Demystifying AI: An Accessible Guide for All

1. Course Overview: Hello everyone. It's a beautiful day to be alive. Today we have a very amazing course that's going to make our life super, super easy. The course is titled I for everyone, the core of the course is to simplify a I make I, very simple for anybody no matter what your background, what your freedom of expertise, what you do in life. We are going to take I, which is a very complex term, very complex field, and simplified, make it easy for anybody involved and that's the goal of our course. A simplified approach which means everything going to be laid out in a very, very easy manner. So we can understand how you can use AI in your daily life, in your work, and in any one of your hobbies. So let's talk about this together. All right, so mainly the courts will cover the main building blocks which are introduction to artificial gellation. So what is AI? How can we define it? We're also going to cover machine learning and diplaring because these are some cubs that are used a lot by AI developers and people in the field. So we're going to simplify the, make them easy fundamentals of AI. So what are the main building blocks of AI? What is AI from technical and also from an intellectual perspective? How can we think about AI? And how, what is going on behind the neighborhood, if you will, behind the curtains. And finally, building projects with AI. So we're going to be thinking about a project. What are the steps that we're going to take note to build a project? And how can we build a great AI project following a very simple step by step approach? Then we're going to apply this project in a reward situation. Okay, So three main points. Introduction to AI, fundamentals of AI, and building a project using artificial intelligence. All right. 2. AI vs Machine Learning vs Deep Learning: Now let's talk about artificial intelligence by giving it a quick definition. Then speak about machine learning, and finally speak about learning. The goal here is to make it easy for us to understand artificial intelligence, machine learning, and learning from a very easy perspective, so no technical words here, nothing crazy. Keep it simple. The goal is to simplify it. And I also added machine learning, a deploying here, because a lot of times they are using articles, online news, et cetera, on newspapers, in your media. So it's always good for us to understand these keywords and make sure that we understand them because they're always used. And simplify them. So we can use them in our daily conversations and we can understand if we're building an AI tool, what are we talking about and how can we use it? So we start with artificial intelligence. What is artificial intelligence? Artificial intelligence, also known as AI, is a field in computing that makes machines think and act like humans. The goal here is to take a machine, computer, smartwatch, Smartphone, smart TV, anything that contains a CPU, Rap, anything that contains a computing, let's say hardware. Let's just think about this from a very simple perspective. Any computing tool and everything nowadays is a computing tool that can act and behave like a human. That's amazing. That's great. An example. Nowadays we have a lot of tools that can speak, that you provide them with some text and they can take that text and they speak to you. We have a lot of tools that can take a picture and understand that picture, which was really hard for computers to do in the early days. It was very hard for a computer to take a picture, understand it, understand what it is, what it does, et cetera. Nowadays, tools can take a picture and they can tell you exactly what's happening in the picture. We have a person running in a forest wearing the following clothing, et cetera. So this is very, very advanced stuff that we as human beings used to be very good at. We used to be, you know what, this is, the only thing that computers cannot do is understanding pictures. Understanding takes understanding language, understanding audio to nowadays. I can do that easily. So this is anytime that you see something that's human like a tool, a, a program that's human like that does human like behavior. For example GPT or open AI, GPD, it does the work. It can generate human friendly text. Text that resembles text that was written by human beings. That's AI. So this is a really quick definition for AI. I hope it makes sense. Second one is machine learning. Machine learning is the process by which computers learn and adapt from data without being explicitly programmed. So let me simplify the definition. Machine learning is a subfield of AI that takes data and learn from data without being program. Programming, if you think about it, is the ability to give a computer a set of instruction. Hey, computer, open this step. Hey, computer, open this tool, rate the tool, enter my name to my E mail into my phone. So we're giving the computer some instruction. That's a program. So anytime that you open a program in your computer, it does step by step approach. It works by giving an instruction. So let's say you have a media player that plays videos in your computer or in your phone. You open it, the program itself will automatically open your video, start playing it. You want to adjust the video and go to, let's say 50% of the video have let's say the video is 2 minutes and you want to go to the 1 minute mark, you can just drag and go to the 1 minute mark. So these are programs, they don't understand human logic or a human intuition. But machine learning takes data and understands how that data is working and understand the person in the data. I'll give you an example in machine learning, it is a very easy example called the churn rate example. So a lot of companies that are mainly making money from subscriptions, they want to understand the churn ery. They want to understand how many people are going to cancel the subscription next month so we can talk to them. Hey, why are you canceling the subscription and what's going on? Can we do something better for you in order to understand that generate or that unsubscribe rate? What happens is that the companies correct data about all the subscribers, give it to a machine learning tool. And the machine learning tool will predict for us who are the people that are going to unsubscribe, how it will understand the patent. For a person to unsubscribe, they need to be between this age and this age. They need to be using their activities between this time and this time. So this is really interesting because nowadays computers can understand the data and learn from it. So machine learning is data heavy, so we just give the computer some data, Hey, computer take the data and then the computer will understand it and we make it easy for us to work with. All right, now let's talk about Diplony. Ploy is a technique where computers learn to identify patterns and make decisions similar to how human brain works. Diplony is a more advanced, let's say, field than