Artificial Intelligence (AI) and Machine Learning for absolute beginners 2023

1. Course Introduction: Hi everyone. Welcome to this course on artificial intelligence for the absolute beginner. And congratulations on taking this step in investing in your skills. Now as far as skills go, you really cannot go wrong with artificial intelligence, which is one of the most exciting technologies of recent times. I mean, to the extent it's been called the Fourth Industrial Revolution. So I made this course, as I know, Ea in machine learning can be very intimidating topic to learn for the average person. I mean, it sounds very technical. Even if you're an IT professional, it sounds very technical. Maybe you're an IT manager or naked guy. I like a cybersecurity expert or maybe you're not in IT and all. Are you you're not an idea, you're just a business guy, but you've seen the transformative potential of artificial intelligence. And you want to understand the AA, but you don't know where to start, right? Well then this course is for you. We're going to demystify what EIA machine learning is. And the whole point of this course is to demystify EIA and make it accessible for everyone and remove the complexity that seems to be going at on this topic. So about me, just a quick overview about me guys. That good-looking person is me. My name is Dan moody style. I've been in IT for the past 100 years or so and on two decades. So I have multiple awards. My field, I was simply a what did the UK global style into visa. So I'm currently residing in London, working for me Buddhist publications, you know, like a CA Magazine isochore. And I've won many awards in the field. Just telling you this so that you know you're in good hands with discourse. Currently, I'm working on cloud security and artificial intelligence now. And what I really like to do is I like to take complex topics and the codon so people can understand it and apply it. So the whole point of this slide is to show you that I know what I'm talking about. And I, I do have some experience so that you're in good hands. So about discourse. So what's this course about? The whole point of this? If you have that feeling that maybe I've made a big mistake by taking this course. I hope you change your mind and make you understand. So I'm going to take away any lingering concerns you might still have about discourse. So this course is for whom you want to learn, AI, artificial intelligence, machine learning, but you don't know how to go about it, right? Everything seems to be too advanced to complex. You don't know programming, you don't know coding, you find it very boring and not interested in that, right? You don't have either answer, agrees and mathematics or like Python and all that other stuff that you see. But it seems to be at only a, so that the discourse is like for you if you really have those concerns, right? So what you're gonna do in this course, you're going to understand what artificial intelligence and machine learning is from scratch for either assume that he has zero-knowledge about this. You're going to create your own machine learning model because I'm not a big fan of death by PowerPoint. I don't get about that. So your project will be to create your own machine learning model from scratch without writing a line of code. That's my promise to you. You're going to deploy Ubuntu Services, ai based on artificial intelligence. And you just didn't, didn't connection and a willingness to learn. I guys, I know this sounds too good to be true, but we're going to see how it goes about it. So just a brief window, electric robot me. I do. I'm available on my YouTube channel and on Facebook, and on my blog. So if you want to reach out and connect with me, please do so. And that's that guy. So I thank you very much. I hope you enjoyed this course as much as I enjoyed making it and see you in the next lesson. Thank you. 2. Overview of AI: Hi everyone. Welcome to this section which is understanding artificial intelligence, a brief overview of the evolution. So the purpose of this section, guys, is to give you a context about EIA, about what EI is and what it isn't. And basically how we reached out to this point in time of AI development. Now, you may not believe it, but EIA has been around for several decades way all the way back to the 1950s. So keeping that in mind, why do you think only now we're seeing so much huge hyperbolic the concept of va, by everything. So I mentioned it's only so many AI different products. And so this is what has changed in recent times. Do make EIA suddenly sort relevant? And everywhere you're gonna see, you're seeing every day, you're seeing all sorts of job postings, your seams, all sorts of companies rushing to adopt AI and machine learning into this often products. So what is students who simply to make a sort of different nowadays? That's the whole point of this section. So first of all, let's get started. First of all, what is the AI guys? I mean, what do people think of money? As mentioned, if you stopped average guy on the street and you ask him game, and what do you think? What is artificial intelligence? What do you think that guy's going to think about? So let's take a look. So first of all, yeah, a lot of times the things that turns, we talk about atoms. We know that they can become into pop culture, you know, movies, books, but Helion, you do well what you see on TV. So do the average limb and artificial intelligence means the machines taking over. Great, like in the movie Terminator 2. I mean, if you remember, it's one of my favorite movie from my childhood, basically a machine cup reciprocal medical Skynet. It becomes self-aware rate and it takes that it tries to exterminate all humans. And it say it creates these machines called terminators to exterminate humanity. So that, that is like what he calls a lot of people think that's what AI is. Odd. If you want to go something more recent like the show called Westworld. What happens in there that humans create this realistic robots were basically slaves. And these robots slowly, slowly, they start waking up and realizing that their sleeves and they start fighting back on. If you're like me, you might have a multi-colored 2000 and a Space Odyssey. It's like a masterpiece by Stanley Kubrick. And in that there's a critical Hal that basically starts becoming malicious and it starts not obeying the orders of humans. That ONE, It's a very, very famous movie. So the reason I took these three movies, and all of these are from two different time periods. What is from the 90 is one and from recent time when it's from the 1970s was 16, I believe. So guys, what is the point? The point is to show you artificial intelligence has always fascinated people. It's always fascinated people have been fascinated by this from all time periods that the concept of a thinking machine that is something which is very, very privileged to our pop culture. So, but, and then you talk about EIA. Mostly people think as humans take the machines, taking over from humans, which is a scary talk, I agree. The concept of machines taking over slowly, the things that humans do. It's a scary thought, but that's not really a is. So let's go back to the beginning. Who coined this term AI? What is the one who coined this term? So the father of motion arcs are more than artificial intelligence is usually people refer to John McCarthy, who coined the term in 1956. And he referred to as, as the science and engineering of making intelligent machines. And that is that I wanted to focus on the last part, intelligent machines and what do you think an Indonesian machineries guys? I mean, if you go back several decades, people with a tada calculators and intelligent machine. And they were talking about something like magic you are just putting in. It's calculating the doing all this calculation by itself. We know not not that's not the case. Slate. I've calculated just as hard coded instructions of programs begin cited. It expects a specific input and it gives you a specific output. There's not much intelligence layer. So when we keep, keeping that in mind, now what is EIA? Eia is mixing the science of computers producing results, but delta being programmed to do so, you don't explicitly programming it. Program the computer system to do anything. Instead, what it does is it has those intelligent algorithms by which it slowly starts learning by itself. And then what happens is it can actually recognize speech. It can recognize faces, it can recognize pictures, it can make predictions, it can make intelligent decisions, and it can mimic human judgment to a high degree of accuracy. That is basically what we refer to when we say artificial intelligence. So I hope you're not able to understand the difference between your normal programs, which are which are basically bounded by their programs and that steer their lives, which had been made an artificial intelligence AI. What you do is you don't explicitly programmed to do anything. You'd give it a specific set of inputs. And then you let it learn by itself how it does it. You'll see later on. But I hope you understand now what the differences now has a very literal history going back like I told you that 77 decades. And I put it in the resources section. I don't want you to bore you by giving you specific by giving you specific dates and everything. So you can take a little bit by itself, it's a very, very interesting. It shows you the Fed milestones that have happened. So I hope you understand guys, what we're talking about and let's now move to the next section. 3. The impact of AI : Hi guys. Welcome to this section. So now that we've understood what AI is and what, how it came about, what was its basic history. Let's go into details on just how big of a deal EIA is. Likewise, it's so important because it's very important for you guys to understand EIS, not just a fad. It's not like a trend which is going to unite something which could be those very popular and just fades away. No. Ai is going to transform almost every aspect of a personal and professional lives. And I don't say that lightly. I'm not just being overly dramatic. I'm going to show you why. Why it's so important to fully grasp. Let me show you some contexts. So, so far in human history, we've had our own, I would say, three major industrial revolutions. And by industrial revolution, what we mean is leg, something that came about which whatever the huge change and the social level enough made businesses, jobs, how cities were, how people work, how the economy was driving. So let's look at the first one, the industrial revolution. Now in industrial action now believe it or not, there was a time when people used to primarily working farms. And your income was based on how much work you put in, how much manual liver you put in it was based on that and you don't know, you might take some help in animals with your labor, but that was pretty much how it was. Farmers were the main source of income and how people used to work. What happened ago of a few centuries back is that automation started coming in. Steam power, electrical power machines. They came in and I cannot understand. I cannot like overstretched to you the impact of what it was because now you had mass production happening in factories on assembly lines. It was something which never happened before in human history. So what happened? Jump started moving to the factory events. People started moving to the cities where there's jobs were. And what happened? The city started thriving rate that economists started Bu, started going up, jump, started moving. And this pretty much ushered in the modern era and people started moving to the cities. So I just combine the first, second Industrial Revolution because the first one was based on steam, water power. The other one is electrical, but the concept remains the same. So these were the first, second Industrial Revolution, which change pretty much how humans use to work. The second one was the digital revolution. Now this should be easy because you're living at right now guys. What was the digital revolution removed from mechanical to digital. What was this revolution? It was computers, digitalization, the Internet, smartphones. And they came and they changed businesses and people's lives forever. There's a reason right now you're not using analog phone. You know, All analog phones you