and it's the same color, it's classified as a color pencil. And also, the thing about AI systems is that they're just heavily limited by its data. They can only really predict based on what they have analyzed before. So, you can see our plant disease classifier fails because it has never seen or analyzed grass before, and it you can see

it clearly that it calls it a Healthy peach, which is not. And this doesn't just apply to images, but also languages as well, too, right? An a AI tool can only respond based on the languages it was trained on. So, in this case, I'm speaking in Hindi like, "What is the weather like today? What is your name? What is 2 + 2?" And because this AI tool has never analyzed Hindi data before, it's just going to keep repeating the same English phrase, which is what it knows, "I am good, and you." Repetitively. So, this really

just goes into this concept of like, what comes in comes out. So, machine learning, if the data is limited, that means the model prediction will also be limited as well, too. Okay, project time for lesson number two. So, you had some time to practice in Nearpod kit, but now it's your time to explore your environment and collect Data for the theme you chose in lesson one. And remember, your theme can always change as you go through the worksheet. Also, now you have to make sure to answer the question for lesson two on your worksheet. So,

for lesson number two is about data collection. So, you're going to tell us what items will you classify and how you're going to gather your data. So, an example response from a student is this. The items I will classify are bananas, Oranges, and avocados. I'll bring fruits from my home or go to the school garden to take pictures. Okay? So, you're going to respond to lesson number two, and also if your teachers allows it, you can also go around your environment to take more photos to add to your or to improve your AI classifier. Okay,

welcome to lesson number three. How do machines train? So, if you remember in the last lesson, We talked about the key parts of machine learning where you have your model, and in that model you give it data, and then finally go through a training process where the model uses that time to analyze all the data that you have given it to find patterns. But, how does the machine learning training process actually work? And so, that's what we're going to understand in lesson three. So, the objective is you're going to explain what an Algorithm is and

give real-life examples. You're going to explain why machines need algorithms to work. You're going to describe the steps a machine follows during training, and then finally you're going to compare how humans and machine process information differently. Okay, so to start this off, let's play the describer drawer game. It's going to require two or more players, but don't worry if you don't have another player with you and you're Doing this by yourself. We do have video demonstration of us playing the game, so you can see how it works. So, by the time that you're able to

find a person to play with, you'll be able to be ready for it, okay? But, let's still walk through the instruction until then. So, the way how the game works is that you're going to be working in pairs, so again you need to have at least two people. So, one person's going to be a describer, and one person is going to be The drawer. So, if you do have a teacher, that teacher is going to make a simple drawing with three to five basic shapes in any position. So, the drawing can look as simple as

this, and it can even look as complex as this. Um if there isn't any teacher, the describer is going to make the drawing. Now, once the drawing is ready, only the describer may look at the drawing. So, the describer is going to tell the drawer how to recreate that drawing only Through verbal instructions. They can only talk about how to recreate that drawing. Your hands should be behind your back, there is no pointing allowed at all. And so, during that time as the drawer tries to recreate that drawing, um after 2 minutes, the drawer is

going to compare their drawing with the actual drawing or the teacher's drawing to see how close they were to it. And so, finally after that is all Completed, that's one round, you're going to switch roles and repeat. So, the describer now becomes the drawer, and the drawer becomes the describer. Okay, so let's start playing this game, okay? So, let's try to do at least two rounds, and I'll see you guys in the next one. Okay, so we're going to be doing an activity called the describer drawer game. And the game is very simple. I, the

teacher, am going to be drawing a Very simple drawing that's just going to contain three geometric shapes. That could be either a square, it could be a circle, it could be a triangle, it could be a diamond, it could be anything, right? But it's basically three simple shapes that you can position any way you want to. And so what I'm going to do here is I'm going to draw maybe something simple as this. So this is a simple example that I'm Going to be drawing. And so what's going to be happening is as the teacher,