AI and machine learning where it's data heavy. So it takes more data than machine learning. And it works using, I'm sure you heard this neural network. So just a way that the computer understands data using the same approach as the human brain. In our brain, we have some neurons that are connected to each other so we can understand, remember, et cetera. We developed a program that does that, but that program only works if you have a lot of data. The churnet example can be applied here. So the unsubscription rate can be applied here. Also. In this case, computer vision. So understanding a picture can be applied here. Voice recognition can be applied here. So pron is just a subpheld. Take about as a subpheld that chooses program a tool that resembles to the human brain. Very simple. The only difference between AI machine and Per requires a lot of data. You need a huge amount of data for the computer to understand the passion, what's going on here? Why is this user making that decision? Why is this image different from another image? So this is the difference. Ai thinking like a human being, machine learning, programming a computer program, by giving it data so it can understand how to make decisions and deep learning. It's a technique, a subfield of AI, where we have a program called neural networks, or neural nets, that resembles things like human brain, you know, and it needs a huge amounts of data to understand patterns and make decisions. So these are three fields, simple, easy to understand, don't give them more thinking that they deserve, very, very simple. All right. 3. AI vs Machine Learning vs Deep Learning Tasks: Now let's speak about the tasks performed by AI. What are some of the tasks performed by AI? So we will be speaking about the tasks performed by AI, the task reformed by machine learning, and the tasks performed by in AI. What can we do? We can do image recognition, so we can recognize images as I told you. So we have an image computer and the computer can't understand what's going on inside the image. This was really held for computers, speech recognition. We can easily give the computer instructions. He computer, Can you please open the file? Hey computer, Can you play the video? Hey computer, can you do this? Can you open my browser? Can you go to my social media? So we give the computer instructions and the computer should take our audio, convert it to instructions, and then perform the instructions in a very accurate manner. That's what super hard for computers nowadays, it can be done. Recommendation systems, these are systems like Netflix, so you watch a movie, Netflix will understand your preferences, what kind of movies you like, and you're going to get recommendations from similar movies. Or recommendation systems are very, very powerful. They mainly take your behavior and based on your behavior, we'll recommend the movies or anything that resembles to what you're doing. So these are also great, great, let's say application of AI, autonomous vehicles. So these are vehicles that can drive themselves. So we see a lot of companies that are currently developing autonomous cars and vehicles that can easily understand the road, understand the vehicles that are perhaps across the road, et cetera. And they can do the task without any human intervention. They are very powerful, but they are still being programmed and being trained. Robotics, all robots that you see in the world are all pace because they need to see computer vision. They need to talk. Special recognition, they need to understand audio recognition recommendations. They need to understand your preference so they can talk to you in a way that makes sense to you. Autonomous, also capability. So they need to be able to move around easily. So these are some of the advancement that we're going to be used in the next couple of years, and I want you to keep an eye on them. All right. Now let's talk about some of the tasks performed by machine. We have spam filtering. Spam filtering is the ability for a computer to filter out your e mails from spam and non spam. So we have an e mail the computer without your intervention, we'll read the e mail and see, okay, this e mail contains a bad link. This e mail is not, doesn't locally it, oh, this is spam. So this is really advanced and this makes our life really, really easy. How it works. We give the computer a bunch of spam and non spam data, and the computer will understand the passions of spam data or spam emails and it will in the future protective and email is going to be spam on. So this is a great application for us to simplify our life. Sentimental analysis is mainly used in social media. So when you are a company or you have your own brand or your content created, you want to understand how your people or how perhaps your users feel about your brand and feel about your content. So if you have a sentimental analysis tool that will take let's say Tweet or Facebook posts or anything. And it will tell you if your customers are happy. They're sad. They're mad. They're joyful. So it will tell you what's going on, so you can take action. So if let's say customers are mad, you can talk to them. If they're happy, you can also take them. So sentimentalysis is the ability for the computer to easily understand the sentiment of an individual using their expressions, their behavior and their text. Okay, fraud detection, this is mainly using the financial industry to understand if a transaction is fraud. So we can detect it and we can stop it. So it works easily by giving a computer what are some fraud transactions and non fraud transaction. And the computer will understand which are fraud and which are not fraud. Image classification is the reason for the computer to take an image and classify it. So this is, let's say, a computer image. This is an office image, this is a car image. Very, very nice. Stock market analysis is also another sub field where the computer will easily use the machine learning. Try to predict the price of a stock or the pride of any financial securities just by looking at the historical data or the data or that financial instrument without any issues. Okay, So very simple applications that are revolutionizing the world, like every big corporation out there is using these tools, is using them on a daily basis to take advantage of the situation. Because if you're not using AI, it can be used against you. So it's really nice to use AI and think about some of the applications to know exactly when we're talking about AI and when we're talking about a simple computer program. All right, now let's talk a little bit about learning. Learning is more advanced, as I told you. It uses normal network. So it's another way, a program that understands data using the same approach as a human brain. And it involves in the following tasks. So we have natural language processing, it's very, very important field that understands human language. So you give the computer some language, it will understand the words, the