see in museums or on the Internet, it was English allophones. And because of the digital revolution, that is the reason you're probably watching this on your home personal PC, or on your smartphone or on your tablet to high-speed Internet. That is the reason they did that is all the benefits of the digital revolution. So now I think you can understand what I'm doing with this yellow strip, the stage. Now RIAA, so important. So now we are coming. So this one, this is the fourth revolution which is right in the middle of it, it's starting. And EI is a big part of the Industrial Revolution was just happening. It is a continuation of the digital revolution. And basically it's building upon it in its financing it. So what has happened is now EI is there. And what seemed to be science fiction a few decades ago is now becoming reality. We are seeing robotic of us in everyday life like Siri Alexa, which is helping us make better decisions. It's improving the quality of a life is like, and it's going to change the lines hopefully for the better. But it has already started disrupting job markets. That started eliminating jobs which don't occur at a high level of human interaction. And new jobs have been created as businesses start looking at them or lazy or given this is where you can plug in AI and machine learning. So I hope you understand why it now AI is so important, why, how it's going to be and how it's going to revolutionize how people work and people will live. I've added a link in the resource section, guys, so you can get more information on the fourth revolution because, believe it or not, It's not just the idea other aspects to it by n, It's very fascinating. It's going to lead people, all the governments are interested in looking at how to manage it. What are the ethics of it? It's just awful. We can have a lecture or we can have a course just dedicated on that. So I hope you now understand why it is so important. So now we've understood that, I mean, you know the importance of a, but why has it become so carbon's suddenly, I mean, what has happened in recent years which we are seeing such a huge certainly Eva's products and services. We're seeing IT job postings companies were putting it governments being interested in because the concept of VAs, exceptions of fifties and the Senate, but nothing substantial happened. And why was that? Well, it was mainly because of three reasons, three deficiencies. There were lack of computing power, lack of data, and lack of talent. So if we look at this, like this, and these are the three things which have gone away now. And we have like basically become available to a large extent. And that is the reason why they've been such a huge surge in the abs service. And let's look at the first row, why it can print in color. Now, what has happened is the main thing is cloud computing. Now, technically it existed before the Cloud, right? Artificial intelligence was the essence of a piece. But recently the Cloud was really the catalyst for AI adoption to accelerate. Why? Well, because AI needs a tremendous amount of processing power. And what happened with the Cloud, you can spin up VMs with CPU, memory or disk, and you can basically put it in as powerful a VM as you want. So now you have this compute on-demand services, which I've only been with the Cloud providers. Additionally, the big problem with providers like AWS or GCP, they even have managed services and AI, which take a lot of the complexity for services. And they removed the barrier for entry also for newcomers. So it's very easy to not get involved in the assembly. We've just simply not possible before. And the second is data guys. Now did the cloud needs data and lots and lots of it. Why? Because the more you data you feed an AI model, the more accurately you can make predictions and decisions. And byte data, I want to, I want to be clear. I don't mean data, just records, but data means Weiss, images, video, geographical location, smartphone data and so on. And this is simply not possible before it was, it wasn't possible to store too much since much data, either because of the cost or the storage itself was limited. Now with big data and the reduced cost of storage that this talent has gone away. And lastly, but not least was the human element which was investment. Governments. We're not seeing that eternal value of our investing in AI. And that has changed dramatically. The elements have now seen the potential and there's investing heavily into the future. Ai. And businesses start-ups are looking how to change the Rocky Mountains to accommodate a in the future. We have University's educational institutions offering dedicated and master's level programs at EIA. The course of the course level at the discourse that you're handling right now. This is simply not something unheard off quite awhile back because nobody would have been interested in the aid was simply not got popular. But all of this has gone away. So that's why it's become so popular in recent times. And okay, so novel understood the impact of AI and why it's become so popular. So how are you using the agar is in every day if you can just think of a few services. So let's take a look Netflix, I think everybody has an AC flux, right guys. You're personalization. Personalization, which happens, you know, for movie recommendations, uses mobile eye you've, as you watch movies and Netflix understands what types of movies you like, it starts recommending what, that's all machine learning. It's looking at the movies they like and similar movies which other people are watching. And based on that, it's model, it's the a and machine learning based model. Recommends movies to you so that he can keep getting interested. I think everybody recognizes anybody who has an iPhone right-click necessarily. What the CV do is use a speech synthesis, speech recognition like National Natural Language Processing, it understands what you're seeing and to a high degree of accuracy, it's able to provide to you what you want and it's able to connect google. I think everybody should know this like Google search. As you're typing, you can see Google filling in the blank. It's already understanding what you'll tend to look for with a high degree of accuracy. It's able to predict that AI is building, built into the Google search algorithms. Similarly that Google Maps, you might have seen Google Maps, right? What does it do? It would be using Google Maps. It's labeled estimate where you had it and how to navigate without any command radioactivity, She coughs, typhoons and everything. Twitter. Today you might have seen the timeline prioritization. It has changed. Instead of just seeing the topmost tweet, it actually looks at what you've been doing, what your interests are, what are the relevant treaty might be interested in. And based on that, it prioritizes your tweets and shows it to you, right, which are the tuition might be most interested in. Similarly, flicking, flagging of her hip speech case. I mean, things which are not acceptable in a modern society. It actually be you can't have a guy just manually going through millions and millions of tweets late. So actually it's mortal, goes through it and the flags, which speech comprising of hate speech and flags it and less visible. I think Facebook everybody's using right? Facebook you through facial recognition, you call new tag your contacts. Through facial recognition is able to understand which are the pictures on what your context is showing up. And it's able to sift through millions and millions of pictures and show it to you. So these are just a few of them and every day implications which are using AI in. So I have to do it contextualizes and it shows how many eyes become and how you became sufficient coming. I mean, you're using it without even thinking about it. Okay guys, thanks. I hope you understood now for whatever wonderfully we're trying to accomplish here. So let's move on to the new section. Thank you. 4. Google AI demo: So guys, now that we've understood AI and its concepts and impact, Let's have a quick demo with Vision API, which is available from Google, and which gives you the ability to analyze images using EIA and gets all sorts of insight from it. It can tell you looking at images that can tell you what objects are there, what emotions are under people's faces, what textures present in their magic, cetera, is extremely powerful. I mean, you can even detect the images in class every day in the classified them into millions of categories, which are already predefined. But Google, some people are using it for detecting text in images. For OCR, it let q process documents even if you're like in a business. Say if you are an e-commerce business and you're accepting images, you can use it for externality to find out if any images have adult content while they're in cages. It's an extremely powerful tool. So let's take a look guys. So in case you're wondering what the, what the screeners and YMCA like, you're seeing two of me. It's simple. I put into wedges of myself, one with a handwritten note and when would have printed one. So let's see, let's see the Google vision and see how much, how much data Google can extract out of it with artificial intelligence. So please visit the page instead in the resources section, just click on it and I'll see you there. So guys, this is a screen. I was telling you whether this is Google Vision API is easily It's agent to cloud dot google.com slash vision. Like a toluene. It has a Vision API so that you can do it then you can just do not need to do any programming anything. You can just drag your image again and see what sort of data it can be extracted from it. So let's look at the hybrid image I put there. Let's open from their gaze on the capstone. So this is pretty interesting guys. Look. So it took my expression, sing possible and possibly I'm happy, am I, I guess I am smiling a little bit better. You can see it get connected, whether it's 98 percent confidence of saying, Nice, Okay, what are the objects? There's a person, definitely the yellow shirt and everything that liberals. So Hand handwriting sleeve just so you can see all of that it detected What's the text lookup. So it didn't designate a i is awesome. Okay, So that much it was able to detect what are the properties, not the colors and everything. That's amazing. Safe search detects whether Lego, thankfully, there's no What you call a document in this picture. Thank God for that. And so you can see how much it was able to detect just for the simple picture from the face, emotions, object, labels, text. Property is safe search. Okay? So let us do one thing. Let us look at the one where I'm holding the printed picture. Okay. So this is the second one. November, I have the printed document of my hand. I just wanted to see the U to show the difference. Okay. Again, I guess I'm happy in both pictures, which is nice. Okay. What are the objects now? Person, interestingly enough, the what he called the shirt and all that rate levels. Okay. Jersey sports uniform. Okay. Does it take my beard also? I like that. Oops, sorry. Text. Look at that. I would not detecting probably a is awesome. Learn how to use it because I guess it was more, my lighting was not legible, but I like that monomial fisher says London, if you can see here, it was even able to detect that. That's pretty amazing. You can see that here, right? Property is okay, the dominant colors and other It's addicting answer if such, again, thankfully, there's not a contact deadline, avoid and salvation image. You can see just how much image able to extract from the motions of objects. And then deliberately took just been printed and 100 handwritten. But I really like the fact that it was even able to detect distinguish. I was not ready for it. And it says London only the BUN is visible and was able to detect that also. So you think I just wanted to give you a quick and fast demo. I'll just hop awfully AS become and how much image is able to detect. I want you to do one thing. I want to take a few pictures and protect yourself. They're just see, just play around with it, see how much images. And the DEC, any bacteria doesn't store it. So you don't have to worry about privacy or anything, but just see how much information it can extract from any of your sample pictures. Okay. Thank you guys and see you in the next section. 