after I create my drawings, now I'm going to have my teachers or my students are going to be working in pairs, right? And so basically, uh when the students work in pairs, there's going to be two roles. There's going to be the describer and there's going to be the drawer. So who here is going to be the describer in this position? I'll be the describer. Okay, Then Karen's going to be the drawer. So again, the whole idea is that the describer is the only person that's allowed to come to the back of this hidden drawing

and basically come back to the drawer to give instructions of how to recreate that drawings. And so the trick is is that the describer can come as many times as they want to, however, they can only give verbal instructions and they must stand behind the drawer to give those instructions, So they don't see each other. All right. So let's actually show a demonstration how this game is going to work. So you guys ready? Yes. Steady, let's go. You can go out, Mason, see the drawing and give the instructions to Karen. You got this, Mason. Okay.

So the first thing is draw a square. Okay. >> [snorts] >> In the middle of the square, draw a circle. Is it going to be touching the edges of the square or in the center? Look. Um it'll be in the center. It won't be touching any of the edges. Okay, done. Okay. And at the bottom of the square, it will be outside the square and it would be a triangle pointing downwards. Okay. Using the bottom line as an edge? Correct. You would use the right edge and left edge and it will go exactly in the

middle as the point of the triangle. Okay. Nice instructions, Mason. All right. All right. Good job, guys. And as you can see over there, they were able to recreate that drawing. And now here's the thing, right? The whole idea of this game is that that was just really the first level. As a teacher, you can increasingly make this game harder by Like doing shapes like this, right? You can put your shapes in this position and maybe make a smaller square and put a triangle in the middle of this one, right? Basically, the key idea is

that we just want to be able to challenge our students to essentially give clear instructions to the drawers so they can clearly understand what to recreate. All right. Thank you, guys. Thank you. And we'll see you guys in the next game. Okay, welcome back. So if you had the chance to play the describer drawer game with someone or you were able to watch the video demonstration, our question to you is this. What strategies did you and your partner use or seen to make it easier to recreate the drawing? So just pause the video and think

about it for a few seconds. Okay, so now that you had some time to think about it, maybe you have noticed That they were using something called step-by-step thinking. So step-by-step thinking, so as you play the game with your partner, you had to give very clear step-by-step instruction for the drawer to understand how to recreate the teacher's drawing. So for example, the person may have said something like this. First, draw a circle. Then inside that circle, draw a small triangle. But you can tell that we gave very clear step-by-step Instruction. So what you just created

or saw was an algorithm. An algorithm is just any clear step-by-step instruction to completing a task. And the thing about algorithms is that it can apply to anything in life. It doesn't have to be technical. As long as that algorithm is a set of step-by-step instruction for completing a task. So for example, if I were to look at this this steps to making a cup of tea, you Know, I had the first step, boil water. Step two, I had to add a tea bag. Step three, I had to add sugar and milk. And finally, step

four, I had to stir and serve, right? So this is an example of an algorithm because I was able to break it down step-by-step to completing a task of making a cup of tea. So we use algorithms every day in our life whenever we do something step-by-step. That could be cooking, that could be getting dressed, solving a Math problem, or in our case, just playing this game as well, too. So my question to you is this. What is something that you can think about that you do every day that can have clear step-by-step instructions or

has clear steps? So just pause the video and just think about it for a bit. Okay, so now that you had some time to think about it, the thing that we really want you to understand, this whole idea of algorithms, is that machines need Algorithms. Again, machines don't understand the same way like humans. And that's one of the biggest reasons why we need to give a clear step-by-step instruction. So when you're playing that game or seeing that game, the describer is like the human giving instruction and the drawer is like the machine following them. Okay,

so now that you understand how an algorithm works, let's understand how we can create an algorithm for learning. So To better understand this, we're going to play a little game called which group does it belong to. And so what's going to happen is I'm going to show you a picture and you're just going to tell me whether or not it goes to group A or group B. Now the trick is I'm not going to tell you the pattern. I'm not going to tell you the rules. Your job is to essentially figure that out yourself. Okay?

