Hi, I'm Dan Mbanga with AWS Training and Certification. Welcome to our introduction to Artificial Intelligence. I have been with AWS for four years and I'm currently responsible for Business Development in Machine Learning and Deep Learning. As part of the Business Development team in Amazon AI, I have contributed to helping our customers bringing their strategy from conception to realization on the AWS environment. In this video, you will learn what Artificial Intelligence is, how it adds value to different businesses, and how Amazon uses AI in its products. We'll look at a use case where AI plays an important role, and you'll learn about some AWS services that you can use to develop an AI ready application. Simply stated, AI is intelligent behavior by machines, that means any device that can perceive its environment and take actions accordingly, has AI. By using AI, a machine can mimic cognitive human functions, like learning and problem-solving. A common example of using Artificial Intelligence is giving machines the ability to scan and interpret their physical environment, so that they can handle moving around and even up and down the stairs. To make the machines act and react like humans, we need to provide them with information from the real world, in order to mimic human intelligence AI relies on something called knowledge engineering. Knowledge engineering is the key component of AI research, machines with AI are expected to solve problems like humans would. To do that, machines need extensive knowledge of the real world. In other words, they need to understand things like the relationships between objects and situations, the properties of an event, cause and effect, and more. This data is then processed and fed to software programs that in turn analyze the data and come up with decisions for a particular problem, the way humans do. In short, the goal is to transfer human expertise to a software program, that can take in the same data and come to the same conclusions as humans would. This process of feeding data to a software program and coming up with human-like decisions is also known as the modeling process. The model, which is basically your software algorithm is consistently refined until its decisions are close to those a human would come up with. If the decision for a particular problem is inconsistent with what a human decision would be, then we go back to the model and debug it until we improve it. As you might expect, this is an iterative process. AI presents us with new possibilities and promotes growth in business, all kinds of companies are using AI to innovate. Companies are making significant investment to improve their products based on user satisfaction, feedback, trends and more, and they are using AI to do it. Here are a few examples of how AI is being used today, detecting and deterring security threats and fraud, resolving users technology issues through automated call center or chatbot, automating repeatable tasks such as payroll, data entry, and audit, anticipating users' actions and providing recommendations, monitoring social media comments, and tailoring advertising content as per search trends. Once you start learning about AI, you start seeing terms like Machine Learning and Deep Learning. Machine Learning also called ML and Deep Learning also called DL, are really subsets of AI. You can create an AI system with the help of ML and DL algorithms, for example a software program to predict user actions and suggest recommendations or a system that understands thoughts and sentences spoken by a human like Alexa. Let's talk about these fields and how they differ from each other. Machine Learning is often deployed where explicit programming is too rigid or impractical. Unlike regular computer code, Machine Learning uses data to generate statistical code that will output the right result, based on the pattern recognized from previous examples of input. Machine Learning starts with the data it already has about a situation. It processes data using algorithms to recognize patterns of a behavior and outcomes, it then interprets those patterns to predict the future outcomes. These predictions are used to make a decision about the next step for the Machine Learning to take. That decision produces results, which are then evaluated and added into the pool of data, the new data would influence the predictions and subsequent decisions made going forward, this is how Machine Learning learns over time. Machine Learning can make predictions from huge datasets, optimize utility functions, and extract hidden patterns and structures from the datasets by classifying data. This enables a software program to learn and make predictions in the future. Deep Learning takes Machine Learning a step further. Rather than telling the machine what features it needs to look for, Deep Learning enables the machine to define the features it needs to look for itself based on the data it's being provided. In this example, traditional Machine Learning requires you to tell the machine how to differentiate between a rectangle and a circle. Deep Learning on the other hand, shows machines several examples of rectangles. It analyzes those examples and infers common features that define a rectangle. At this point, it can identify on its own whether it's looking at a rectangle. In the same way our brains process information using neurons, Deep Learning processes information using similar but artificial processing structures known as artificial neural networks. It builds these structures from