5. Summary: Hi everyone. So finally we've reached the end of this section. And I really hope you enjoyed learning about what is its background and also how big of an impact it's having in our everyday lives. Now we learned about the reality of AIA enemy really distinguish it from the science fiction version, which is unfortunately present in some people's minds. And we also understood why EIA suddenly everywhere nowadays know what, what, why, why is that? What are the factors that have contributed to it? And lastly, with the quick and dirty demo of Google Vision API, which uses image recognition to like really amazing effect. And this EPA's available to anyone who wants to use it. I would definitely encourage all of you to experiment with it. See the different results that come up with objects, texts, expressions. Try plotting something with a 100 and uploading something which printed to really get a feel for how accessibility and he's become an idea, energy, and how intelligent it is. So now that you have a solid foundation, it's now time to do a deep dive into how AI works and what are the key concepts. And this is acquired guys before we start building out when the AI services. So this is definitely the upcoming module is definitely the most important one. So I will see you in the next section. Thank you. 6. AI Basic Concepts: Hi everyone. Welcome to this section in which you will discuss the details about how AI works and what machine learning is. And this is very important guys. We will discuss in detail the different types of machine learning models, how they work and Michael Butterworth, sweet situation. The importance of this model is that once you have this foundation in place, we can start building our own models. So that's why I easily, this is the most important section of this course. And I would definitely do need your full attention here, please. So simply put, if your basic scenario, if your basics are in place, then creating AI services become much easier. So let's get started. So first of all, I would like to clear up a few common terms which you heard a lot and which people use interchangeably. And that is AI and machine learning and deep learning. So while the I used together, they are definitely not the same. So we need to get this clarified, guys. Artificial intelligence we've discussed. It is a way to describe any system that can replicate tasks that previously required human intelligence. Almost always. I mean, this is a little bit some sort of complex decision-making. The human judgment would be required, you know, like most USCA system AAA would involve making predictions. Classifications are decisions with a high degree of certainty in a way that's similar to human judgement. So this is the entire field of making intelligent machines that do. The next step is machine learning. Now that is a subset of the eye. And like the name says, I mean, you could understand what it is from the name itself. It involves giving a competitive the ability to learn with our acid actually, but to make decision by itself. And this is definitely the area we will spend the majority of our time and dispose. Or why is that? That is simple. Nearly all of the AAC systems today I created using machine learning and AI can be created without machine learning. I'm not gonna lie to you, but right now, machine learning is the primary method for creating AI systems. Similarly, you can use machine learning for things other than the eight. But right now the majority of machine learning is a related. So we'll go into great detail in the chapter about machine learning and how it works. What are different types? Lastly is deep learning, or deep learning is a subset of machine learning. It's basically machine learning take into the next level that leads the best way to understand it. Deep learning models didn't. They can make their own predictions completely independent of human beings and they use neural networks. Well, what does that mean? Basically, it's inspired by how your brain works. The biological neural network, which isn't human brain. It analyzes data with a logical structure similar to how human beings. We know the drive to conclusions. And honestly guys, some very, very complexity of subject. I won't be going into too much detail. And instead we will be focusing the majority of it on the middle one, which is machine learning, as mentioned. So let's do a deep dive into machine learning, and I will see you in the next section. Thank you. 7. Lets understand Machine Learning: Hi guys, Welcome to this section, which is easily the most important section of this course, which explains in detail how machine learning works. Now, as I explained earlier, machine learning is a subset of artificial intelligence. And what does it do? It enables a program or software to learn from its experience and its improved, its improved self-care tasks without being explicitly programmed. Now it sounds really strange, but how do we do about it? So if you know anything about traditional programming or how computers work, you know how computers work. Oh, what a computer does is you feel you take some input and you write a program rate and you tell them I can put like, Okay, this is what it's going to come. This is what you have to do, and it uses that program to generate an output. That's how computers have always worked pretty much. Nobody comes to machine learning. It's slightly different. You actually give it the input and you tell it what the expected output is. And the computer itself is going to work out a program. You call it the color itself, a model. It's going to use that to generate what the output is. So if we look at it one by one, so you get lots and lots of data. But it's going to do and you give it him and got to come to understand that data. Good guys, this is what it is going to be. Yeah, the machine itself will build a model. We are going to use this to predict something which is happening, happened yet. You haven't sold it yet. Now I'll give it more detail. Not ever going to see if it's working or not. Your model is correct or not. If there was less, I'm not going to give it more data. You're going to have to return it multiple times. We find that it give it more use guesses until the desired output is formed. What is happening? The machine is basically running it on its own. And there was those to become more and more accurate over time as if he debater. So let's see. More diagrammatic representation of this circle, make it more clear. So machine-learning guys. So while you have, you have training data and yeah, so let's take a look at this. What is happening in the machine learning process starts with inputting training data into an algorithm. Okay, what does it then go to them? First of all, I'll go with them is just a way for the computer to understand what data you're fitting it and make sense of it. So what is going to happen? The machine is going to take this algorithm, take this data, and it's going to build a model. This model is what he called a. So basically what is the model? The model is the output of your algorithm and your data. So that's what basically it's called, it's going to use this to predict something that has not happened yet. So now we have FedEx data, we have Delgado from the sales created a model. Now let's see what's happening. If you give some real data. Let's take some actual data now and feed it into the model. So what happens then? Prediction is going to make a prediction and it's going to let see if it's correct or not. So what am I going to assess those are, so let's assume the prediction is not going to go back. If the prediction is not as expected, then the algorithm nutrient multiple times until the desired output as phones, what happened that this enables a machine-learning algorithm continue to learn exon and produce the most optimal answer. And it will increase in accuracy over time. The more data you feed it is going to increase in accuracy. So in a nutshell, the bail decision machine learning is, and if you remember what we discussed way back, we set d by d Theta is so important for AI and why the possibility to have so much data, that's why you're seeing so much AI services increasing. Well, this is one of them uses. A machine learning model is only as good as the data you feed it as because it learns from historical data fed into it and it built its prediction algorithms to predict the output for a new set of data will determine the accuracy of the models. If it depend on the quality and amount of input data, you give it a large amount of data, is going to build a better model and produce more accurately. So I hope now takes away some of the mystery of machine learning. So let's go to the next section and see the different types of machine learning that they are. Thank you guys. 8. Types of Machine Learning: Hi guys. Welcome to this section, which in which we are going to go into the different types of machine learning. So machine learning is a complex object in itself. And that's it isn't, It's been divided into two main areas, which is supervised learning and unsupervised learning. So each one has a specific purpose and action. You know, within machine learning, they produce different results and they utilize different types of data. I would say approximately 70 percent of machine learning is usually supervised learning. Unsupervised learning ranges from 10 to 20 percent. Main difference between the two usually comes in the data which is factored, admittedly, the label data or unlabeled. So what's the difference in the data labeled it? I, you know, it's pretty simple to understand the important outputs. Then you feed it into the model, solidly understood. So you tell the machine this is what the data is and this is what you, what output you're extracting. And when unlabeled data and limited as it doesn't have the output, doesn't have the input line in one of the parameters as missing. So the good thing about labeled IT IS you tell the computer already what you're expecting. But it'll cause more human effort because now you have to label all the data. And that can take quite a long time. And in unsupervised learning, you don't have to tell it what the data, so you can spit out on all the human effort. But the problem is of course, that it will cause more complex solutions. So let's do this. Let's go into detail in supervised learning. So I want you to understand supposing you're doing a task and you're doing something for the first time and you have a support as a standing over you and it's jetting whether you're doing, it's something that a cue on correctly. If it wasn't connected heat as you can during this long, do it properly. So this is basically what supervised learning is, that neither the explanation is there and then the image sell the machine, does it under supervision. Well, what does that mean? You give it a label data when you're turning it. And what happens is basically the detail you're feeding it. It already comes back with the answer. The alcohol Dalton should come up with. So a label database. And let's take an example. You, you're feeding in different types of flowers, you know, maybe roses, daisies, daffodils. So whenever you're giving it the data here, what did the flower is? It's already there in the data. So what happens is benefited a new set of data and you image the model is going to compare it with the examples you already gave it. And to put to predict what the new image. This is basically supervised learning, unsupervised learning. It's the exact opposite. The reality is cleaned perfectly level data is not that easy to come by. Or if there's a lot of human effort which goes into it. So, and sometimes that's what the research shows that asking the questions, I don't know the answer themselves. So this is the unsupervised learning comes in guys. In unsupervised learning, the machine learning model is given a set of data without clear instructions on what to do with it. So it can be a number of examples of data without the outcome or the correct answer. So what's going to happen? The model is going to pass this data itself and it's kinda find the patterns. It's going to find structures in the data by extracting features. So I'll show you in detail lately. Let's have a graphical flow like we did before to get up early ideas. But before we go, I'd guys, like I told you, like, 70 percent is usually much Sheila