So let's start with the first picture. Okay, so does this go to group A or group B? Just take a guess. This here goes to group B. Does this go to group A or group B? This one here goes to group A. Does one go to group A or group B? This one here goes to group B. Does it go to group A or group B? This one here goes to group A. This one Does this one go to group A or Group B? This one here goes to group A. Does this one go to

group A or group B? This one here goes to group B. Now here's a tricky one. Does this one go to group A or group B? This one goes to group A. And I'm sure you probably figured it out already, but does this go to group A or group B? This one here goes to group A. And finally, Does this one go to group A or group B? This one goes to group B. Okay, so what was the pattern? Well, if you have probably already found out, if there was any blacks at the left end,

that was group A. And if there was any black at the right end, that was in group B. And so I would think what's really interesting is like how do you get better at Deciding between whether or not it went to group A or group B when I didn't need to tell you any rules or when I didn't tell you any rules or patterns. So feel free to pause the video and just reflect on this for a few seconds. Okay. So the human learning algorithm can look something like this. Basically, you take a guess, then

you essentially check your answer to see how far off you are from the from The actual answer, and then you adjust your focus on other characteristics, and then you just keep repeating this process until you come to a point where you can distinguish between different groups without having to be given rules. And so the thing I want to point out about humans, what makes this really interesting, is that we look for what feels important. So as humans, we're just naturally very curious. We have These feelings and this strong desire to find meaning and connecting ideas.

So for example, if I just see a black square out of nowhere, I'm like, okay, well that black square definitely stands out to me. And like why is it in that position? You know, at the end of the day, I'm just so curious about how does this all connect and what does this all mean? And so when the human learning algorithm is in action, you know, what happens is Like first of all, you have to make an educated guess, right? Because you don't know what's going to happen, you don't know the characteristics. So whenever I'm

asked, does this go to group A or group B, you know, I just say, okay, well I guess it's group A since we're just getting started. And so this is where things get interesting. So at step number two, that's when you start checking your answer. So in step number two, when you Make a guess and you see the result, that is going to make you feel something. So for example, whenever I said that this is actually group B, you can see that I'm a little bit shocked. I'm like, huh, really? Maybe this has to do

with being on the right side. And so the thing about checking answers is that what it's going to do is give you emotion of either feeling surprised, maybe curious, interested, you know, whatever it is, you're just going to Ultimately wonder what this is all mean and how does it all connect together. And then step number three is when you start to adjust your focus. So based on how you felt after your last guess is going to affect what you're going to focus on in your next one. So, you can see over here that when I

ask the same question, does this go to group A or group B? I'm like, "Huh, okay. Well, it's on the left side now. So, maybe this has to do with being on group A." And when I say that this is correct, you can see I get immediately very excited. I'm like, "Oh, I think I'm starting to get this, right?" And so, as you keep repeating this process, over time, you're going to learn which of these characteristics matter most for telling the groups apart. Okay. So, it turns out we can actually simulate this learning algorithms into

machines as well, too. Okay. So, last time we talked about how We can simulate this learning algorithms into machines. And so, before we understand that, the thing we need to understand is what exactly is a feature. So, features are just pieces of information that help us tell things apart. So, take for example these two things here. Well, the first thing I notice is that they're different in terms of shape. One of them is a square, the other one is a circle. And they're also different in terms of color. So, This is actually a yellow square,

and the one that's on the right side is a blue circle. So, these are basically features or or information for me to help me tell these two things apart. And so, the way how humans see features is that, you know, anytime we look at a image, we basically break it down into different characteristics, and then we naturally focus on the one that feels most important, right? So, if we looked At all of these different features of this image, what we find is that because this black square position is by itself, and it just stands out,