the data it analyzes, and then infers features about its subject matter based on the data. Then it weighs those features according to certainty and commonality, and organizes them into layers of hierarchies and relationships with each order. To return to the circle and rectangle example, if the Deep Learning machine looks at its reference data on what a rectangle is, it can infer that rectangles are built from four sides at right angles. Unlike Machine Learning, the Deep Learning machine doesn't have to be told to look for the number or angle or sides, instead, it recognizes the sides as a common feature of the reference data on its own. It can then look at the big blue rectangle, see that it has four sides at right angles, and determine with strong certainty that it's a rectangle. It can also determine that the purple square is probably a rectangle, since it also has four sides at right angles, even though its four sides appear to be equal and it's not of a color that is included in the reference data. To help understand the differences between AI, Machine Learning, and Deep Learning, let's go through a very high-level example of how these three might be applied to common task of facial recognition. In this example, an Artificial Intelligence wouldn't necessarily know that it was looking at three people, unless it has been thought what to look for in order to spot people. This requires a lot of trial and error on the part of the developers creating the algorithm, and it doesn't involve the machine having to learn anything about what humans look like, other than what the developers tell it to look for. The machine may be provided with the ability to identify head shapes or skin tones, but without the ability to learn, the machine could fail simply because of the wide range of diversity in what humans look like. For instance, it might not recognize a person because of a beard, which could generate a false negative. With Machine Learning however, you can give the machine a rough framework for what a person looks like and the ability to iteratively process and learn other human appearances through experience. So here the machine can recognize the figure in the middle, since it's the closest to the figure example it already knows with a similar facial shape and hair shape. Once it confirms that these new appearance is a person's face, it becomes more confident in its ability to recognize humans based on facial and hair shape, but less confident in brown hair color. With this new information, it might now be able to recognize person three as its confidence in facial shape is high enough to overcome its lack of knowledge in other areas such as hair-shape and skin tone. But because the machine was not prepared to recognize facial hair ahead of time, it still doesn't have the ability to recognize person one. That's why deep learning is such a popular choice for facial recognition. With deep learning, the machine is provided lots of facial reference data upfront, and unlike traditional Machine Learning or AI, it isn't always told exactly what features to look for. It uses it's highly advanced data processing capabilities and neural networks to derive the important features it needs to look for from the data itself. Rather than the developers telling the machine ahead of time how to recognize specific, how to define things like facial hair, the machine simply looks for the common features that define all of the humans in this data, and looks for those in the things that it sees. In other words, the machine defines the essential features of its subject rather than the developer. That's what distinguishes deep learning from the traditional machine learning. Now that we understand what AI is, let's talk about how to establish an effective AI strategy. You can establish an effective AI strategy in your organization with the help of fast computing environment, data gathered from various sources such as social media, browsing trends and more, and advanced learning algorithms. Let's start with the data. More data means better analytics and better analytics results in better products. Better products means more users and that in turn generates more data for you. This in simple terms is the flywheel for data. You can gather data from a number of sources like clickstream and user activity, then you can analyze it using tools like Hadoop, and Spark, and Amazon Elastic search surveys. Using the analysis, you can feed the AI and machine learning algorithms to form pattern recognitions and generate predictions. Then, you can use those predictions to make your products better and drive more users do it. By using a combination of programming models, algorithms, data, and hardware acceleration with infrastructure such as GPUs, you can develop a framework that helps with AI enabled features like image understanding, speech recognition, natural language processing, and autonomy. These combination of programming models, algorithms, and data is usually what forms the basis of machine learning and deep learning frameworks, and the underlying hardware infrastructure supports the frameworks. Today, AI is being used all across Amazon. On amazon.com, users see recommendations suggested by Amazon's recommendation engine, which improves their shopping experience. We also use AI to spot trends in the customer's experience so that we can develop new products and enhance existing products. In the fulfillment and logistic departments, robots pick, pile, sort, and move boxes around so that they can be shipped to