but supervised learning and then I would say 10 to 20 percent is. And suppose that there was another one also which is called reinforcement learning, which is not used that much often. But I do want to discuss with you in case you had about it, reinforced. Reinforcement learning is more about trial and error. So basically it's a way in like you might play a video game, you know, first time you play with the level, you don't know what to do, right? You're fumbling around, you're doing this, doing that trial and error, you make mistakes and you understand those. This is basically what reinforcement learning is. It gives them doing stuff, making mistakes, and understanding what's happening. And it finds itself. This technique is usually used for training robots. You know, it makes a series of decision tasks like Venice, doing an autonomous vehicles like they're driving by itself, or managing inventory. So this is basically where reinforcement learning comes in, but we won't go into too much detail because it focused on supervised and unsupervised. So now that we understand these two main models which are used, Let's see a graphical representation of them to get a better idea. So firstly is supervised machine learning. So in supervised learning, like I told you, we use known or labeled data. And since the data is known, the learning of the phosphorylase like to say, you know what output is, I'll give you a simple example. So supposing you have a child with you, right? You show them a picture of a dog and he said this is a dog and is shown a picture of a cat and you would find her, Hey, it's a cat. Now you show him enough pictures. The venue showman, new picture. Now he's gonna know it's going to recognize that he's going to learn to differentiate between them because what has happened, you have to meet him and he's able to recognize different breeds of dogs. It because even if you haven't seen that, because he knows what their basic features of a dog is. Okay. I hope you understand. So let's see. So now you have a number of pictures. You have this data rate. You've put a label of a dog. And if we didn't have the machine learning model and the algorithm change, what it does it now you have a supervised learning model. So now you have this data. Let's feed it some unknown data. So you're going to fit a picture of a new dog is something which haven't been living before. What goes it's going to pick up action. Yes, it's a doctor, so I understand it's pretty simple. That's what's supposed learning is guys, you train the model enough. It understands mobile world, what it does is going to come into it. So, and now let's look at unsupervised learning. Now, like we said earlier, it unsupervised learning. The data is not labeled, it's not known. You give the model absolutely more healthy, but the data is going to come no supervision at all. So you feed some data to it, like as an example, cats and dogs and pictures, but you don't tell it, you don't tell the model what it is. So this data is going to be fed into the machine learning algorithm that's going to be used to clean the model. So then what do you think the model is gonna do? Is gonna search for patterns. It's gonna see OK. Amanda, cats and dogs look different. They have different activities going to pass the data and it's gonna classify it. These two animals look similar to each other, but it's going to look at the differences between those. So the algorithms, machine learning by itself and it discovers a pattern on a structure. And since it's working with unlabeled data, it has to find out by itself what are the common features and separate them. So yeah, it's going to separate the cart separately and it's going to separate the dog separately. So you understand what the advantage of unsupervised machine learning is. It is, it has the ability to go to the unlabeled data. So humanly, for it is not acquired and labeling all this, making all the data before readable allows for much larger ecosystem about the bill because he can skip all the human labor. But what the thing is, it becomes much more complex. So I hope you understood now harvest supervised and unsupervised machine learning books. And so I'm going to move on to the next section in this demo is one of these algorithms because sometimes people do ask, well how these algorithms work. So I need scene the next section, guys, thank you. I hope know you've understood all these different models work. 9. KNN Algorithm: Hi guys. Welcome to this completely optional section. It's up to you. You can skip it if you wanted. So this is basically about, because a lot of people do ask me sometimes about these machine learning algorithms. You know, how do they work? Like what can mechanics behind some of these algorithms? If you want to know the details of this, I've chosen one simple machine learning algorithm. It is one of the easiest and simplest to implement, which is a machine learning algorithm, which is called the Kim nearest neighbor key. And an algorithm for short. It's a supervised machine learning algorithm. And what it does, it, it's based on the simple principle that similar things exist in close proximity. I mean, you William neighbors are, you're basically like that. That leaves a is B, It's in the name also the nearest neighbor. Or it's very uncomplicated and easy to understand. So what happens is it groups the existing data and what it does it when you give it new data, it puts the data into the same category, whether similar things are. And based on that, it makes a prediction guilty guys distinguish probability that. So let's take an example on what's this diagram is. Supposing you have an image of a creature elected a cat or dog, right? So what happens is you've fed the model, the data about cats and dogs based on the height and weight. So it has group that data usually decide on the street and bulbs or this ICT industry. So as you might want to put a new picture there of a cat and a dog. And I won't build the model, whether it's a cat or a dog. I only tell him it's it's heightened width. So what's going to happen? You're going to group it and you're gonna check what its nearest neighbors are, ionic either cats or dogs. And based on that, it's going to make a prediction. So let's take a look. So I feed it a new picture I have. I don't tell it what it is, right? What what he's gonna do is gonna check height and weight. Let us can fall into Italy had a catch, its nearest neighbors like that. So it's pretty much the same. It back to 6 o'clock. Either one I created a picture of a dog without telling it what's going to happen. It's going to check the height and weight, okay. It's for somebody, it's falling into the nearest neighbors, which are dogs. So based on that is going to say, okay, this is definitely a dog. Yeah. So you can understand the advantage of this mortar strategy to understand people do Python scripting of this, you can easily find models of this on the Internet. The disadvantage of this as the, because it needs a lot of data you can understand to make predictions properly because it's going to have to categorize it's similar model. Accuracy will increase based on the number of data it has. The disadvantage of visitors becomes slower as the volume of data increases. You know, because if you have what he called an environmentally, you need to make predictions quickly. Knn might not be the most optimal model gotten for you. But supposing you have some sufficient computing resources, you know, your, your computers are so powerful it can actually handle all the data you're going to use, then it can be a very good model to use. So I hope you understood now some of how much inelegant bottom of the electorally huge amount of algorithms. I won't go through all of them. I just wanted to show you that this is the back-end of how usually algorithms work. Okay guys, Thank you. As you in the next section. 10. Summary: Hi guys, saw we reached the end of this very important section about machine learning and the key concepts behind AI. I hope you understood the core concepts of AI machine learning. We took, we took a deep dive into machine learning and we learned different types of it. The different types of models that sub-arrays and unsupervised. And we also took a look at one of the algorithms, which is called the k-nearest neighbor. So like I said, this was the most important section of the direct costs. So the good thing about this is now theory is over. The theoretical part of this course is over now you have enough information, enough knowledge to start creating your own machine learning projects, which you're not about to do. So this is really exciting guys. I'm gonna see you in the next section. Thank you. 11. Time to build a Machine Learning Model: Okay guys. So welcome. I'm happy to tell you that near finished the conceptual part 1, 2, 3. Hi guys. Welcome to this section. And I'm happy to tell you that we have finished the conceptual part. Of course. And now we're ready to get cracking on building actual artificial intelligence based services. So since machine learning is passion of mine, so I thought it'd be good idea, Let's build a supervised machine learning model for ourselves. So what are you going to do? We're going to train a machine to recognize liberal images and then present it with fresh data to check whether the model we made was successful or not exactly like we discussed earlier. So all the theory that we discussed a god, you're going to put it into action. So the good news is, we're not going to give you don't need to do any programming as they're under the Tools present, which can manage the entire bag on complexity for us. It is something we could not have imagined possible a few years ago. So in order to do it, Let's look at one of my favorite tools which is teachable machine by Google. Now teachable machine, it's, what is it exactly? It's lit. It's a tool that makes creating machine learning models very, very fascinating, easy anybody can do it. There is no technical barrier to overcome. It was launched in 2017. And what it does, it makes machine learning accessible to everyone. It's fast, it's easy and it's very easy to do. You can create, you know, machine learning models for your application insights and expertise. According as a quiet, you could even explored the models you create for your projects and just donated for me to study. So let me just show you let me go there. The link for this is in the resources section, guys. Hi guys. So this is teachable machine. But when I was telling you about, so this is where you can create models you can create. Now using this, you can create models that can, let's say you can identify images using a web camera files. You can teach them what watertight and if i sound samples, even my favorite is that you can even teach it to classify body motion. Suppose a switch you strike in front of your webcam. So just to show you, yeah, you can see. So this is for images, sounds, you can do something even pauses. So all of this you can do. So let's start with the first trend laymen. But every show you receive, if you see it in action, we understand it much better. Let's get started, guys. And let's do an image product. Okay? So what are we going to do? We're going to basically what he said, This is how this is the most complex machine learning, sorry, teachable machine gets these other Lanisha classmen class to these other type of data, we're going to refuting it. So let's do one thing. Let's create a machine learning model to identify different types of fruits. So they're going to train it for some live data and then see how it does, how well it can identify different types of images which you present to it and what he calls. So this is all you need to do it. So let's do one thing. Let's get started, guys. Let's create like first-class, It's called Apple. Okay, second class, Let's make it. Let's add one more guess. That would be I think got it. Okay. Okay, so now that we've created these few different classes, what we need to do is we need to give it some data. So heavy, we need to feed it some images so it can actually understand. So when we're presented with live data, so let's do one thing. Let me turn on their webcam and presented for some images. Okay guys. So this is me. I'm going to hold up an apple and 2008 as much as possible. So you see this, this is actually gathering all the data. It seemed