way that you write sentences, the text. And it can generate text for you. Ors can give you a response based on your answer. So natural language processing, Anamaly detection. So this is a great, great way for the computers to understand anything that is abnormal. So let's say abnormal behavior by users in your account. Abnormal behavior by financial instruments, abnormal behavior by your computer. So this is really used for cybersecurity and also of the financial industry. Anamaly detection to detect any abnormal behavior, facial recognition. So this is also great for, let's say, security aspects. So you want to open your computer or your smartphone using your face, Facial recognition. We understand your characteristic, your facial characteristic. And we'll open your computer or your laptop if it's really who is opening the smartphone, the computer. Customer segmentation, another subfield of deplaring or application of deple, where we have a pulse of customers and we segment them based on some behavior. We have customers that like product A, customers that like product B, customers that like product, so on and so forth. Text generation, so this is what everybody's talking about nowadays like Ch GPT. So when we generate text, we ask the computer a question and we get some text. Hey computer, can you please write 2,500 articles about Machelleuniformy? And the computer will write this beautifully with crafted article. We have a lot of tools such as GPT that can do that nowadays. And they can do it in a very, very advanced manner, so you don't even know if it's a human or an AI tool. All right, so these are just some tasks, some applications of deep learning. And I wanted to share with you deaf so you can see. Okay, so what is, what's not, what's machining, what's not machinery? What's ply and what's not deploy? So we can make that distinction between I and just simple computer programs or software. All right, so this was it for the tasks. Now let's talk a little bit about the main I buildpl. 4. Building Blocks of AI: So what are AI building blocks? I created a very simple chart for you that takes four building blocks of AI. So in order to create an AI tool, you need four main building blocks. We have data algorithms, computional power, and deployment, and maintenance. Don't worry, I will make sure to explain each one of them in a very, very easy manner. But these are the main building blocks that you can think about. Every time that you're thinking about AI. You want to develop an AI tool in your company, in your job, in your daily task, as a content creator, in your personal life, something for you and your family. You can do that using AI. But you have to think about the main building block data. Do I have data to do that algorithm? Which algorithm should I pick? Which tool should I pick, if you will, Computational power. Do I have a strong computer to do that deployment and maintenance? Where can I deploy? Where can I perhaps put this tool so I can use it in my family or my co workers can use it? Okay. Very simple. Very easy. Now let's dive deep into each one of them. 5. Data: Data. Data is the fuel of AI time that you hear about data, this is the fuel of, without data, we will never have AI. Ai needs data to function without it. It cannot operate. If you remember the spam e mail example. In order for us to develop an AI based spam filter, a tool that will filter your E mails from sperm and non spa, We need a tool, or we need that tool, to, to get a lot of e mails and for us to filter out, this is spam and this is not spam. So the computer can understand the versions of a spam e mail and the persion of a non spam e mail. So this is very important. It's the fuel of data. Also for the financial transactions, we need a lot of historical data, but a stock for example, to understand, okay, how the prices behave and move Quality and quantity matter data itself. I want you to always please do me a favor. Think about quality of data is the data that I have. Quality data, for example, if you have a lot of spam data, but it's not clean, it's not high quality. There is no pure distinction between what spam email and what's a nonspam mail. Your AI tool will not behave really well. So it's just like us as human beings. We learn from books, we learn from courses. If the course is high quality, you will learn easily. So the same because AI is learning, we're training AI. The data itself should be high quality. Better data will lead to a more reliable tool. That's why sometimes we'll have two companies that are performing the same task. Text generation from company A and text generation from company B. Company A is doing much better. The only reason that they are doing that is because of their data. The data that they use in order to train their tool is high, high quality. Okay, now let's talk about data preparation. Whenever you talk about data, there is always a task that goes hand in hand with data, which is data preparation. So the data, once you have it, let's say you are in a company and you tell your colleagues that you need a lot of data to develop a tool. That data should be exported or taken from your colleagues or let's say you want to develop it in your house, take it from your computer or smartphone or whatever, and then clean it. Remove anything that's not going to be useful if let's say it contains times spam, or let's say a date or whatever, you need to remove that. If the data contains a description that's not going to be used by your to remove it. So clean the data, make sure the data is pure. So the case of e mail, we have the e mail title. The e mail, let's say subject. And then we have the e mail content or the e mail body. We give it to the computer. The computer remove, let's say, anything unrelated that would not help the AI tool understand and take action. Sometimes the way that I clean data is I ask myself the question. If I was doing the tasks manually, will this data help me in my task? So think about cleaning the data from a human perspective. I, we're always going to use human intelligence in this case. And it's really good for us to clean it and follow the best practices in order to clean the data. Okay, now let's talk a little bit about some examples of data in AI. So whenever we talk about data in AI, we have four main distinctions. And these are just examples, we have text data, so any type that you have text based data, so let's say e mails, books, articles. These are text data. You can give it to AI, the AI will understand it and it will give you some output or some results. Image data, so this is really good for computer vision if you're trying to develop a tool that will understand images. So you can easily give it a bunch of images in any format and the tool will understand it. Numerical data. So we have a lot of data. It's a stock market data, financial transactions data. Let's say if you are a teacher, you have some grades from your students. You can just give it to the AI tool and analyze those grades. So this is really great. Numerical data is by far the number almost used data in the AI field because it's easy to understand and it's also matematical, so the computer can easily understand it on audio data. This is for audio text. If you want to generate some audio or the computer to understand text that you give it in a speechal recognitional manner. So we have an online assistant or an AI assistant, so you can just speak to it. It will understand what you need and it will perform the task. So those are all some of the main data points I want you to remember. So we have text data like an e mail image, data like your picture, numerical data like your financial transactions, and audio data like your audio recording or an online assistant. So these are some examples that I want you to remember. 