it just feels like it's most important out of all of these other type of features as well, too. However, the thing about machines is that they aren't guessing with feelings, they're actually making calculated predictions. So, machines don't treat these features as feelings, they treat These features as numbers, right? They don't actually see or feel anything. So, for example, in order for them to actually process it, you know, they turn the white into one, and the black into zero. And so, a good way to really see the difference between humans and machines and how we process

features is that for a human, you know, whenever we see something that stands out like a black square, we're going to be like, well, That feels important, so we're going to basically take a guess. Now, the thing about machines is that whenever they You can see over here that they don't actually see it, they basically change that into numbers, and then they basically multiply that by some sort of importance of which of these things are most important at determining the outcome, and then they essentially make a calculated prediction. Don't worry about the math, right? The

important Thing I want you to understand is that there's there's just a lot of math that's involved in the process. So, the machine learning training algorithm, you can see it has like similar amount of steps where it's like you you make a calculated prediction, then you measure the error, then you adjust your math equation, and then you keep repeating this process. So, the thing about machine learning is that they basically calculate the prediction, Right? So, instead of actually seeing the feature, they take the feature, and they multiply it by, okay, how important are each of

these features at determining the outcome, right? Is it mostly feature one? Is it mostly feature two? And so on, right? It's trying to decide that to see how it gets to that answer. So, let's see the machine learning algorithm in action. So, step number one, you basically Calculate your prediction. So, they start by calculating the features as numbers, and then they basically add them together with random important values, cuz remember, in the beginning, you know, very similar to humans, we had to make a random guess. So, also these machines, they are just putting something random

of which of these features is actually important. And so, they come up with some sort of calculated prediction. So, then we After we make that prediction, then we have to measure error. So, what the model does is that it basically compares how far off the calculated prediction was from the actual correct answer, and this distance is what we call the error. And so, this error is going to be very important to us because what we do is that we use that mirror error, and we use it to help us adjust the importance of each feature

differently in our Equation, aiming for less error in the next one. So, basically, what we do is like we take that error, and then we basically, you know, update it differently for each feature to see which one is more important determining the outcome. And then we basically come up with some sort of updated equation that we're going to iteratively update. And so, after enough repetition, you basically have a trained model that can Make predictions. And so, you can see over here that we have a nice fine-tune math equation that can make some pretty accurate decisions,

right? You can see it says 87% group B. So, the key idea, and it's okay if you don't understand the math that goes into it. Like, the only thing I want you to understand is that a trained model is is based on math equations. So, trained AI models basically turn data into numbers, they do a lot of calculations into it, And then they also output a number which later is converted into something that a human can understand. Okay? And so, this is why every single detail of our image or data really matters because every detail

matters in AI models because every number matters in an equation. Right? AI doesn't process information the way we do, so its predictions may not actually match our expectations, right? So, you can see over here that each of these can be representing of a Number that we're going to be calculating in that equation. And so, this idea just applies to just in general any type of AI model, right? So, even like language models or conversational AI tools, they basically take your words, you have to turn them into numbers, you do math on them, and then you

turn them back into words so you can basically understand it, right? So, basically, a lot of it is doing a lot of math equations and calculations At the end of the day to come up with a prediction. So, the key thing that I want you to get out is that AI isn't magic, it's basically math that's guided by human choices. And I say this not to undermine it, but I I want it to be very distinct that it really is just math that's guided by human choices. So, for example, whenever we look at AI adoptedness,

right? AI adopts basically using the learning algorithms that was Run from humans that runs on machine. So, as we mentioned and as we saw, this requires a lot of math calculation, which takes a lot of energy. So, Nearpod is basically using the energy or or is calculating right on your device, right? So, a lot of the heavy math is going to use a lot of energy, making your device get warm, and it's going to drain some batteries, like so. Now, the reality is the environmental cost of AI is that while Nearpod works right on your

