Introducing AI

At IBM, we define AI as anything that makes machines act more intelligently We like to think of AI
as augmented intelligence We believe that AI should not attempt to replace human experts, but
rather extend human capabilities and accomplish tasks that neither humans nor machines could do
on their own. 
The internet has given us access to more information faster. Distributed computing and IoT have led
to massive amounts of data, and social networking has encouraged most of that data to be
unstructured. With Augmented Intelligence, we are putting information that subject matter experts
need at their fingertips, and backing it with evidence so they can make informed decisions. We want
experts to scale their capabilities and let the machines do the time-consuming work. 

How do we define intelligence? 

Human beings have innate intelligence, defined as the intelligence that governs every activity in our
body. This intelligence is what causes an oak tree to grow out of a little seed, and an elephant to form
from a single-celled organism. How does AI learn? The only innate intelligence machines have is
what we give them. We provide machines the ability to examine examples and create machine
learning models based on the inputs and desired outputs. And we do this in different ways such as
Supervised Learning, Unsupervised Learning, and Reinforcement Learning, about which you will
learn in more detail in subsequent lessons. 

Based on strength, breadth, and application, AI can be described in different ways.

Weak or Narrow AI is AI that is applied to a specific domain. For example, language translators,
virtual assistants, self-driving cars, AI-powered web searches, recommendation engines, and
intelligent spam filters. Applied AI can perform specific tasks, but not learn new ones, making
decisions based on programmed algorithms, and training data. 
Strong AI or Generalized AI is AI that can interact and operate a wide variety of independent and
unrelated tasks. It can learn new tasks to solve new problems, and it does this by teaching itself new
strategies. Generalized Intelligence is the combination of many AI strategies that learn from
experience and can perform at a human level of intelligence.
Super AI or Conscious AI is AI with human-level consciousness, which would require it to be self-
aware. Because we are not yet able to adequately define what consciousness is, it is unlikely that we
will be able to create a conscious AI in the near future.
AI is the fusion of many fields of study.
Computer science and electrical engineering determine how AI is implemented in software and
hardware.
Mathematics and statistics determine viable models and measure performance. Because AI is
modeled on how we believe the brain works, psychology and linguistics play an essential role in
understanding how AI might work. And philosophy provides guidance on intelligence and ethical
considerations. 

While the science fiction version of AI may be a distant possibility, we already see more and more AI
involved in the decisions we make every day. Over the years, AI has proven to be useful in different
domains, impacting the lives of people and our society in meaningful ways.

What is AI
There's a lot of talk about artificial intelligence these days. How do you define or what does AI mean
for you? There is a lot of talk and there's a lot of definitions for what artificial intelligence says. So one
of them is about teaching the machines to learn, and act, and think as humans would. Another
dimension is really about how do we get the machines to- how do we impart more of a cognitive
capabilities on the machines and sensory capabilities. So it's about analyzing images and videos
about natural language processing and understanding speech. It's about pattern recognition, and so
on, and so forth. So the third axis is more around creating a technology that's able to, in some cases,
replace what humans do. I'd like to think of this as augment what humans do. To me personally, the
most important part of definition for artificial intelligence is about imparting the ability to
think and learn on the machines. To me that's what defines artificial intelligence.

AI is the application of computing to solve problems in an intelligent way using algorithms. So

what is an intelligent way? Well, an intelligent way may be something that mimics human intelligence.
Or it may be a purely computational approach and optimization approach but something that
manipulates data in a way to get not obvious results out,

I think, is what I would classify as being artificially intelligent. I would define AI as a tool that uses
computer to complete a task automatically with very little to no human intervention. 

To me AI is really a complex series of layers of algorithms that do something with the information
that's coming into it. 
Artificial intelligence is a set of technologies that allows us to extract knowledge from data. So it's any
kind of system that learns or understands patterns within that data, and can identify them, and then
reproduce them on new information.

Artificial intelligence is not the kind of simulating human intelligence that people think it is. It's really
not about intelligence at all. But I think another word that describes AI more accurately today is
machine learning. The reason I say that is because machine learning technology is all about using
essentially mathematics on computers in order to find patterns in data. Now this data can be
structured or unstructured. The only difference between machine learning and the technologies that
came before it is instead of us, as humans, having to manually hard code these patterns, and these
conditions into computers. They're able to find these patterns on their own by using math. That's
really the only difference here. So what I'd say artificial intelligence is, is it's a set of mathematical
algorithms that enable us to have computers find, very deep and patterns that we may not
have even known exist, without us having to hard code them manually.