6. Algorithms: Now let's talk a little bit about algorithms. The second building block of AI is algorithms. The brain of AI is the algorithm. Ai uses algorithm as its brain to process the data and make decisions. Once you have the data, of course, we need a way for the computer to process the data. We have an algorithm that does that. Algorithm is just a set of steps to tell the AI how to process the data and understand the path. Don't worry, all the AI algorithms are open source, available online. If you want to develop an AI tool, you don't have to develop the algorithm from scratch and build it and test it. Every great algorithm is available online, open source, that you can use, implement, and start working with it. So that's the beauture of AI. Second point that I want you to think about, anytime that you're thinking about algorithms, types of algorithm. Ai has different algorithms for different tasks. Machinery predicts with historical data, while learning is great for image and special recognition. So think about it. So for every task that you are performing, you have to think about, okay, what type of algorithm should I use if I'm trying to process images? What are the best algorithms, the best tools that can take my data and process it for an image task. So you always need to take about a good algorithm that can do the job. So just link in the puzzles. This puzzle goes, this puzzle. How can we do it easily without any issues? Okay, Learning from the data. So I learns from the data using different algorithms. So we have supervised algorithms and supervised algorithm reinforcement algorithm. Don't worry about the names. The goal is to simplify the terms. So you just need to understand that your goal is true. Number one, prepare the data. Number two, take the data and find what is the best algorithm that can perform a great job for my task sample. Don't even worry about it. Okay, next. Now let's talk a little bit about some examples of algorithm. So these are just some examples. I just put here four examples for us. We start by linear recreation. Linear regation is an algorithm available online, open source, that we can all use to develop an AI to, that takes numerical data and predict future data. An example would be sales. You're a company and you want to predict next month sales. You will give an AI to sales for the last ten years and you will use linear recreation algorithm available online for free. Once you give it to the linear regration algorithm, the linear regulation tool, if you will. The computer, we understand the data and there we go, we're going to have your answer. What's the next month's sales? The computer will do the work for you. So anytime that you are using or dealing with numerical data, stock market is a great example as well. Give it to the computer and lead the computer to the work. Okay? Decision trees. Decision trees are amazing because they are used for classification. Let's say example. So you have a classroom with a lot of students and you're trying to predict which students are going to do creating an exam and which are not going to do creating an exam. What you can do is give an AI tool, a decision tree. In this case, all that data and the computer will predict this is going to pass and this student is going to fail. So this is a great tool if you have a decision. So you're trying to predict if a customer is going to subscribe and subscribe an e mail or so. Spam, spa. We have a tree. We have a decision tree that's going to be built based on your data, don't worry about it. It's available online so you can use it without any issues. And this is really great for non numerical numerical decision. Anytime that you are doing something that is non numerical, it can be put in a decis entry. Neural networks is the one that I explained previously. So I can do it again. No worries. So it's mainly an algorithm, a tool that uses the human brain functionalities in an algorithm. The people that developed a neuron, they went and they studied the human brains. In the human brain, we have some neurons, They connect together to make a decision. So let's develop a tool that uses the same approach, and let's try to give it to a computer and see what happens. It's just an emulation as a representation of the human brains. That's why it's called neural network. So it's a network, it's a bunch of points that are connected with each other. Takes data to make a decision. This is really good for image recognition. It's amazing for text generation. Any task that is very, very difficult to do, you give it to a neural work which is available online for free and it does the work for you. A random forest is a bunch of trees. So if you have a very difficult decision that you're trying to make, so we're trying to protect, let's say in these particular clients, which client is most likely to buy a product from us? So this is a very difficult decision because who buys and who doesn't is not an easy answer. Now it will take a lot of data at our purchasing behavior when, how is the client always purchasing from us in July? So we have to think about these questions and give them to the computer. Let's say we collect a lot of data about the customer and we give it to the computer in the random forest format. And there we go, we are going to have our answer using random forest. So it's also good for decision based algorithms or decision based problems, and it's great for complex decisions. Linear cation, numerical data, decision tree and random forest, non numerical data. Mainly decisions norangework for complex tasks such as computer vision. You're trying to understand an image, voice recognition, you're trying to understand the voice, speech recognition, et cetera. So these are some of the examples. All the best algorithms are available online, so you never have to program them from scratch or learn about programming, et cetera. All you have to do, use them and you'll get your up okay. 7. Computational Power: Now let's speak a little bit about computational power. So in this slide, I want to focus mainly on why do we need computational power to develop an AI tool? All AI tools, all AI programs, they need computational power to process the data. The