customers. Our employees used to have to walk miles each day. By using AI, we save time and free up our staff to serve more customers faster. Now AWS is making AI tools broadly available so that businesses can innovate and improve their products. Amazon Web Services offers a range of services in AI by leveraging Amazon's internal experience with AI and machine learning. These services are separated here according to four layers, AI services, AI platforms, AI frameworks, and AI Infrastructure. They organize from the least complex to the most complex going from top to bottom. Let's take a brief look into each of these layers. Our AI services are each built to handle specific common AI tasks. These services enable developers to add Intelligence to their applications through an API called to pre-train services rather than developing and training their own deep learning models. Amazon Recognition makes it easy to add image analysis for your applications. With recognition, you can detect specific objects, scenes, and faces like celebrities and identify inappropriate content in images. You can also search and compare faces. Recognitions API enables you to quickly add sophisticated deep learning-based visual search and image classification to your applications. Amazon Polly is a service that turns texts into lifelike speech, allowing you to create applications that talk and build entirely new categories of speech enabled product. Amazon Polly's text-to-speech service uses advanced deep learning technologies to synthesize speech that sounds like human voice. Amazon Lex is a service for building conversational interfaces into any application using voice and text. It provides automatic speech recognition for converting speech-to-text and natural language understanding to recognize the intent of the text. That lets you build applications with highly engaging user experiences and life-like conversational interactions. The AI platforms layer of the stack includes products and frameworks that are designed to support custom AI related tasks, such as training and Machine Learning model with your own data. For customers who want to fully manage platform for building models using their own data, we have Amazon Machine Learning. It's designed for developers and data scientists who want to focus on building models. The Platform removes the undifferentiated overhead associated with deploying and managing infrastructure for training and hosting models. It can analyze your data, provide you with suggested transformations for the data, train your model, and even help you with evaluating your model for accuracy. Amazon EMR is a flexible, customizable, and manage big data processing platform. It's a manage solution in that it can handle things like scaling and high availability for you. Amazon EMR does not require a deep understanding of how to set up and administer Big Data Platforms, you get a preconfigured cluster ready to receive your analytics workload. It is built for any Data Science Workload not just AI. Apache Spark is an open-source, distributed processing system commonly used for Big Data workloads. Apache Spark utilizes in-memory caching and optimize execution for fast performance and it supports general batch processing, Streaming Analytics, Machine Learning, graph database, and ad hoc queries. It can be run and managed on Amazon EMR clusters. The AI frameworks and infrastructure layers are for expert machine learning practitioners. In other words, for the people who are comfortable building deep learning models, training them, doing predictions, also known as inference, and getting the data from the models into production applications. The underlying infrastructure consists of Amazon EC2 P3 instances, which are optimized for machine learning and deep learning. Amazon EC2 P3 instances provide powerful NVIDIA GPUs to accelerate computations, so that customers can train their models in a fraction of the time required by traditional CPUs. After training, Amazon EC2 C5 compute optimize and aim for general-purpose instances. In addition to GPU based instances, are well-suited for running inferences with the training model. AWS supports all the major deep-learning frameworks and makes them easy to deploy without AWS, deep-learning Amazon machine image, which is available for Amazon Linux and Ubuntu, so that you can create managed, automatically scalable clusters of GPUs for training and inference at any scale. It comes pre-installed with technologies like Apache MX net, tenser flow, Cafe and Caffe2 and auto-populate Machine Learning software such as the Anaconda package for data science. Now let's go through a few use cases. Almost all industry domains are now innovating with AWS AI. For example, for fraud dot net uses Amazon Machine Learning to support its Machine Learning models. The company uses Amazon DynamoDB and AWS Lambda to run code without provisioning and managing servers. Fraud.net also uses Amazon Redshift for data analysis. What are the benefits that they get from that setup? Fraud.net lunches and trains Machine Learning models in almost half the time it took on other platforms. It reduces complexity and makes sense of emerging Fraud patterns. It saves customers about a million dollars each week. To summarize, you can create an impact in your business by automating repetitive and manual tasks, engaging customers and optimizing product quality using AI. I hope you learned a little something and we'll continue to explore all the courses. I'm Dan Banger with AWS AI and thanks for watching.