Machine learning or AI is not simulating a human mind, but what it does try and do, is it tries to open
up new doors for computers. It tries to enable computers to understand certain kinds of data that they
couldn't have understood before. So for example, if you take a look at what we as humans are so
special at, the fact that we can understand natural language, in fact, we are the only animal to have
such a complex ability of being able to communicate in natural language, even if through something
that I have not directly witnessed or seen or heard evidence for, I can still describe that imaginative
concept to you, that is actually really wonderful.

We're also great at processing visual data, like when you look at someone's face, the fact that you
can instantly recognize them. When you look at someone's eyes, you can tell exactly where they're
looking, that's really an amazing ability. These are things that computers cannot do because they are
fundamentally limited to mathematics. They can only understand numbers and mathematical
operations. But by using machine learning technology, you can actually take these mathematics,
and use them to understand patterns in vast amounts of both structured 
and unstructured human data.

In this lesson, you have learned:

 IBM Research defines Artificial Intelligence (AI) as Augmented Intelligence, helping experts scale their
capabilities as machines do the time-consuming work.
 AI learns by creating machine learning models based on provided inputs and desired outputs.    
 AI can be described in different ways based on strength, breadth, and application - Weak or Narrow AI,
Strong or Generalized AI, Super or Conscious AI.
 AI is the fusion of many fields of study, such as Computer Science, Electrical Engineering, Mathematics,
Statistics, Psychology, Linguistics, and Philosophy.

Impact and Examples of AI

AI is here to stay, with the promise of transforming the way the world works. According to a study by
PWC, $16 trillion of GDP will be added between now and 2030 on the basis of AI. This is a never
before seen scale of economic impact, and it is not just in the IT industry, it impacts virtually every
industry and aspect of our lives. 

AI means different things to different people. 

For a videogame designer, AI means writing the code that affects how bots play, and how the
environment reacts to the player. 
For a screenwriter, AI means a character that acts like a human, with some trope of computer
features mixed in. 
For a data scientist, AI is a way of exploring and classifying data to meet specific goals. AI
algorithms that learn by example are the reason we can talk to Watson, Alexa, Siri, Cortana, and
Google Assistant, and they can talk back to us. 

The natural language processing and natural language generation capabilities of AI are not only
enabling machines and humans to understand and interact with each other, but are creating new
opportunities and new ways of doing business. Chatbots powered by natural language
processing capabilities, are being used in healthcare to question patients and run basic
diagnoses like real doctors. In education, they are providing students with easy to learn
conversational interfaces and on-demand online tutors. Customer service chatbots are improving
customer experience by resolving queries on the spot and freeing up agents time for conversations
that add value. AI-powered advances in speech-to-text technology have made real time transcription
a reality. Advances in speech synthesis are the reason companies are using AI-powered voice
to enhance customer experience, and give their brand its unique voice. In the field of medicine,
it's helping patients with Lou Gehrig's disease, for example, to regain their real voice in place of using
a computerized voice. It is due to advances in AI that the field of computer vision has been able
to surpass humans in tasks related to detecting and labeling objects. Computer vision is one of the
reasons why cars can steer their way on streets and highways and avoid hitting obstacles. Computer
vision algorithms detect facial features and images and compare them with databases of face
profiles. This is what allows consumer devices to authenticate the identities of their  owners through
facial recognition, social media apps to detect and tag users, and law enforcement agencies to
identify criminals in video feeds. Computer vision algorithms are helping automate tasks. Such as
detecting cancerous moles in skin images or finding symptoms in x-ray and MRI scan. AI is impacting
the quality of our lives on a daily basis. There's AI in our Netflix queue, our navigation apps, keeping
spam out of our inboxes and reminding us of important events. AI is working behind the
scenes monitoring our investments, detecting fraudulent transactions, identifying credit card
fraud, and preventing financial crimes. AI is impacting healthcare in significant ways, by helping
doctors arrive at more accurate preliminary diagnoses, reading medical imaging, finding appropriate
clinical trials for patients. It is not just influencing patient outcomes But also making operational
processes less expensive. AI has the potential to access enormous amounts of information, imitate
humans, even specific humans, make life-changing recommendations about health and finances,
correlate data that may invade privacy, and much more. 

There's all kind of different applications, obviously, there's healthcare, there's finance. The one that's closest to my
heart, of course, is robotics and automation.

In this lesson, you have learned:

AI-powered applications are creating an impact in diverse areas such as Healthcare, Education, Transcription, Law
Enforcement, Customer Service, Mobile and Social Media Apps, Financial Fraud Prevention, Patient Diagnoses,
Clinical Trials, and more.