idea here is we need a very strong computer, a very strong hardware. You know, I wouldn't even go to the DTSCPU Ram. Let's keep it very simple. We need a strong computer to take that data, process it, and understand how it operates. And this is a main baiting block of AI, because without Dutch computational power, it will be hard for the computer to process the data. Let's say we have billions of data points, billions of let's say customers that we are trying to process. Of course, we need a very, very strong computer with a strong B CPU memory and all the hardware techniques, a computer computer in order to process that data and give us an answer. So it's the engine of AI. Without a strong computer, a strong tool, hardware itself, we are not going to be able to do anything. So AI is going to be used on a daily basis by a lot of companies and a lot of people. So the use of computational power is on a growing demand. A lot of companies and a lot of individuals are now using cloud computing. So you go to Amazon or acrosoftazu or Google Cloud. You go to one of the cloud providers, one of the big cloud providers, and they will give you that computational power. But you have to pay for it, obviously. And this is something that you need to understand that because AI needs a strong computational power, there is a growing demand. Everybody is now trying to allocate and pay for the computational power in order to use them. And have an amazing, amazing experience with developing hydro. So, because computational power is in demand, a lot of companies like Intel and all the computational companies out there that are developing hardware are now creating a more advanced hardware. So a better processor, a better memory, a better graphical processing unit. They are creating a lot of these amazing tools and amazing functionalities that can be used by AI developer by us in order to do the work faster. So that's the power of it. In this age that we're living in, we are in an amazing opportunity. Why? Because now computers are strong, now we have a lot of data, and now we have a lot of gums that we can use. So we are in a perfect time, in a perfect place to use AI in our daily life, in our businesses, and everything that we do. All right, so think about computation power as what are some of the hardware things that I need to do in order to get the job down, in order to make my life easy without any issues. And we can now do that easily because of companies such as India, such as Intel. Because they are providing us with all of these things and we just can use them without any issues. So that's something to keep in mind. And they perhaps play a critical role in the development of A, and advancement of AI. Because if you remember with me just a few years ago, we didn't have strong computers and you had to pay a lot of money to get a very strong fast computer nowadays, you can easily get it or you can just, if you're developing an AA tool, you are going just to go to a cloud provider, pay them some money, and they will allocate for you that resource where you can put your AI tool and make it work. Okay, now let's look at some of the examples of computational power. We have quantum computers, These are types of computers that are still developed. They are advanced computers that are super fast and they can be used a lot in cybersecurity. So these are also great for AI because they have super speed, fast computing power. You know, they cannot be used nowadays in a fast scale because they're being developed. But they are the future. And I just wanted to share with you that they exist and they are out there. Cloud computing is the main part of the AI development life cycle. Why? Because we have a lot of providers that can provide us with these computers. They have data centers. They have people that are managing these data centers. And they provide us with this computation power and we just pay. And they provide us with you services. So AWS, Microsoft Azure, Google Cloud, all they can do that for us without any issues. Graphical processing unit. Graphics processing unit, so GPU. Graphics processing is a tool, is a hardware that processes the data in a very fast manner. So it's in every computer, so every computer our day we'll have a GPU and Text In Vida and other companies that are working on GPUs, we have very fast GPUs that can take data processing. So thanks to the advancement of GPUs, we can process a lot of data and we can have great, great results. We have TPUs, or Tensor Processing units. These are just AI accessed applications developed by Google. So Google is a company that currently owns, let's say, a tool called Tensor Flow. It's a library used for AI. They also provide us with TPUs. And these TPUs are able to create applications that are dedicated for AI. So anytime that you have an AI system, you can use Google TPU. They can, because they're Google properties, so they can be mainly using Google Cloud and Google infrastructures. And they can also play a major, major role in computational power and AI. So think about, these are just examples. You don't need to go deep into them. At the end of the day, let's say you are developing a tool. You're just going to a cloud provider like Google Clouds or AWS allocate some resources from them and everything's going to be taken care because they already have the infrastructure. Okay, So quantum computers are very, very advanced, they do the work for us. They are mainly used for cybersecurity and they're still under development, so we still don't know what they can do. Cloud computing, mainly the ability to use huge computational power and huge computational resources using a cloud provider. So just like AWS Zu, Google Cloud graphic processing units are small hardware units that are used to process the data. Tensor processing units are Google provided processing units that work on also analyzing the data. And they are mainly used for complex tasks. Complex I task, okay. 8. Deployment and Maintenance in AI: Now let's talk about model deployment and maintenance. Once you are done and your tool is working, your AI tool, let's say you're developing an AI tool that helps predict, let's say, sales in the next month. So you have a company and you're trying to predict sales for the next month, okay? What you can do is easily, without any issues, just create a tool that can predict sales based on previous sales. Very nice, very easy. What can you do? What can you do in this case is take the tool and deploy it. You have to put the tool in the cloud or in your personal computer or anywhere where it can be used and accessible. If you're the only one that is going to use it, you can deploy in your personal computer and work with it. But if you are providing the service to allow people, and this is perhaps the significant example or the main uses of AI, What you can do in this case is just easily, without any issues, go to a cloud provider. So AWS or Google Cloud sign up create an account and then buy some resources and