device, the thing about most AIs is that they run on giant buildings filled with thousands of computers, right? So, these buildings use as much electricity as small cities and billions of gallons of water just to keep themselves cool in order to do a lot of these different types of math calculations. So, it could definitely be a huge cost to our environment. The other thing is now AI's autonomy. Well, AI autonomy is like, you know, AI makes predictions by finding its own patterns in massive amounts of data. And so, as humans, we basically provide examples or

feedback to really help guide its prediction, right? So, for example, you were the one that collected data to train your Nearpod app. And so, there's actually, you know, in in the big scheme of things, there's a lot of humans that are actually behind these AI. You know, with Nearpod, it was just Either you or maybe a few other friends that were collecting the data, but most AI requires millions of people to clean up the data and to guide those AI models. And so, without those human choices, the machine just wouldn't know how to find those

right patterns. So, at the end of the day, really what you want to understand about how AI works is that AI tools are really just a product of human choices. Okay, lesson number three, project time. So, it's time for you to start thinking about your features. So, based on your lesson two data, I want you to think a little bit more deeper. Think about the key features that you used to distinguish your groups apart. Was it maybe the color? Was it the size? Was it the shape? What what other features were you thinking about that

distinguished your groups apart? Okay. So, you're going to answer that For lesson three question on your worksheet. And if you finish early, just keep improving your AI classifier. If your teacher allows it, go out there to collect more data so you can continue working on your AI classifier. So, collect more data if needed. So, an example response for lesson three, which is identifying features. What features will your classifier use to categorize the data? So, in this example, the student said this. I will Use color, shape, and size. For example, bananas are long and yellow. Oranges

are round and orange. And avocados are oval and green. These features make them easy to tell apart. Okay, lesson number four, can machines be responsible? So, what we're going to be learning lesson four's objective is we're going to explain how data bias leads to inaccurate predictions, explain why humans are responsible for AI and how to reduce harm. We're going to Explain simple ways to protect people's privacy, and we're going to recognize how bias, privacy, and responsibility connects to responsible AI. We're going to do a challenge called the barely a dog challenge, and your goal is

to build a model that tells bears and dogs apart that passes all test cases. Your constraint is that you have 3 minutes, you can pick up to 10 bears, and you can pick up to 10 dogs. To find this challenge, you go back to your Pocket app, you're going to see something called Try Bias Challenge, click on it, click start challenge, and you're going to immediately see that you can select up to 10 bears and the 10 dogs. And just like how you've been using your Pocket, after you collect your data, you go to training,

and then you go to your model where you're going to run all test cases and see if you pass everything. Okay, so let's get started with this Challenge, and I'll see you guys in the end of it. Okay, so welcome back. So, after playing the Barely a Dog Challenge, here's a question for you. What were your results, and why do you think they happened? So, did you pass all the test cases? Did you pass some of them? Did you fail all of them? So, just pause the video and think about it for a few seconds.

Okay. So, if you were able to pass all the test cases, this could be because of human bias. Now, the thing about humans is that we all see the world differently because, well, we all have different life experiences. And we call this limited perspective of the world bias. So, this image does a good job of explaining it. You know, you have these three people, and the person on the left has like the trunk of the elephant and thinks it's a snake. Um the person in The middle thinks it's a wall, and the person with the

tail thinks it's essentially a rope. So, they all have different experiences from the same elephant. And so, that is a reason why they have these limited perspective. And so, what happens is that our bias, or our limited perspective, can really affect the features we choose in our data. So, take for example the Barely a Dog Challenge. So, maybe someone might have just only picked black bears and White dogs because maybe that's what they're most familiar with. But, the reality is bears can also be white, and dogs can also be black as well, too. And what

this biased data can result to is something called algorithmic bias. So, whenever our biased data shows AI a limited view instead of the full picture, um this basically leads to inaccurate results. So, for example, you can see over here that remember in our dataset we made Sure that the bears were all black and the dogs were all white. So, whenever it saw an animal that was white, it associated as a dog. It anytime it basically saw an animal that was black, it associated as a bear, right? So, basically our bias was only showing a limited