Some of these applications include:

 Robotics and Automation, where AI is making it possible for robots to perceive unpredictable environments
around them in order to decide on the next steps.
 Airport Security, where AI is making it possible for X-ray scanners to flag images that may look suspicious.
 Oil and Gas, where AI is helping companies analyze and classify thousands of rock samples to help identify
the best locations to drill for oil?

Some famous applications of AI from IBM include:

 Watson playing Jeopardy to win against two of its greatest champions, Ken Jennings and Brad Rutter.
 Watson teaming up with the Academy to deliver an amplified Grammy experience for millions of fans.
 Watson collaborating with ESPN to serve 10 million users of the ESPN Fantasy App sharing insights that
help them make better decisions to win their weekly matchups.
To learn more about the applications of AI from IBM:

 Expert Insights: AI fast forwards video for sports highlights

 IBM Watson creates the first movie trailer for 20th Century Fox
 USTA uses IBM Watson to enhance player performance

Cognitive Computing (Perception, Learning, Reasoning)

AI is at the forefront of a new era of computing, Cognitive Computing. It's a radically new kind of
computing, very different from the programmable systems that preceded it, as different as those
systems were from the tabulating machines of a century ago. Conventional computing solutions,
based on the mathematical principles that emanate from the 1940's, are programmed based on rules
and logic intended to derive mathematically precise answers, often following a rigid decision tree
approach. But with today's wealth of big data and the need for more complex evidence-based
decisions, such a rigid approach often breaks or fails to keep up with available information. Cognitive
Computing enables people to create a profoundly new kind of value, finding answers and
insights locked away in volumes of data. Whether we consider a doctor diagnosing a patient, a
wealth manager advising a client on their retirement portfolio, or even a chef creating a new recipe,
they need new approaches to put into context the volume of information they deal with on a daily
basis in order to derive value from it. These processes serve to enhance human expertise. Cognitive
Computing mirrors some of the key cognitive elements of human expertise, systems that
reason about problems like a human does. When we as humans seek to understand something
and to make a decision, we go through four key steps. First, we observe visible phenomena and
bodies of evidence. Second, we draw on what we know to interpret what we are seeing to generate
hypotheses about what it means. Third, we evaluate which hypotheses are right or wrong. Finally, we
decide, choosing the option that seems best and acting accordingly. Just as humans become
experts by going through the process of observation, evaluation, and decision-making,
cognitive systems use similar processes to reason about the information they read, and they
can do this at massive speed and scale. Unlike conventional computing solutions, which can only
handle neatly organized structured data such as what is stored in a database, cognitive computing
solutions can understand unstructured data, which is 80 percent of data today. All of the
information that is produced primarily by humans for other humans to consume. This includes
everything from literature, articles, research reports to blogs, posts, and tweets. While structured data
is governed by well-defined fields that contain well-specified information, cognitive systems rely on
natural language, which is governed by rules of grammar, context, and culture. It is implicit,
ambiguous, complex, and a challenge to process. While all human language is difficult to parse,
certain idioms can be particularly challenging. In English for instance, we can feel blue because it's
raining cats and dogs, while we're filling in a form, someone asked us to fill out. Cognitive systems
read and interpret text like a person. They do this by breaking down a sentence
grammatically, relationally, and structurally, discerning meaning from the semantics of the
written material. Cognitive systems understand context. This is very different from simple speech
recognition, which is how a computer translates human speech into a set of words. Cognitive
systems try to understand the real intent of the users language, and use that understanding to draw
inferences through a broad array of linguistic models and algorithms. Cognitive systems learn, adapt,
and keep getting smarter. They do this by learning from their interactions with us, and from their own
successes and failures, just like humans do.

Lesson Summary

In this lesson, you have learned:

Cognitive computing systems differ from conventional computing systems in that they can:

 Read and interpret unstructured data, understanding not just the meaning of words but also the intent and
context in which they are used.
 Reason about problems in a way that humans reason and make decisions.
 Learn over time from their interactions with humans and keep getting smarter.