Apple presented for vol, as much as possible. I'm going to give it as much data as possible so it recognizes. So you have an uncle 25 and Wendy's, I think that should be enough. Okay. That's it for Napa. Okay. Same thing now for the banana guys. Okay, so let's hold it. This is just to give it as much data as possible. Okay? It's building up all of this image library. Okay? I think that pretty much the same way as we did for that. Okay. One more. Okay. Let's record it. Let it keep all the data alert, try to rotate it as much as possible. So it does as much as data to build what rate? Ok. That's pretty much I think as much with yeah. Okay. So now we've fretted data like we did. We learn a supervised machine learning, we've fretted actual data now we need to train the model. So let's click on this. Yeah. So now what is it doing? It's going to claim start building its machines, supervised machine learning model, mishandled golfer. And based on the data we've fretted, do not guys, very important thing to Nazgul away from this page because then it stops. So it isn't running it by there. Yeah, exactly. This is what I was referring to. So now what's it doing? It's turning on its supervised machine learning algorithm. And it's cleaning the model for recognizing live data when it happens. So it's almost done now. Okay, Well, so it's search ready now, do we accepting data. So you don't worry about that. So you see this, they did it shrink the confidence level. So the audience don't worry because it's so let me put up my hand up. Are you see the snow? It's not understanding what the hand is. Just think it's an apple. That's pretty funny because we haven't funded died, but he did and let's predict some actual data and see what happens, okay? Okay, honestly just okay. And let's see an apple with extensive knowledge. You see this a 100 percent. It's saying this is an app and you see it's not confusing it. Okay? And we would know it what happens? I think it'll save you see if I'm moving my hand thinking that happened behind, is that different? Yeah, I think keeps up a less funny. So okay. You see the banana and it presented things matter. So you see you just recognizing there's still some kinks to Sorter tool. You can see it's looking at the backend and for some reason ticking it's an audience because some of the data captured, we haven't defined it fully. But you see how these ideas now, if I put an apple moment saying it's an apple, from putting into banana a 100 percent of signature bonus. I see the husband upon the data. So that was pretty exciting case. So now you have created a model. What can we do with it, guys? So this is just to show you, if we go to the export model, you can actually now export this model. If you click on upload my model, what happens is it actually gives you a live link. If yes, he does this link, you can use it'll WHO said this link for free? Or if you wanted to like download it, you can actually download a zip file with all the coding done. So you see how much heavy lifting Google has done for you behind the scene guys. So this is, I just wanted to show you how easy it is now to create machine learning models. And you saw an actual application of the purity we learned in the previous course. Of course, there's a lot of polishing we can do with this model. My goal is just to show you how easy it is. So I have an assignment for you guys. I want you to go and do an audio or a post-process and play with this amazing service for yourself. You won't believe people have built some amazing tools you think teachable machine. And I really want you to experience it for yourself instead of just watching me do it, do it for yourself. So once you've done with it and you've played around with the nervously in the next section. Thank you. 12. AI Services in AWS: Hi guys. Welcome to this section. Now I hope you enjoyed the last lesson with really created a very first machine learning model. So now we are ready to try some news AI services, but this time not with Google, but that Amazon. Now Amazon offers some of the most amazing machine learning services or own absolutely free of cost, like Google, they want to make machine learning accessible to everyone. So before we get started, one thing I need you to do is you'd cue you to create a free AWS treaty at account. If you don't, if you don't already have one, it's very easy to do. Just go to the Resources section and I put the link there, god there. And please fill out the form and you'd have it if you haven't done it, please go there and do that first. The what is, it basically gives customers the ability to explore and try that a lot of AWS services completely free of charge up to specified limits for each services. Now, he does on this limited services are provided on those basically against each service on the page. If your application you use exceeds the 3D elements you will play as simply play a standard pay-as-you-go service charges, which are present for each of us has but the good news is we won't be exceeding them. We won't be using those services that much. So if you haven't done it already, please go ahead and create your 50 or service. On this page you can see this is the whatever the footing to Amazon.com slash few such machine learning this linkage, there are no resources section. These are the three Machine Learning Services which you can do on AWS on the free tier. So it's amusing, you have text-to-speech, street to text or machine learning translation you're going to do, you're taking three of the most prominent services and using them. So you can see we have Amazon Polly, which basically converts text to speech. Okay? Whatever takes your coverage can actually make it live a life like that. Okay? Like what you call, if you have like some peas or blog, you can actually point amazon polygon is going to convert into like a completely make it audio. And Amazon Transcribe, which is opposite for just speech-to-text. Okay? You can understand this idea actually take speech and mixed texture. You have a video, it's going, it can completely transcribe and write it down. Seems McLean is actually machine learning models, but this is completely managed service, so you don't have to do any of the complexity down little complicated. Backend. Amazon Lex is a chatbox, you know. So you can actually create your own chart box, which you see on websites which stopped being in natural language and understand what you want. Automated image and video analysis comprehensive lots of them in there. You're going to be picking three of these most prominent services, which is Amazon Polly, Text-to-Speech. Amazon Transcribe just speech-to-text. And Amazon Lex, conversational AI for chatbox. And we're going to try and make three of these services. So this is pretty exciting. Let's go ahead into the next section. Started building it. Thank you. 13. AWS Transcribe: Hi guys. Okay. So the first thing, the first service, I want you guys to take a look at this Amazon AWS transcribe. So what does AWS transcript in one sentence, it is speech to text. What it does is basic uses, deep learning, which if you recall from earlier, it's like a subset of machine learning. What it does it convert speech to text, their process called automatic speech recognition. So what it does, it opens up a lot of exciting possibilities. If you are a developer, I mean, you can have some very powerful functionality to the applications like video subtitles. If you have an e-learning application, you can add subtitles to it. All. You can transcribe constant or worse recording so you can search them like you such any text. Or you can automate taking minutes of meetings. You know, I mean, the possibilities are quite amazing. So AWS transcribe, it's driven by the AWS Machine Learning platform. So what does that mean? It actually gets smarter over time. As it learns in learning, it gets smarter over time. So let's do this in action actually. So I've, you've created your AWS account by now as I want you guys to do it yourself instead of just watching me doing stuff. That the best way to do something is to do it yourself. So this is a small we do it given days I've taken this, you will find this as an MP3 file in the resources section also. So I'm going to conscript disomy. So let's listen to, on 1520 seconds of it started with a question. So teams seem to make a composite object in the universe. Time synchronous. Hi, it's not changing. Artificial intelligence. Most about. Okay, so that was quite interesting days. I mean, this is non-formal small midi clip from BBC on artificial intelligence. So like a dual do, this is good in the resources section also, this is quite interesting. You can listen to the whole MP3 clip if you want to know, I want you to have, I want to convert this video into text as English skills. So let's do this using AWS transcribe back IS, so this is the idea with console. So let's go to transcribe glucose transport. Let's open this case. Okay, So there are multiple ways of doing that. I mean, if you want, There's something called real-time transcription. So we can actually, as you're speaking into the microphone, it can transcribe it, but I don't want to do it through that. Let's do it using a transcription job. So we wanted to transcribe that MP3 file we just listened to. So in order to do that first, you need to put it into an S3 bucket. I mean, if you don't know what S3 is, it's like a folder in AWS. It's like Dropbox or OneDrive, a few slides before you have to create that folder and upload at Embry the file there. So I've already done it. But just to show you guys, she would not know, go to S3. So I already have a bucket called S3 demo, Udemy, demo AI, and I've put day, you can see them but the file already there. So if all you have to do is go here, create a bucket. You have a name for your bucket. It has to be unique guys, just to let you know and you can pretty much keep the, all the default vertical options that are there. And that's uploaded empathy file there. So I guess if have an S3 bucket and we've put the file there. So let's go back to it at least transcribed. Okay, so let's get a job in order to transcribe it. So let's call it a blue as well. We can, we can use all the default options, guys, this is focus, at least specific language, English. So this is, yeah, this is very exhausting where the file is. So let's browse history. Yeah, let's choose this one. Yeah, this is the file we want to transcribe choose. Okay. So, yeah, it's cool. Next, there's some other options you're going to contract from a world. Basically, we don't need to get into this. It can basically keep all the default options. Okay, so now you can see our transcript demos running so that that clip is around like a few minutes lighting. So it can take around 15 to 20 thirty-seconds effort to transcribe. So yeah. What are you going to do is give us the what he called transcribed job has been let's see it in action and see how accurate it was. I mean, did it did it transcribed the first 30 seconds which feel listened to and live with any mistakes I would like a little, let's see how much how accurate it was. Let's refresh it and see stock. Okay. It's in progress, guys, sometimes it takes a few minutes, lapse rate. Okay, guys, Let's completed. Okay, so let's click on them and see what happened. Let's go down. Here. Does okay. Talker. So this is actually, this is where the transcription happened, guys. So let's see. Let's take a look at the knobby have to operate in front of us. And if you remember what the videos, so what does AIA? It had a sentence with it put a space with me. Hey, let's start with a question right now you can see, so it's surprisingly accurate right down to the pauses into sentences. I would recommend, I would recommend reading this and then listen to them side-by-side. And then comparing the two to get a feel for it, how accurate it was. So in available, of course, you would schedule the jobs to run business specific events or time periods or you wouldn't make EPS from other services. I would recommend you guys play with it with some other files yet, understand how it works and let me know. So I hope you've got a good feel for how it works and how accurate it is. Like a do loop. It opens up a lot of possibilities. I mean, you can transcript call centers. You know, people want to search for their call center recordings or they want to automate ticking