view than the reality. So, yes, our bear was misclassified as a dog. >> [laughter] >> Now, the thing I want to really be up Front about though is that, you know, being biased is a perfectly human thing, right? It only starts becoming a concern whenever you apply it to others without considering their own experience and values. And so, basically our biased data could potentially guide our AI to exclude the very people that we're actually trying to help. So, some examples of this could be like inaccurate medical results, right? So, sometimes AI may analyze X-rays, and

you Know, if they are not able to analyze enough patient during training, they can actually make mistakes. So, you can see that it gave the incorrect diagnostic decision. And the thing about doctors is that if they overly trust the results, right, and they don't double-check, well, someone can actually get the wrong diagnosis, and that could be really bad. The other thing is is that there's actually been cases where there has been arrest, right? So, sometimes AI system Has misidentified people because they basically lack full training data on various different types of faces and race. And

so, this has actually led to people like Robert William being wrongly arrested as well, too. And this doesn't even just apply to just images as well, too, but also just language as well, right? Sometimes AI tools fail because their chain data doesn't have enough of that language, that local dialect, or that slang. There Was like a time where a man in Arabic was greeting good morning, but that was actually mistranslated as attacked him, causing the man to be wrongly arrested, but, you know, later released. So, there's tons of these different cases where when we're not

accounting for somebody else's experience or situation uh because of our bias that it can cause inaccuracy. So, here's my question to you. When Mistakes do happen, who is responsible, the AI or humans, and why do you think so? We'll answer that question in the next one. All right, so let's have a look at this question. When mistakes happen, who is responsible, the AI or humans? Now, if you haven't had a chance to think about this question, just quickly pause the video and think about it for a few seconds. Okay. So, one of the most important

Things that we want you to understand about these AI systems is that they're just tools. People are ultimately responsible. AI tools don't care about the results of their actions. From what we've seen, underneath it is a lot of math. It's trying to find patterns between its inputs and outputs through a math equation. So, there's no emotions or feelings involved. And what we have seen, these AI tools have been used to cause false arrest, to misdiagnose Somebody, and to even cause fatal accidents. And so, the reason for this is because it's just a tool. So, the

people who build and use it are ultimately responsible for what happens. And so, now that we understand that humans are at the center of this, what are some of the ways we can reduce harm from these AI mistakes? So, pause the video and think about this for a few seconds. So, we can look at this in terms of two Roles for humans and how they can reduce these harms, from a creator perspective and from a user perspective. Now, as a creator of these AI tools, it is important that you add a variety of high-quality and

diverse data to reduce the bias to the best of your abilities. But, remember, all humans are biased because we all see things in a different perspective. So, it's really important to listen to other people's different perspectives and to be honest about the Limitations of your AI tools. Now, if you're a user of these AI tools, it's always important that you double-check the AI tools' results when there's any risk involved in your decision. And to make sure to share feedback to help improve these AI tools to the creators. So, now let's work a little bit on

this exercise on how we can improve our bear and dog classifier. So, besides the black and white color features that we Talked about, where we talked about black bears and white dogs, what other features and variation of those features could help improve our classifier? Okay, so pause the video and think about this for a minute uh and see what other features and variety of those features we should also consider. Okay. So, there's tons of things that we could think about and how we can reduce the bias and and add more perspective to our AI models.