To learn more about how to get started with cognitive computing, read this post:

 How to get started with cognitive technology

Terminology and Related Concepts

Before we deep dive into how AI works, and its various use cases and applications, let's differentiate
some of the closely related terms and concepts of AI: artificial intelligence, machine learning, deep
learning, and neural networks. These terms are sometimes used interchangeably, but they do not
refer to the same thing. Artificial intelligence is a branch of computer science dealing with a
simulation of intelligent behavior. AI systems will typically demonstrate behaviors associated
with human intelligence such as planning, learning, reasoning, problem-solving, knowledge
representation, perception, motion, and manipulation, and to a lesser extent social
intelligence and creativity. Machine learning is a subset of AI that uses computer algorithms to
analyze data and make intelligent decisions based on what it has learned, without being explicitly
programmed. Machine learning algorithms are trained with large sets of data and they learn
from examples. They do not follow rules-based algorithms. Machine learning is what enables
machines to solve problems on their own and make accurate predictions using the provided
data. Deep learning is a specialized subset of Machine Learning that uses layered neural networks to
simulate human decision-making. Deep learning algorithms can label and categorize information and
identify patterns. It is what enables AI systems to continuously learn on the job, and improve the
quality and accuracy of results by determining whether decisions were correct. Artificial neural
networks often referred to simply as neural networks take inspiration from biological neural networks,
although they work quite a bit differently. A neural network in AI is a collection of small
computing units called neurons that take incoming data and learn to make decisions over
time. Neural networks are often layered deep and are the reason deep learning algorithms become
more efficient as the datasets increase in volume, as opposed to other machine learning algorithms
that may plateau as data increases. Now that you have a broad understanding of the differences
between some key AI concepts, there is one more differentiation that is important to understand, that
between artificial intelligence and data science. Data science is the process and method for
extracting knowledge and insights from large volumes of disparate data. It's an
interdisciplinary field involving mathematics, statistical analysis, data visualization, machine
learning, and more. It's what makes it possible for us to appropriate information, see patterns, find
meaning from large volumes of data, and use it to make decisions that drive business. Data Science
can use many of the AI techniques to derive insight from data. For example, it could use
machine learning algorithms and even deep learning models to extract meaning and draw
inferences from data. There is some intersection between AI and data science, but one is not a
subset of the other. Rather, data science is a broad term that encompasses the entire data
processing methodology. Well, AI includes everything that allows computers to learn how to solve
problems and make intelligent decisions. Both AI and Data Science can involve the use of big data
that is significantly large volumes of data. In the next few lessons, the terms machine learning, deep
learning, and neural networks will be discussed in more detail.

Machine Learning

Machine Learning, a subset of AI, uses computer algorithms to analyze data and make intelligent
decisions based on what it has learned. Instead of following rules-based algorithms, machine
learning builds models to classify and make predictions from data. Let's understand this by
exploring a problem we may be able to tackle with Machine Learning. What if we want to determine
whether a heart can fail, is this something we can solve with Machine Learning? The answer is, Yes.
Let's say we are given data such as beats per minute, body mass index, age, sex, and the result
whether the heart has failed or not. With Machine Learning given this dataset, we are able to learn
and create a model that given inputs, will predict results. So what is the difference between this and
using statistical analysis to create an algorithm? An algorithm is a mathematical technique. With
traditional programming, we take data and rules, and use these to develop an algorithm that will give
us an answer. In the previous example, if we were using a traditional algorithm, we would take the
data such as beats per minute and BMI, and use this data to create an algorithm that will determine
whether the heart will fail or not. Essentially, it would be an if-then-else statement. When we submit
inputs, we get answers based on what the algorithm we determined is, and this algorithm will not
change. Machine Learning, on the other hand, takes data and answers and creates the
algorithm. Instead of getting answers in the end, we already have the answers. What we get is
a set of rules that determine what the machine learning model will be. The model determines
the rules, and the if-then-else statement when it gets the inputs. Essentially, what the model
does is determine what the parameters are in a traditional algorithm, and instead of deciding
arbitrarily that beats per minute plus BMI equals a certain result, we use the model to determine what
the logic will be. This model, unlike a traditional algorithm, can be continuously trained and be used in
the future to predict values. Machine Learning relies on defining behavioral rules by examining
and comparing large datasets to find common patterns. For instance, we can provide a machine
learning program with a large volume of pictures of birds and train the model to return the label "bird"
whenever it has provided a picture of a bird. We can also create a label for "cat" and provide pictures
of cats to train on. When the machine model is shown a picture of a cat or a bird, it will label the
picture with some level of confidence. This type of Machine Learning is called Supervised Learning,
where an algorithm is trained on human-labeled data. The more samples you provide a
supervised learning algorithm, the more precise it becomes in classifying new data. Unsupervised
Learning, another type of machine language, relies on giving the algorithm unlabeled data and
letting it find patterns by itself. You provide the input but not labels, and let the machine infer
qualities that algorithm ingests unlabeled data, draws inferences, and finds patterns. This type of
learning can be useful for clustering data, where data is grouped according to how similar it is to its
neighbors and dissimilar to everything else. Once the data is clustered, different techniques can be
used to explore that data and look for patterns. For instance, you provide a machine learning
algorithm with a constant stream of network traffic and let it independently learn the baseline, normal
network activity, as well as the outlier and possibly malicious behavior happening on the network.
The third type of machine learning algorithm, Reinforcement Learning, relies on providing a
machine learning algorithm with a set of rules and constraints, and letting it learn how to
achieve its goals. You define the state, the desired goal, allowed actions, and constraints. The
algorithm figures out how to achieve the goal by trying different combinations of allowed actions, and
is rewarded or punished depending on whether the decision was a good one. The algorithm tries its
best to maximize its rewards within the constraints provided. You could use Reinforcement
Learning to teach a machine to play chess or navigate an obstacle course.