minutes of meetings, all those you can do. It's a very powerful density. You can see just how easy it is. So I hope this was useful to you guys. I'll see you in the next demo. Thank you. 14. AWS Lex: Hi guys, welcome to our last demo of Amazon Machine Learning Services. So we're going to be doing a demo of amazon Lex, this timer, which is a bit more complex than the last two. So, so far we've done text to speech and speech to text. So let's make something more dynamic this time, which makes a lot of this stuff together. I'm, I'm absolutely a 100 percent sure you must have interacted with chatbots at one time. And under when you visit a website, you know, what is a chat bot is basically a machine learning program that stimulates a conversation you might have on the phone, over the counter with the person, but it's not a person you are interacting with virtual machines. So amazon Lex, it uses machine learning which recognizes speech or text, and it can take actions. It can fulfill orders based on what the customer has spoken or written. It's the same technology which follows Amazon elixir also. And you can build extremely powerful chatbots for you of obstetrics minimum technical skills. So let's actually go over to the Amazon console and start working on it. And I will see you there. Okay guys. So now the hair on the console, Let's go to amazon, Lex. Okay, here we go. Okay, so this is not going to regenerate it. So sometimes the services that you're looking at, it's not in the regions. So I'm currently in US Eastern Ohio and let's go to London. Okay. So this is Amazon Lex. Let's get started now. Okay. So you might see a few things here that you might seem confusing initially, buddy, I'll explain. So zooming want to peel, let's assume we want to create a chalkboard to other burgers, right? So what does an indent intend is basically what he wanted to do. I want to order bogus. I want to book a flight or book a trip. Okay. That's what the intent is. Attendance is what the user says. Types, like our notebook, I want to order a burger, I want to book a hotel, I want to book a flight. Slots are basically the parameters which you give, are going to give a video live. What type of burger, what time? Those sort of things. Fulfillment happens at the end. This is what after everything has been done, it places the other. Okay? So this is basically it's just different names for stuff, but it's a pretty much the same thing. So, you know, Amazon has already predefined a lot of those sample box. So you don't have to actually do anything. You can just do something predefined, but I want to do it ourselves. So let's do it as a lev, simple creators several bought which cannot, we're going to take order for burgers further student or the burgers, that's an MFA, but let's take English, UK. Okay, She's output ways. We can continue Amy. Hello, my name is Amy. Okay. That's the session timeout after Richard lift their user doesn't vertical. No responses here, it'll reset it. So these are all the default options, copays basically, if you're taking some sensitive information about minors and all that. So no, it's not. Confidence threshold is getting raised basically indented. How much intelligence is easy to discern what the user's same with Augustinian legacy. It's basically how much confidence in detected they use it as a rule. Is the user saying, is, is it close to what we think? He's saying? Basically the intelligence which it uses. Okay, so what is left now? Let's unfold. Okay, what is left now? I think we put everything here. Session timeout. Okay. Five minutes. Yeah, sorry, I forgot to put the session timer. Okay, let's go ahead and create it now. Okay, so like I said, first thing is the creating the intended rate. We've created this border or what do you want to do? Like I said, we want to order burgers. Your eyelid expanded scale. So let's do it. Create content. Yeah, I want to create an intent. So the document the same name of the bottle order. Okay? And okay. So okay, here you wonder what the user but room. What do you think the user will type a rookie in order to initiate the board? So whatever you're detecting what they do, some nitrate or murder. I want yeah, you can keep adding it as much as you want. Miss OK. Ok. Ok. And now a dislodge side, what did the milliwatts, the information you are going to be ticking from the user actually. So let's, let's think about engagement. What do you want? If the guy says, I want a burger first thing, what are you going to ask him? You going to ask you what type of program? That's what I'm going to ask him about that for the bot is going to ask. So this is basically what the datatype is. So we're going to just click alphanumeric. It has a lot of predefined Amazon. You can see email, phone numbers, time, date, it's pretty it's very good at Mr. Gardner. Okay. What type of burger? Sorry. Sorry, just to be burger. Burger. This is a, this is actually the name of the slot, not prompt. This is the prompt. Okay. So okay, after you've taken the type of bugger, we can ask yeah, the location rate. We don't know where the guy is. So let's take the city. That's what he's going to ask. Okay. 99. One more thing. Okay. You've had the direct runoff or delivered every day. So I think Amazon already has an API data or deviation rate. Yeah, dying. I'm sorry. What time? Delivery? Yeah. I think this is a leg just for a simple audit. I think this is more than enough not to get too technical about it. Now all of these adequate later, I see only one is checked now we want all of them. Yeah, the first one will be what type of vulgar? The second one would be the priority. If you look at the proteins is the sequence in which it last two questions. First, priority is a type of bogus seconds, the location. Third is that time. Okay? Confirmation prompt. This is basically yeah, you sure you want to order. So basically they are you sure you want to restore order? Yeah. Council, if the user says no. Okay. Now for filament is basically once you've taken everything like I told you, placing the order, now we're not going to be looking at your business logic. Obviously. We're just going to save it and up in biomarkers back to the user. But let's not just ridden some genetic message left-sided here. Okay, So when he goes Let Thank you. Your order for let's see, then you've broken buckets numbers and basically what you're taking from the slot here, you're under populated dynamically. Slot. For the first slot has been list, deliver location at the literary down. Okay. So I think this is more than enough. Gaze on just clicking. Let's see. Let's hope I haven't goofed up some method. Let's build his bottom. See what happens. Yeah, let's build a bot. Let's see. Oops. No value for reject statements curve and reject statement.me, invalid. Okay. Let's see what happened goes. Okay. Your order has been answered. I think this is I think I forgot to put the radical. If these accounts as daughter. There was no other statement there. Let's build it again. Okay. It's building up for some things like that, what is happening? It's like if you've built any program or a computer program is basically compiling the bottom, okay, everything was put in there. It's basically building it in the back-end and creating the bot ready. Okay. Is there any no. Okay, So now you can actually add the right screen. You can see this is what he called the bot is now ready for us to interact with just completing the bills that it led to build complete. And then you can actually Desktop conversations with it. Let's see. So we should be able to place an order for a burger and get it reflected back to the customer if you've configured properly. Okay, so it's now ready. Guys, Let's see it. Okay. Let's see. Let's put I want okay. What type of burger? You can see now you can see that there are slots below. It's checking what I want to say. Okay. Where do you live? I live in the body. What time you want to deliver than bm? Yes. This is the conformation thing we put in, yes. Yes. Okay. For this happened but they didn't show the confirmation messages. You put an interestingly enough Let's see what happened. Response. I think I think I didn't save this, unfortunately. Ok, My mistake guys. I forgot to say that because this is a very generic message you want. There's no Okay. Thank you. Your Honor. For let's just put a type list. Okay. Okay. Yeah. Because we want this message to each on local distributor. Thank you for order for type has been blessed us with dielectric. I think it's case sensitive, also listed each int and set the preview. So now the response to the customer's funds to the goblet, but to be shown not just genetic one. Let us work out fine. I don't want this. I want it to be like customized for the user because it's not going to use a fungicide. But anytime you make a change case, you have to build it again just to let you know. Let's see if I'm benefiting at insipid luster module. Okay. Let's begin again. Okay, so it's the limit again, basically detect by those fun little bit of time because it's compiling the bark from scratch again and basically check so all the bottom it doesn't do anything. Just to the left. Okay. Love in London. And BM? Yes. Okay. Good. Now you can see the customer sponsor gave not the genetic language we didn't want. So this isn't guys, you've created your own bot and it's very, very powerful. You can, you can customize it as much as you want. Of course, the willpower webcam will convince you integrate it with some business logic choice should pick disorder and basically take all this parameters and place a liquid to a restaurant system and therefore the order comes directly. But now you've created your own bot. It must have been collected with it quite a bit today. And if you see this, you can put it on Facebook. You can put it on Slack. These two haven't used, but it's really only can monitor this box also how many requests have come in and there's an issue. So this is the powerful over that SHE using all the natural language processing to basically interact with the users. And if you put sometimes luckily you can really check, I want you to check how the bot intellectually, if you just put something that like confusing, if I just put something I haven't put in. Yeah. It can you see that? So saying sorry, can you please repeat that, less clear this again. Let's put something I'm not actually going to check the context of what am I doing. Okay? Yeah, you see, is actually in, I didn't put one of these utterances, I put something else, but it Understood. So this is very powerful. I wanted to really glazed, want you guys to take a look at it, understand how it works and create your own box. That's of any, this is a much more complex than what we used before. So I hope you enjoyed and gaze and OPIA, it takes away some of the confusion which happened before with the gods toolbox. For now, I want you to create your own custom board. Let me know. Thank you. 15. AWS Polly: Hi guys, Welcome to the second demo of AWS Machine Learning. So this family we're going to look in at another service which is called Amazon Polly, which is just the exact opposite of AWS transcribe. So instead of speech-to-text, we will be doing text into lifelike speech. So what this polytope polices machine learning could synthesize speech sounds very natural and you can make applications that basically talk when using syntax of poly that, but the punctuation in a text you might be like if you have what he called commas or foodstuffs that's actually used for cues for parsers and books Use Cases laminar for this is too many to mention. I mean, if you, if you made a website or a blog, you can basically have polyploid odd you did lead to your visitors. Or if you have an e-learning up, you can convert divergent course material into a completely order driven costs. Or maybe you're a bank rate. You can have polyploidy and interactive voice response. So when the customer requested upon balanced polygon received that he doesn't automatically played back to your customer, but human interaction. So I hope you get the idea. So let's create some text for this and check it out how it works. So this is just something I wrote guys and I mean, you can pretty much it. I just thought this right now on the fly. I just want to put this in Amazon, Polly, and check it out. Flexicurity, extremely straightforward. I will see you in the AWS console. Okay