use them. So deployment is the ability to take an AI model, AI tool that you developed and put it in the web, put it in an infrastructure, a cloud provider, let's say, or in your own computer, so it can be used by you or anybody around you, okay? Scalability, Anytime that we think about AI, there is a very, very good concept that I want you to understand, which is scalability. You have to make sure that your AI tool can be scaled to as many users as possible. You came up with a great idea, which is a tool that is used by businesses to predict their sales. The tool, once you set, deploy it in our website using Google Cloud for example, it should be able to scale. It should be able to maintain the same level of performance no matter how many users are actually using the tool. So if you have 100,000 users that are using your two simultaneously, 1 million users, it should work perfectly without any issues. And that's why it's recommended to use a cloud provider because they can do the work for you without any issues model monitoring. So monitoring AI models is essential to catch and fix issues caused by changes in the real world data. So it's really good for us to always monitor if our AI tool is doing great work and get feedback from the user. So once you have the sales data, I am going to be 100% operational, working perfectly. Of course, you're going to get some feedback from you, use that data, give it back to the tool, and the tool will improve and we'll do a better job. So that's something to keep in mind and that's something that is very, very important. These are some of the examples of deployment and monitoring tools. And I, these are just tools, so I don't want you to be perhaps intimidated or these are just names of some tools that you can be using that are all used for AI deployment and maintenance. In order to take your I put it in the marketplace. Okay, we have Docker is an open source tool that can be used to deploy your AI without any issues. We have Azure Machine Learning. It's also a cloud service provided by Microsoft Azure that can be used in order to deploy your machine learning tools. Okay Cubes. It's also a tool that goes hand in hand with Docker that is used for scalability. You have a lot of AI technologies and you have a lot of users that are using these technologies. Cubans is a tool that you install that will make sure that your users are all happy with the performance that, let's say one of your tools goes down. You have a replacement easily. So just a tool to maintain scale, make your users happy. Flow is a tool that's used to maintain. So monitoring maintenance, making sure that if there is a puck, you get notified, you can fix it. Some tools that can be installed that can be used without any issues, that for every one of them we can do a Task docker for deployment. Let's say I'm done with working on my AI tool. I want to put it online. I can use Docker Azure Machine Learning is a sub field, let's say a sub service of Turqets. I can just go to Azure, pick that service and I can buy it and deploy my AI to Cubonetis's, what you call an orchestrator. So pretty much it monitors all the AI services that are running. And if one of our services goes down, there is a replacement. Scalability is also there. We have a lot of users, no user shifts that performances beds or really good airflow for monitoring. Making sure that all the services are working perfectly without any problem, okay. 9. Building an AI Project: Now it's time for us to talk about the AI project life cycle. How can we build an AI project from step number one to step number five easily without an issue? We start first, we have to identify the problem. So before you do anything, you have to sit down and you know what. Let me think about the problem. What am I trying to solve? Am I trying to protect sales? Am I trying to understand images? Am I trying to create a tool that enables me to fix or remove all spams by filtering out the spam? And as well so identify the problem, understand what kind of problem I'm trying to fix using it. So this is a very important step because if the problem that you're trying to affix is not clearly defined, what will happen is that you're going to have a mismatch. Okay? Collecting and preparing the data. Once you identify the problem and you understand exactly what you're trying to solve, the next step is you have to collect and prepare the data. We have to go and start collecting the data. If you have a company you're working in a company, it will be easy. You can just talk to your colleagues and get the data. If you don't have the data, you can go to an open source data provider. So go online, find free data that you can use without any issues. So open source data, if you don't have the data, you can also go create a survey, talk to your customers and get some input from them. So collecting and preparing the data, collected data, put it in your computer or put it in a cloud provider. And then clean it. Make sure that you remove all the stuff that you don't need from it. Choosing and implementing an AI model. So once you collect and prepared the data, the next step is I have to pick an AI model and I have to implement it in my computer. So as I said, so AI models are available online, so all AI technologies are open source. So if it's a numerical tool, you can use the sellinar regression. If it's a decision you can use decision tree. So just pick and choose over this case, what can I use once you do it? You can just implement it. Whenever you implement it, just run it. You just have to click on the Run button to get it to run training and evaluating the model. So once you implement a model, what you need to do is you have to take the data, put it in the model to train it too. They need to be trained. You have to give them the data so you can understand the situation and based on the situation, they can make decisions. So we train it and we evaluate if the model is doing well. Is it going to correct if it's giving us the correct answer all the time? These are some of the questions that you need to ask and we have a lot of tools nowadays that are free that can do the training for us and also the evaluation, so you don't need to do any heavy lifting, deployment, and maintenance. So this is the last step. We take the tool, put it online, let's say in a cloud provider. If it's offline, you can just put it in your own computer, put it in your smartphone. So just deploy it in a particular system and start working with it. And make sure that you maintain it. Make sure that if you have new data added to it, that's very, very easy. So we start with it and define the problem, then we collect data and we prepare the data. Then we choose which model we're going to use. We train it and we evaluate it. And finally, we deploy it and maintain it. Okay. 10. Identifying the Problem: Now let's talk about perhaps a real world situation where we are going to implement the step by step approach. And we are going to understand how we can