Machine Learning Techniques and Training

Machine Learning is a broad field and we can split it up into three different categories, Supervised
Learning, Unsupervised Learning, and Reinforcement Learning. There are many different tasks we
can solve with these. Supervised Learning refers to when we have class labels in the dataset and we
use these to build the classification model. What this means is when we receive data, it has labels
that say what the data represents. In a previous example, we had a table with labels such as age or
sex. With Unsupervised Learning, we don't have class labels and we must discover class
labels from unstructured data. This could involve things such as deep learning looking at pictures
to train models. Things like this are typically done with something called clustering. Reinforcement
Learning is a different subset, and what this does is it uses a reward function to penalize bad
actions or reward good actions. Breaking down Supervised Learning, we can split it up into three
categories, Regression, Classification and Neural Networks. Regression models are built by
looking at the relationships between features x and the result y where y is a continuous variable.
Essentially, Regression estimates continuous values. Neural Networks refer to structures that
imitate the structure of the human brain. Classification on the other hand, focuses on discrete values
it identifies. We can assign discrete class labels y based on many input features x. In a previous
example, given a set of features x, like beats per minute, body mass index, age and sex, the
algorithm classifies the output y as two categories, True or False, predicting whether the heart will fail
or not. In other Classification models, we can classify results into more than two categories. For
example, predicting whether a recipe is for an Indian, Chinese, Japanese, or Thai dish. Some
forms of classification include decision trees, support vector machines, logistic regression,
and random forests. With Classification, we can extract features from the data. The features in this
example would be beats per minute or age. Features are distinctive properties of input patterns that
help in determining the output categories or classes of output. Each column is a feature and each row
is a data point. Classification is the process of predicting the class of given data points. Our classifier
uses some training data to understand how given input variables relate to that class. What exactly do
we mean by training? Training refers to using a learning algorithm to determine and develop
the parameters of your model. While there are many algorithms to do this, in layman's terms, if
you're training a model to predict whether the heart will fail or not, that is True or False values, you
will be showing the algorithm some real-life data labeled True, then showing the algorithm again,
some data labeled False, and you will be repeating this process with data having True or False
values, that is whether the heart actually failed or not. The algorithm modifies internal values until it
has learned to tell from data that indicates heart failure that is True or not, that is False. With
Machine Learning, we typically take a dataset and split it into three sets, Training, Validation
and Test sets. The Training subset is the data used to train the algorithm. The Validation subset is
used to validate our results and fine-tune the algorithm's parameters. The Testing data is the data
the model has never seen before and used to evaluate how good our model is. We can then
indicate how good the model is using terms like, accuracy, precision and recall.

Deep Learning

While Machine Learning is a subset of Artificial Intelligence, Deep Learning is a specialized subset
of Machine Learning. Deep Learning layers algorithms to create a Neural Network, an artificial
replication of the structure and functionality of the brain, enabling AI systems to continuously learn on
the job and improve the quality and accuracy of results. This is what enables these systems to learn
from unstructured data such as photos, videos, and audio files. Deep Learning, for example, enables
natural language understanding capabilities of AI systems, and allows them to work out the
context and intent of what is being conveyed. Deep learning algorithms do not directly map
input to output. Instead, they rely on several layers of processing units. Each layer passes its
output to the next layer, which processes it and passes it to the next. The many layers is why it’s
called deep learning. When creating deep learning algorithms, developers and engineers
configure the number of layers and the type of functions that connect the outputs of each
layer to the inputs of the next. Then they train the model by providing it with lots of annotated
examples. For instance, you give a deep learning algorithm thousands of images and labels that
correspond to the content of each image. The algorithm will run the those examples through its
layered neural network, and adjust the weights of the variables in each layer of the neural network to
be able to detect the common patterns that define the images with similar labels. Deep Learning fixes
one of the major problems present in older generations of learning algorithms. While the efficiency
and performance of machine learning algorithms plateau as the datasets grow, deep learning
algorithms continue to improve as they are fed more data. Deep Learning has proven to be
very efficient at various tasks, including image captioning, voice recognition and
transcription, facial recognition, medical imaging, and language translation. Deep Learning is
also one of the main components of driverless cars.
Neural Networks