guys, so heavier. So unlike care transcribed, we don't need to have an S3 bucket or something that should be pretty straightforward with volume because my intent is just to show you how it works. So let's go to Amazon. Polly, Here we go. Now, okay, so easily, like I said, it's very, very straightforward here. You can just put things here and what he call it replayed back services. My name is John Allen Wilson textbook. Hi. My name is Joanna. I will read any text you type here. Okay, that was pretty nice. And you can see you can actually change this from male to female and all that. Hi, I'm Matthew. I will read any text you type here. Okay. So let's put architecture. Hi everyone. This is a demo of Amazon Polly for your Udemy course. Polya's converting this text-to-speech based on the parameters you have set to show you how Amazon AI services work. You can use this to create interactive voice response for call centers and other applications based on your use case. Or you can use Amazon Polly to create audio files from blogs or websites. Hope this was useful. Bye. Okay, let's take it through female human waste. Hi everyone. This is a gentle of Amazon Polly, for your Udemy course, Polya's converting this text to speech based on the parameters you have set to show you how Amazon AI services work. You can use this to create interactive voice response for call centers and other applications based on your use case. Or you can use Amazon Polly to create audio files from blogs or websites. Hope this was useful. Bye. Okay, you see suddenly you can check how accurate are two draws rate. And even though you can put lexicon, what I lexicons maybe in some pronunciation, some birds I pronounced differently. Some words are there which are not pronounced the way you want to have slang and vocabulary, you can actually upload those here and it'll read it and it'll change those virtual how you want it to be pronounced. So it's pretty straightforward, guys. It's like it's much more simpler than transcript is straightforward. I want you to play with it, check it out, and see how it works. Okay. Thank you. 16. Summary: Hi guys. So finally we reach the end of this section, and I hope you enjoyed it. We find the applied what we have learned in theory into practice land implemented those concepts also. So we created a machine learning modelling Google teachable machine, which between to differentiate between different types of objects, images. And I give you some homework to do the, use some audio, use simple as this and see, are you able to get a good machine learning model? A model. And lastly, we used a Amazon 50 or services for machine learning for transcribing audio data, speech recognition, and creating conversational chatbots, which use natural language processing to take orders, are going to use all those free services. Do you actually implement some basic EI services? I hope this has helped to solidify the concepts that you've learned earlier and now you have a better understanding of the A1 before. Now let's move on to the last section which you've app everything up. Thanks. 17. Why is governance necessary ? : Hi guys. Welcome to the section which is an extremely important part of the spores, which is pertaining to artificial intelligence, governance and standards, and more specifically how AI can be misused. Now this might seem to be like a little bit of a confusing thing because how can he be misused? You might be asking, well, the sad fact is that the eye is AI is like any other technology in this regard. There will always be people who will try to explain technology and music for unethical activities. And let's take the example of say, the Internet. Now the internet is like a massive technological leap forward for humanity. Millions and millions of people's lives have been changed to a diagonal. But just, you know, how many cybercriminals have taken. Their whole concept of cybercriminals that really took off because of the Internet and how they took with them, taken advantage of it. So just like any other technology, AI is also like that and it can be misused. And so in order for these things to be control and we need to have some sort of a regulation sounds of our standards for AI. So why do you think I need to be going or not? Let's be honest. Nobody likes regulations, right? I'm thinking, especially in information technology. It is usually seem to be a blocker for generation. It seems like something which is impeached growth and innovation and technology. And AI historically was a self-regulated industry. There was, wasn't that much regulation. Governments didn't pick that much. It costed at initially. And what happened was when lots of things happen, AIs mass adoption suddenly really started. Covid-19 was a game changer for a lot of companies across the globe. And he angles one of them. A lot of companies, what they did was they accelerated the technology roadmap rate. They be accelerated adoption to the Cloud. Actually accelerated adoption of also the digitalization, e-commerce and AI. And the company is really relate. It cannot be like, just like the Wild West Nordic election is present and they release for the AI systems to be trusted by the customers, you need to have a broad regulatory framework to be present. And this is what the question comes in. Now, trust, how do you measure trust him and VA system? Because mixed Mike seems to be some sort of a paradox here, right? Yeah, It's not like humans, like the egg doesn't have emotions as an empire says. So right there, concept of trust command. Now this is where the interesting thing comes up. Now give me any bias. That's a good question to ask because humans have bias, right? Humans or prejudices, non-human is completely objective. That's what may get some sort of subjectivity, honest keeps, and that's why we need checks and balances. So the sad fact is, you remember what, how AI works like you need to give a training data on beyond the basis of which it starts to collect data and understand all the datasets that you're feeding it. Not biases during data collection from humans. Damn, keep in Belgrade, it comes down to it that if, let's take an example, you, you're creating an algorithm for facial recognition. And the trading data that you feed it. It is not representing all the groups. So let's say the training data you gave it, its 75, 75 percent male faces and only 25 percent female. And out of all of them, 80% of the total faces you find a demo by people and only 20 percent are black. Now what do you think will happen? Do you think that the model will recognize white people and males better? Of course, a good light because that cleaning data was not uniform, it was not fair. You didn't give it the full all the data that was needed. So let's take an example. Now this doesn't hurt anything. It was like Kim and I put in the links below and you can check it. Facebook's ads are a lot of them. It was found out that it was discriminating by gender and race. So what was happening was like when it comes to relate jobs of nurses and Secretary missed. It was targeting a descending node, family to woman. And if it was locally in jobs, no, like taxi drivers of something, it was mainly sending it to minorities. And it was high-paying jobs like, you know, like selling a house or something like that. It was targeting white people. So what had happened was, like I said, those are the biases in my algorithm that February reptilian because of the training data that was filled. Now this was also something very quite amazing. The racial bias for them and better health care musical bar comes and what didn't happen knows this hospital was trying to find people who would need more targeted medical care. They were at a higher risk. And what didn't it was targeting based on how much they were spending. Right now, what happened was black people were actually had more medical issues, but they were spending less. White, white, white people were spending more. So on the basis of that, that algorithm decided that black people need as much medical care as white people. And it didn't started, and it started, I think white people based on that. So you see how it happened, right, with condemn income inequality. That thing happened. And this was a very famous case and it was where it was in the news predominantly also, a fairly famous US justice system called compass, which we found out was actually biased towards black people. And this you more details in the coming slides, but I hope this gives you an idea. Ai can be prejudiced actually, it's not completely objective if you do not treat it properly. So this is how many? This is an example I was talking about. That system was called compass. So this is an actual example where biases and artificial intelligence can cause damage. To relax people really like damage we left consequences to people which can have lasting impact on their lives. Compass was a system being used in the United States. What does it use being useful? It was being used by the query, by the justice system to forecast which humanists were likely to re-offend. Okay, like let's recall what was the probability of them committing crimes? Again, based on assessments judge would make the judges would make decisions when everything from the amount of GLD people buy, the amount of bail. So you can imagine like people's lives are being impacted and what happened? Well, they find out that there was a bias against black people better than this algorithm. So black people, by giving them, giving higher points that they went, there was a higher chance of them are higher risk of reoffending despite the fact that, that similarly, I like white people who had a much higher criminal histories, they were being given low and low scores. As you can see here on the left, that this person, Bernard Papa, he had very few benefits, is very small, meaning offensive, but the white guy, yet like quite significant criminal record. And you can see, but the risk level is completely skewed. And the black guy has been given a high risk as opposed to the white guy. Same thing on the right side. During the early Robert Kennedy can see. So Robert Kennedy had been begging offense, but Becky tryptophans and no subsequent offenses get him his voice was six. While it seems really was loaded because clique so you can see how it was happening. What had happened was the data was being tethered to the questioners. They were being asked 137 questions based on which the system was giving the score. And it was not taking into account all the different things that for their advice and all that. So based on this, there was actually, I believe there was a non-profit organization called ProPublica. They highlighted this and it also became a big scandal for this company. And the report is publicly, it's publicly available. You can download it. So you can see now I hope you can see what are the differences. Not if you're a judge and you're assessing a Ts, right? And if a guy comes up to you, even if the guy has not done anything significant before, but if he gets a high risk of 10, don't you think that will impact you subconsciously, that bias will come in and you might give them a like a higher jail sentence or something like that. Now I hope this gives you an understanding of how dangerous it is and how much it can impact you. So let's see what we can do about this, okay. 18. Types of AI regulations: Hi guys. So in this section, as we discussed before and I showed you how systems could accidentally have prejudices like a crept into them because of the trading data and the biases that represented the data that aspiring to that. So the question now is how do we create trust? Neither society we should always be striving to develop AI technology that is completely fair for everyone. And even as for businesses, like as you become more and more land or machine learning artificial intelligence, you have to tackle this. So you're going to draw, you can imagine what would happen if you're using an and brought them on AI system. And it turns out that it is discriminating against certain customers. So that will be a huge repetition of back for you and return people or inform you. So what we need to create trust? Now, how do we create trust? Well, crust relies on four things which is integrity, explainability, fairness, and recently it's not. Let's see what happened. Well, what these means? The