easily use AI to solve the problem. We have been receiving a lot of emails and some of them are spam. And oh my God, we have so many bad emails that we don't need to like, we need to find a solution. We're running a small company as a team. And we decided, you know what, let's create a I spam fat. Sounds like a good idea. And that tool can also be used by our colleagues and other partners, let's say in our company partnerships with other companies that we have. We sat down together and we said the problem is we're trying to develop a tool that will help us solve the issue of receiving spam, e mails on a daily basis. Okay, how can we think about the problem? In a very easy diagram, I receive an e mail. The tool will simply tell me if it's a spam or non spam. Very, very easy spam, not spam. Okay, now let's talk a little bit about identifying the problem. The task is to automatically identify and filter out e mail. This is a very classification decision problem, so we can use decision tree in this case. And it will work 100% because we have a decision to make. I received an E mail, is it a Spa or not? Span Super easy. Super good. So we start with identifying the problem. We want to automate the detection of Spa images in our system. So we have a small team and I would like to just identify if the images that our company is receiving. Span or span. So far so good understanding the context and constraint. So we have to make sure that we understand these two points which are the context and constraint. So do we have a budget to do this? How much time will it take us to develop the system? Do we have people who understand I a little bit, who can work with AI, who can do the work? We have the technical team to do the work. This is also part of the problem identification, to understand our environment. Can we do this? Can we solve this problem Yes or no, Setting success metric. So we have to make sure that for this system, in order for us to say, wow, we did it. Our system is working, we are successful. What is our success matric In this case, use this satisfaction level. So we are going to leave the users once they use our system with a user satisfaction. Let's say quick form where they can just say if, if they're happy or not. So if you're happy with the system, give us a thumbs up. If not, give us a thumbs down. Okay? So these are some of the question I just want to put you in the mindset of thinking about AI in a very, very simplified manner, all right? 11. Collecting and Preparing the Data: Now let's talk about collecting and preparing the data. So we have to first correct the data. Take all the emails that are received by our company. Anybody, any employee that is receiving emails in our corporation. We're just going to correct the cleaning them. Remove anything that we're not going to be used. Images from the emails, like anything that is not going to be relevant, MGs, whatever. Clean them, make sure that the purely subject and text and then splitting the data, this is something that I left out for this part of the slides or the course, is that once you have the data, you have to take your data and split it into training data and testing data. So let me explain. You give the computer the lesson he computer learn the difference between spam and non sperm email. Okay, so far so good. I'm happy. Then once the computer learns, you have to test if the computer actually understood what you meant. How we have a small testing data, we can give it to the computer and we see if the computer is going to be able to detect if the email is sperm nonspam. So once you collect the data, you need to split it into a training set. So obviously, the training will be huge. So let's say you have 100,000 E mails. 70,000 is going to be given to the computer to learn. 30,000 is going to be used in order to train, to test, to see if the computer actually unders, and this is the same concept in the scoring system. So we have, let's say, a semester of learning, so we learn about history, about it said a language about computer science. And then at the end of the semester have a quiz or an exam, a mid term, a final, whether we test our abilities, if we actually get the material or not. So this is the same concept we just, it was modeled from the human behavior. As you can clearly see, where we take data, we split into training and testing, okay? 12. Choosing and Implementing the AI Model: Now let's talk about choosing and implementing AI model. So the first step is we have to choose the model. So for every situation, as I explained, we have to think what is the best model for this case for spam? It's very easy. It's decision based. And anytime we have a decision, we need to think, okay, because this is a decision, we are going to choose a decision tree because we're giving a computer some data and the computer is thinking spam, non spam, very binary, very easy spam and non spam decision implementing the model with which we can use let's say Python for example, which is a really famous programming language for implementing these models, Give it the data and Python can implement the decision three for us without any issues. Okay, so once we choose the model, we have to understand what are the tools that we're going to use. Of course, in this case programming language. Are we going to use Python, Java? Are we going to use an online tool that already contains a decision? Three, we're just going to give it some data. So these are some of the questions that as a team we need to think about before implementing this IMO. 13. Training and Evaluating the Model: Now let's think a little bit about training and evaluating the model. It's very important for us to think about the training and validation and testing. Any time you think about training and validation, we have to put ourselves in a situation where everything makes sense. We start with cleaning the model. We take the data that we clean, let's say in this case sperm and sperm give it to a decision tree, which is the tool that we picked a decision, just an algorithm A. And then between the model and the model calls and reads the text. Okay, So this is an E mail, this is spam. Those are the patrons that I found in a Sam. This is non spam, this is Y. And the computer will just go over the data and find the patrons, find the patron. Understand, just like us, like human beings, understand how these things operate. What makes a spam e mail and what doesn't make a spam e mail. Okay. So we can split or separate between spam spam. So far so good. Validating the model. Once we're good, we have to make sure that the model operates, run it, see if it's working, give it a new e mail, see if it's going to operate perfectly nice. Then the third step is testing the model we have to take. If you remember, we take our data, we split it into 70% training, 30% testing. We take that 30% give it to the model and we see if the model is going to get an Or, an plus. So to see if the model is going to do a good job or not. So if the model is answering all the questions correctly, in this case identifying all the E mails as spam or non spam without any issues. We know that our model is operational and we have a high success rate. And we can easily deploy it. And we can start using it in our company if the model is not doing a good job. Wait a minute, let's get more data, let's work on it more, okay? So this is a really, really good process. So training the model, validating the model, and testing the model. These are three steps that are very, very important here. And this should not be ignored because I know that it's really good practice or best practice to train your model. First, give it the data validated, so run it, see if it's working. And then testing it by giving it that test data to see if it's operational. All right. 