An artificial neural network is a collection of smaller units called neurons, which are
computing units modeled on the way the human brain processes information. Artificial neural
networks borrow some ideas from the biological neural network of the brain, in order to approximate
some of its processing results. These units or neurons take incoming data like the biological neural
networks and learn to make decisions over time. Neural networks learn through a process called
backpropagation. Backpropagation uses a set of training data that match known inputs to
desired outputs. First, the inputs are plugged into the network and outputs are determined. Then, an
error function determines how far the given output is from the desired output. Finally, adjustments are
made in order to reduce errors. A collection of neurons is called a layer, and a layer takes in an
input and provides an output. Any neural network will have one input layer and one output layer. It
will also have one or more hidden layers which simulate the types of activity that goes on in the
human brain. Hidden layers take in a set of weighted inputs and produce an output through an
activation function. A neural network having more than one hidden layer is referred to as a deep
neural network. Perceptrons are the simplest and oldest types of neural networks. They are single-
layered neural networks consisting of input nodes connected directly to an output node. Input layers
forward the input values to the next layer, by means of multiplying by a weight and summing the
results. Hidden layers receive input from other nodes and forward their output to other nodes. Hidden
and output nodes have a property called bias, which is a special type of weight that applies to a
node after the other inputs are considered. Finally, an activation function determines how a node
responds to its inputs. The function is run against the sum of the inputs and bias, and then the result
is forwarded as an output. Activation functions can take different forms, and choosing them is a
critical component to the success of a neural network. Convolutional neural networks or CNNs are
multilayer neural networks that take inspiration from the animal visual cortex. CNNs are useful in
applications such as image processing, video recognition, and natural language processing .
A convolution is a mathematical operation, where a function is applied to another function
and the result is a mixture of the two functions. Convolutions are good at detecting simple
structures in an image, and putting those simple features together to construct more complex
features. In a convolutional network, this process occurs over a series of layers, each of which
conducts a convolution on the output of the previous layer. CNNs are adept at building complex
features from less complex ones. Recurrent neural networks or RNNs, are recurrent because they
perform the same task for every element of a sequence, with prior outputs feeding subsequent
stage inputs. In a general neural network, an input is processed through a number of layers and an
output is produced with an assumption that the two successive inputs are independent of each other,
but that may not hold true in certain scenarios. For example, when we need to consider the context in
which a word has been spoken, in such scenarios, dependence on previous observations has to be
considered to produce the output. RNNs can make use of information in long sequences, each layer
of the network representing the observation at a certain time.

BM Research creates innovative tools and resources to help unleash the power of AI. In this hands on lab, you will
use a new kind of neural network, called a generative adversarial network (GAN), to create complex outputs, like
photorealistic images. You will use a GAN to enhance existing images and create your own unique, custom image.

Follow these steps to work with a GAN:

1. Access the demo here: This landscape image was created by AI.

2. In the Co-create with a neural network section, under Choose a generated image, select one of the existing
images. For example, choose the 11th image.

3. From the Pick object type list, select the type of object you want to add. For example, click on Tree to select it.

Figure 1 - Original generated image:

4. Move the cursor onto the image. Click and keeping the mouse button pressed, drag your cursor over an area of
the existing image where you want to add the object. For example drag a line in the red area highlighted in the red
rectangle below to add a tree there.
5. Choose another object type and add it to the image.

Figure 2 - Trees and grass added:

6. Experiment with locations: Can you place a door in the sky? Can you place grass so that it enhances the
appearance of existing grass? 7. Use the Undo and Erase functions to remove objects. 8. [Optional]
Click Download to save your work.

For more information on the capabilities of GANs, follow these steps:

1. In the What’s happening in this demo section? Click What does a GAN “understand” and read the text.
2. What does the text say about placement of objects? Does this explain the results you saw earlier?
3. Click Painting with neurons, not pixels and read the text. How does the GAN help you manipulate
images?
4. Click New ways to work with AI and read the text. What are some of the use cases for GANs?