first term is integrity. What do you mean by integrity of your system under simply put, this is just like we check a home, right? If you buy a home, I needed to have a check to check the foundations to make sure it is strong and sturdy. Same thing like this. You need to check what does the training data definition fed into the system? What are the controls over this knee back? How was it built? Hey, how's it going? Goodness, something from start to finish so that no changes are happening in an algorithm is not being changed and continuous monitoring of it. So that's integrity, explainability. Well, that's pretty straightforward. What does it mean that you should be able to explain? The reason is a model makes a particular production and you should be able to understand the results should not be a black-box so that nobody knows how I did that. I got a commutative but particular decision. And that is essentially for costing it, especially if you're going to take decisions based on those acts, Richard Gibbs. So you need to be able to explain why and how a motto produced an output. And it should be able to do like presented to us at this, this is very exploitability comes up. Fairness, not this is called wind back and Oracle slice of tissue or do you want to trust any systemic fee if they're not fair? So it is, it should be built to be as free from bias as possible so that you know what the data you feed. It needs to be relevant, appropriate, and it should be allowed by the people to be short. So let's take an example. If you are EIA bought Occam is like giving schools to people and based on a course to coordinate since no, sorry, digitizer cognitive poor and they won't be able to pay back loan. So that is a bias, right? So you need to have to draw them with and need to be monitoring it continuously to make sure that they had no biases creeping into the training data. And resilience. Resilience is the technical robustness and compliance in other security, the risk management. Nobody's trying to compromise the algorithm and muddy anyway, somebody is trying to like what effects controls, you know, AI systems should be strong enough to resist such that x, that is y was against complex. So these are the four attributes which we're talking about. So I hope you understand if your AIE system is matching the sport requirements, then you can see, if you see that it does, it should retrospect. Now how do we get, What's the mechanism? But across the world, people are getting regulations across the world. And they're usually usually based like they're things that UAE, things in the US and across the border, you see lots and lots of governments clearing your collections. So one thing we despite significant on April 23rd, 2021, what has happened, the first concrete proposal, EIA has been saint like a draft I was created by the European Union. And this is, this is likely to perform the effectively the back and debate on artificial intelligence and the law. Whole lot of companies, be it small and large companies, they will hardly ever be using artificial intelligence as you can go through it. So if you're familiar, gdpr, GDPR was a Data Protection Regulation. And once it was largely, it quickly became the global standard for how other nations were looking at it as a blueprint. So the same thing will happen here also. That's for pretty much expected because the GDPR sets the tone for the whole world. Similarly, this recognition could set the tone for the whole world. This will probably come into effect in a second after 2022 in a transition period. So you can imagine by the second half of say, 2 thousand credit for that is, you can start at coming into effect and performing if they like, the mechanism of this thing coming into play. So what, what does this dream of doing exactly? So it basically divides the systems that could take categories mixed up people with high risk and limited and the limitless care systems in the last two, I like pretty much the same. And this is how they want to be when he called classifying and dealing with the I-Systems. So three main levels which they have divided, neglected certain acceptable risk, high risk, and limited with endless. An acceptable risk is basically systems which are considered to be like a threat to the safety and livelihood and rights of people that simply bank regulation of simply batting average, that it's like a zero-tolerance, those sort of things. And the limited and the minimum was gay. They are simply small things which are using Gately keeping your spam folder, you know, it uses machine learning to know how many images are coming in just span. It doesn't concern the main focus is on hydrosphere. What I'd also like basically, it's a high-level definition like a system's between evaluate to see customer, consumer credit worthiness or the critical magical blaze biometric identification, like those sort of things. And those will be subject to very strict obligations which before they can be put into the market, you will have risk assessments happening. The datasets we need to really heal. You need to provide information to be used, how it's happening. You need to have human oversight over it and highlighted robustness of security and accuracy. So how would you call what? Well, if you develop an AI system, it needs to look over it like something called the conformity assessment. So basically what it does, it leave somebody with technical documentation and the quality of the system so that it is conforming to go back nation. And if it does, it gets registered and what are you going to be placed on the market? And so supposing something changes, something changes empathy a system than you need to completely go back and we certify it by this li, all the things that you mentioned about fairness and integrity and all those things, they would be actually becomes a mechanist. So you can see this purple regulation into the secretory globally. So I hope you understand now guys, how what he called the risks which are presently a systems and how dominant across the world and the pudding and measures and putting it into the steps to make sure that ecosystems can be trusted. Thank you. 19. Way forward for you : Hi guys. So now we've finally reached the end of this course and this is our final section. I hope means I just journey. And you've gained like a thorough understanding of the basics of artificial intelligence and machine learning. So now the $1 million question is what? No, Where do you guys go from there and build on what you've learned in this course. So if you want to pursue a career, a well, the good news is that you have chosen a professional person, which isn't a huge demand. Even with the pandemic destroying millions of jobs. You have big giant investing heavily in AI. You have globins investing your startups indistinct. It is expected by 2030 that 1 third of that book in the US will be replaced by automation and robots. So I mean, this is just like there's something that I took from LinkedIn. I mean, the number one most in-demand job was artificial intelligence specialist hired like a 74 percent annual growth, which is absolutely amazing. So again, some doctrine analysis. Gardner, like I said, not even 2030, which as I like 2024 like 32 fourths of enterprise and operationalize a. So the good news is if you want to pursue a career in EIA, the good news is you've chosen a profession which means demand. Okay? And in the resources section, I've put in a link to a recent study by the UK government on how the, how to grow the artificial intelligence industry in the UK. And it's not just that you get this is happening pretty much on a global scale from Europe to the Middle East. Additionally, just know the European Union is port for the proposal for regulating EIA, which is the first-ever legal framework on the a, which addresses the risks of AI's ability when I CA is not going anywhere, guys are good. That's a good news. We're looking at what's the bad news? Well, the bad news, I don't know if you can call it bad news, but just something you have to keep in mind. It is a very technical field and you will need to dig deep into some technical subjects like programming and data science. If you want to create your own machine learning models, and honestly, if you want to do anything of worth NEA, that's the nature of the beast, that is AI. And there's no way to skip from it. Whether you want to become an AI engineer or a data scientist. I put in what, in my experience, what sort of skills you need to pursue. See the screen, yeah. Then numerous postgraduate, bachelor's and master's level programs available from accredited universities and institutions, which can take you from scratch and you can really build upon the skills. So if you are interested in skill and that's great, you know, a lot of these programs are not available. While ago you had to really it was like a really specialist programs with the knowledge become much more common. So it's much more easier to pursue a proper career. And now I mentioned programming language. So that's the question I get asked a lot to say. If I do okay, I do want to pursue AI and I want to pick up a programming language and which one should I focus on? Well, there are numerous options available for artificial intelligence, but my recommendation is almost always buy. There's a reason it's the most popular coding language for machine learning. The reason for that is it's filled with 316 liabilities. So you already a lot of work has been done and can help you get a running start. And it's very simple to understand by Ethan has always been popular, I'll be honest, but with AI and machine learning, it really took off. It's definitely, they will be like no. Another one is list. But it's an option, but all the water, most of it honestly it's not present in Python. It's not as user friendly as Python. It's all does not have the rich number of liabilities exciton has. Java is also quite popular and it's a strong contender against Python. It'll also as values and user-friendly. I think anybody who's worked in a large organization has experienced a dollar. And the last one, if you are a number cruncher, then r, which is an emerging language. It might be for you. It became very properly with them because it's good for statistical analysis. And it even better than Python when it comes to number-crunching. It has very powerful support for data mining and advanced data analysis. There are other language also like C plus plus prologue and I can go on and on about it. But if you honestly want my opinion dice, then you really cannot go wrong with Python. And there is a good reason, it isn't a moment trace for AI professionals. And then numerous courses available, you can take it up and you can get a very good, solid foundation for starting a career in ai. Okay, that was just a guide. Just to help you get a running start on the ion starting your Acharya, what to do? So I think we had reached a conclusion. Now. I'll see you there. Thank you. 20. The End !: Hi everyone, and congratulations, Thank you for completing the course and I hope you've learned something now about AI and machine learning. And it's no longer such a scary topic as it seemed to be at the start. So just a quick somebody, I mean, these other things, these other achievements that you have now done, you've understood the basic concepts of machine learning. You've created your own machine learning and AI models, and it sounds incredible, but yes, you have done that. And you've also realized the dangerous of the misuse of AI. Invite nice governing and some risk management frameworks we put in place. And you also understood, I hope I've given you the tools you need to have in place. If you really want to take your career forward. And like a really invest in artificial intelligence. I mean, the sky's the limit guys. Really, there's no stopping EA. And he's like, you're investing yourself for the future. So I hope this has really helped me out and please do give me your feedback and comments. I would love to have honest feedback that will help you to create more courses and you'll find this course more. So just a quick self-promotional again, I do have a YouTube channel, the Cloud security guy. This is my Facebook page and a blog on which adequately post. So I wish you the absolute best in your careers and good luck on your Acharya and see you in the future. Bye.