14. Model Deployment and Maintenance: Now let's talk about deployment and maintaining your model. So it's very, very important for us to understand if the model, if the I want we tested it and we trained it and we did all the work if it's going to be used locally, so in our computers inside the organization or the T or we need to use it, perhaps cloud provider like AWS Sure. Or any one of the cloud providers out there to deploy the model and start working with the model online. So we have website, let's say that we're accessing all together and we use that website to filter out all our e mails. If it's a tool, we can also provide it to other companies. So we have to think about these are just decisions and deploying is very easy. You just need to pick a tool and just host or take your tool and put it in the Cloud provider. And you're good to go very, very easy. So deploying the model simply, you can use Docker in this case, or Azure Machine Learning in any one of the tools. Monitoring the model is very, very important for us to always monitor the model, see if it's operational, see if it's working. What's missing? Are we getting good feedback from the people that are using the model? Are people happy with what they are doing with the model? Are they excited? Do they feel that the model is changing their life, the daily tasks and they're not getting any spam e mail anymore and maintaining the model. So anytime that we get new e mails, we can give it to the model. So we can always be up to date with the spam emails and non spam email. So anytime that we are getting spam email, that is tricky. We can easily give it to the AI model and the AR model will update. You know, upgrade, understand, oh, you know what this is. There is a new patron for spam email. I'm going to watch out and make sure next time to flag it and help the user. So very important that deployment and maintenance comes hand in hand with monitoring. So we deploy the model, Let's take it, put it online, monitor it, make sure the model is working. We're getting thumbs up, everybody is happy, and then maintaining. Anytime that we're getting new data, we can just give it to the model without any issues easily and quickly. Okay. 15. AI Spam Email Filter: Now let's just take a look at the diagram that I created for us to understand how the entire life cycle of the AI spam filter, or the AI tool that will filter all our sperm works tip number one, Collecting and preparing the data we can use. In this case, I just gave an example of technology Python. Python is a very powerful programming language that is mainly used for AI, data analysis, data processing. It can be also used for collecting data. Let's say you have e mails and you want to collect and put them all together in a small file or in your computer locally or in the cloud. You can use Python. Python contains a lot of great libraries that can do the work, that can collect all the data easily for us, and it can also prepare the data in Python. We have a lot of libraries such as Pandas Numpy. I can definitely talk about those in more advanced courses where we can collect the data and then clean it very fast, very easy. Once we collect and clean the data, the next step is choosing and implementing an AI model. In this case, we're going to be using a decision treat, because when making a decision spam, spam very easy. Thanks to Python, we have a library called SK Learn, or Psychic Learn, that can easily provide us with the tool for free without paying anything. This is really great. Python can also do the work where we can implement the model, or implement the tool if you will, to make it simply to make it simpler for us and we can easily use it without any issues. Once we're done, we can just train and evaluate the model. Once we implement the model using Python, we can also train and evaluate the model. Python can also do the test. If you can just take a look at the diagram correct in the data, choosing and implementing AR model, training and evaluating the model. All of these tasks are going to be done using Python. Python is a great tool because it can do correction, it can do the implementation of the AR model and it can also do the training and evaluating the model. Thanks. So it's amazing library in Python. A lot of libraries, a lot of tools if you will, can be implemented by us as AI developers without any issues deploying the model. This is the last step perhaps, or before last step if you will, where we take the AI model and we put it in the cloud. So in this case I picked Microsoft Azure as, as my cloud provider And I'm just going to de take my model. Once it's working, I tested it. It's operational, it's working without any issues. Take it and put it in the cloud and easily without any issues. Make sure that I'm using it. Make sure that it's operational. Make sure that it's working. Of course. Let's not forget that we need to maintain it and make sure that it's working 100% without any issues. Because we don't maintain it, we might have an issue of what's missing in our program. Are our users satisfied with the output? So these are questions that are very important in this case that we should keep in mind while using AI. The last step, monitoring and maintaining. So in this case I gave you the logo of Get Lab. So Get as you know, is a tool that is used to control the different changes in your code and the different changes in your AI or any application that you are developing. The Colso provide some tools that are very, very good for monitoring and maintaining your AI's. Say there are some changes in your code or in your AI tool, you can just applaud them or put them in your Cloud provider easily without any issues. Maintaining your tool, making sure that every time that you're getting a new data you're adding to the tool is very critical and important in this case. So that was it for our course. Thank you very much for taking the time and taking the course. I'm so happy to have you on, and I'm so happy for your time. I cannot wait to see you in a more advanced course where we talk about Python, AI, and machine learning, and how can we use them in a daily life. Thank you and have a beautiful, beautiful, beautiful day. See you soon.