Use the Discussion Forum to talk about these questions with your fellow students.

Lesson Summary

In this lesson, you have learned:

 Machine Learning, a subset of AI, uses computer algorithms to analyze data and make intelligent decisions
based on what it has learned. The three main categories of machine learning algorithms include Supervised
Learning, Unsupervised Learning, and Reinforcement learning.
 Deep Learning, a specialized subset of Machine Learning, layers algorithms to create a neural network
enabling AI systems to learn from unstructured data and continue learning on the job.
 Neural Networks, a collection of computing units modeled on biological neurons, take incoming data and
learn to make decisions over time. The different types of neural networks include Perceptrons, Convolutional
Neural Networks or CNNs, and Recurrent Neural Networks or RNNs.

In the Machine Learning Techniques and Training topic, you have learned:

 Supervised Learning is when we have class labels in the data set and use these to build the classification
model.
 Supervised Learning is split into three categories – Regression, Classification, and Neural Networks.
 Machine learning algorithms are trained using data sets split into training data, validation data, and test
data.

[Optional] To learn more about Machine Learning, Deep Learning, and Neural Networks, read
these articles:

 Models for Machine Learning

 Applications of Deep Learning
 A Neural Networks deep dive

Key Fields of Application in AI

So can you talk about the different areas or categories of artificial intelligence? Now, there are lots of
different fields that AI works in. But if I were to on a very very high level group some of the major
areas where artificial intelligence is applied, I'd like to start off with natural language. Because
natural language is, I'd say, the most complex data for machine learning to work with. If you
see all sorts of data, whether that be a sequence to genome, whether that be audio, whether that be
images. There's some sort of discernible pattern. There's some sort of yes, this is what a car sounds
like or yes, this is what human voice sounds like. But natural language is fundamentally, a very
human task. It's very human data source. We as humans invented it for humans to understand. If I
were to, for example, give you a book title, there's actually a very very famous book, and the title of
the book is there are two mistakes in the the title of this book. Now, there's actually only one mistake,
the two the's. The human brain doesn't realize that. What's the second mistake? That there was only
one mistake. So this is a sort of natural language complexity that's involved here. Humans we don't
view natural language literally. We view it conceptually. If I were to write a three instead of an E, you
will understand it because we don't mean the three in a literal sense. We mean that in a symbolic
sense to represent the concept of E and you can contextualize that three to figure out that, "Yeah. It
means in E" and not an actual three. These are things that computers aren't capable of. So natural
languages that number one field that I'm most interested in when it comes to machine learning.
Second, I'd say the most popular would be visual. Visual data understanding, computer vision.
Because it enables us to do so many things. As humans, our primary sense is vision. In fact, a vast
majority of your brain's processing power at any given moment, goes to understanding what it is that
you're seeing. Whether it be a person's face, or whether it be a computer or some texts, or anything
of that sort. Third, I would say audio-based data. So text-to-speech, speech-to-text these are very
very complex. The reason it's complex is because it combines a lot of challenges into one. First of all,
you've got to support many languages. You can't just support English and call it a day. You've got to
support other languages. You've got to support other demographics. Another challenge is that even
within languages, there are absolutely infinite number of ways that any human could represent a
language. Everyone's going to have a different accent. Everyone's going to have a different way of
pronouncing certain words. There's no standardized way that every human will pronounce ice cube
exactly like ice cube. That doesn't exist. If you take a look at another challenge, it's that audio data is
fundamentally very very difficult to work with. Because the thing is, audio data exists in the natural
world. What is audio? It's vibrations of air molecules, and vibrations of air molecules are fast.
Audio is recorded at overpay say 44 kilohertz. That's a lot of data, 44,000 data points every single
second. There are usually only 44,000 data points in an individual low-resolution image. So of
course, there are lots of challenges to work around when it comes to audio. But companies like IBM,
Google, Microsoft have actually worked around these challenges and they're working towards
creating different services to make it easier for developers. So again, on a very very high level,
there's natural language understanding, there's computer vision, there's audio data and of course,
there's the traditional set of tabular data understanding. Which is essentially, structured data
understanding.