But, you know, as you've Seen, there's a limitation of bears and dogs that we have. So, you know, just considering different types of bears, different types of dogs, deciding whether or not those bears and dogs are in their baby and adult forms, the different environments that they're in, and especially the different sizes as well, too. And there's tons of things to think about, but these are some of the things that you can consider. Okay, so we understand that diverse data Matters, but just because we can collect data, should we? Now, data can come from anywhere,

just like how we could take photos in many places. So, for example, this picture of the park, we have photos of humans and dogs, and these are potential data that we can add to our AI model. However, not all data is available to use. And so, we can look at it in two ways, public versus private data. So, public data is information that anyone can see, but may Have rules for how it's being used. Now, private data, that's information that can identify someone or share something personal about them. And this always requires permission for use

and seeing. So, some examples of private data could be like your face, your emails and phone numbers, your health or medical info, name or signature, financial info, and even your home address. All of this information could provide personal data about you. So, I want you to think about this question. What if an AI tool was trained on your personal photos without your permission? How would that make you feel? So, pause this video and just think about this for a few seconds. Okay, so maybe you feel a lot of different types of emotions. Maybe you feel

violated. Maybe you feel embarrassed. Maybe you feel angry. The important thing to say is that no matter what emotion you have, you have a right To your privacy. So, you have the right to control how your personal data is being used. The thing about creators is that they should be very, very clear about what data they collect and respect your your choices if you say no. So, let's say that you had, you know, different types of private data. You can give permission to them to see your face, but may not give them any information to

see your financial info. But, the thing is there's been cases Where creators have misused private data and have faced consequences. There was a company called Ever Album that was actually accused of using people's photos to train AI models without proper permission. And the US government caught on and ordered them to delete their data and AI models. So, given that we understand this now, I'm going to ask you this question. How can we protect privacy when creating and using these AI tools? Pause this video And think about it for a few seconds. Okay, so we can

look at it in terms of two roles, from a creator perspective and a user perspective. Now, as a creator, you should always ask for permission before using personal data. You need to be clear about how it's being used and remove any personal information that is not needed. Now, from a user perspective, you should read how your personal data is being used, and you should ask questions on how your Data is being used. You have the right to your personal information and can say no or limit what is being shared. Okay, so let's do this exercise

and how we can protect privacy in our bear and dog classifier. So, let's say that we have this photo over here and that we want to add it to our bear dog classifier on this dog. How would you make sure this photo respects privacy before using it in our bear versus dog classifier? And just a Heads-up, this dog in this photo is not owned by anyone. Okay, so pause the video and think about this for a few seconds. Okay. So, the thing is there's a few private data that we found inside this photo. We have

this person's house address, their car license, and the person's face inside there. So, let's say that we want to use the photo as is. In those cases, we should always ask for permission. Now, the thing is once we actually have this photo, you know, the other thing is is that just remove all the private data that's not needed. The thing about our bear and dog classifier is that we only care about the dog. So, this is a way to protect the person's privacy by just taking all the data out. And then you can also blur

the private information as well, too, if you decide not to cut the whole entire picture. Okay, so we talked about this idea of Data bias and data privacy and how at the end of the day humans are responsible for these things. This is what we call responsible AI. It's about making sure people are responsible creators and critical users of these AI tools. And one of the most important things I want you to remember is that every one of us plays a role in shaping how AI tools are made and used. Remember, and this is not

just specific to AI, but technology in general, that Technology is really just the product of human choices. Okay, so it's project time for lesson number four, our last lesson. So, I want you to now expand your perspective. So, now think about the features that you used to distinguish between the different groups from lesson three and see how you can add more variety, explore other features, or even get feedback from other people to reduce that bias. Remember, no classifier is Perfect, and that's just part of how AI works. Explain its bias and explain how you might

reduce it. So, this is the last lesson. Congratulations. So, you're going to be working on your lesson four question, and if you finish early and if your teacher allows it, you can keep working on your classifier. All right, so this is our last question, building responsible AI. What can you do to add more variety to your features to Reduce bias? And so, this is what the student responded. I will take pictures of fruits from different places so the AI does not only analyze from one place, and I will also ask my friends to check if

it works well for them, too. And if I use someone's face or something private, I will ask for permission first. So, remember, again, back to the private information, if you use anything that contains any personal information, make Sure to ask for permission.