What is Artificial Intelligence?

 

Hi, I'm Ankit currently I enrolled for AWS Training and Certification.

 

In this article, 

We 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 we'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.

Full video link is here

Hi, I'm Dan Mbanga with AWS Training and Certification. Welcome to our introduction to Artif: Added to Selection. Press [CTRL