Natural Language Processing, Speech, Computer Vision

Some of the most common application areas of AI include natural language processing, speech, and
computer vision. Now, let's look at each of these in turn. Humans have the most advanced method of
communication which is known as natural language. While humans can use computers to send voice
and text messages to each other, computers do not innately know how to process natural language.
Natural language processing is a subset of artificial intelligence that enables computers to
understand the meaning of human language. Natural language processing uses machine learning
and deep learning algorithms to discern a word's semantic meaning. It does this by deconstructing
sentences grammatically, relationally, and structurally and understanding the context of use. For
instance, based on the context of a conversation, NLP can determine if the word "Cloud" is a
reference to cloud computing or the mass of condensed water vapor floating in the sky. NLP systems
might also be able to understand intent and emotion, such as whether you're asking a question out of
frustration, confusion, or irritation. Understanding the real intent of the user's language, NLP systems
draw inferences through a broad array of linguistic models and algorithms. Natural language
processing is broken down into many subcategories related to audio and visual tasks. For computers
to communicate in natural language, they need to be able to convert speech into text, so
communication is more natural and easy to process. They also need to be able to convert text-to-
speech, so users can interact with computers without the requirement to stare at a screen. The older
iterations of speech-to-text technology require programmers to go through tedious process of
discovering and codifying the rules of classifying and converting voice samples into text. With neural
networks, instead of coding the rules, you provide voice samples and their corresponding text. The
neural network finds the common patterns among the pronunciation of words and then learns to map
new voice recordings to their corresponding texts. These advances in speech-to-text technology are
the reason we have real time transcription. Google uses AI-powered speech-to-text in there Call
Screen feature to handle scam calls and show you the text of the person speaking in real time.
YouTube uses this to provide automatic closed captioning. The flip side of speech-to-text is text-to-
speech also known as speech synthesis. In the past, the creation of a voice model required hundreds
of hours of coding. Now, with the help of neural networks, synthesizing human voice has become
possible. First, a neural network ingests numerous samples of a person's voice until it can tell
whether a new voice sample belongs to the same person. Then, a second neural network generates
audio data and runs it through the first network to see if it validates it as belonging to the subject. If it
does not, the generator corrects its sample and reruns it through the classifier. The two networks
repeat the process until they generate samples that sound natural. Companies use AI-powered voice
synthesis to enhance customer experience and give their brands their unique voice. In the medical
field, this technology is helping ALS patients regain their true voice instead of using a computerized
voice. The field of computer vision focuses on replicating parts of the complexity of the human visual
system, and enabling computers to identify and process objects in images and videos, in the same
way humans do. Computer vision is one of the technologies that enables the digital world to
interact with the physical world. The field of computer vision has taken great leaps in recent years
and surpasses humans in tasks related to detecting and labeling objects, thanks to advances in deep
learning and neural networks. This technology enables self-driving cars to make sense of their
surroundings. It plays a vital role in facial recognition applications allowing computers to match
images of people's faces to their identities. It also plays a crucial role in augmented and mixed reality.
The technology that allows computing devices such as smartphones, tablets, and smart glasses to
overlay and embed virtual objects on real-world imagery. Online photo libraries like Google Photos,
use computer vision to detect objects and classify images by the type of content they contain.

Self Driving Cars

Can you tell us a little bit about the work you're doing with self-driving cars. >> I've been working on
self-driving cars for the last few years. It's a domain that's exploded, obviously, in interest since early
competitions back in the 2005 domain. And what we've been working on really is putting together our
own self-driving vehicle that was able to drive on public roads in the regional Waterloo last August.
With the self-driving cars area, one of our key research domains is in 3D object detection. So this
remains a challenging task for algorithms to perform automatically. Trying to identify every vehicle,
every pedestrian, every sign that's in a driving environment. So that the vehicle can make the correct
decisions about how it should move and interact with those vehicles. And so we work extensively on
how we take in laser data and vision data and radar data. And then fuse that into a complete view of
the world around the vehicle. >> When we think of computer vision, we usually think immediately of
self-driving cars, and why is that? Well, it's because it's hard to pay attention when driving on the
road, right? You can't both be looking at your smartphone and also be looking at the road at the
same time. Of course, it's sometimes hard to predict what people are going to be doing on the street,
as well. When they're crossing the street with their bike or skateboard, or whatnot. So it's great when
we have some sort of camera or sensor that can help us detect these things and prevent accidents
before they could potentially occur. And that's one of the limitations of human vision, is attention, is
visual attention. So I could be looking at you, Rav, but behind you could be this delicious slice of
pizza. But I can only pay attention to one or just some limited number of things at a time. But I can't
attend to everything in my visual field all at once at the same time like a camera could. Or like how
computer vision could potentially do so. And so that's one of the great things that cameras and
computer vision is good for. Helping us pay attention to the whole world around us without having us
to look around and make sure that we're paying attention to everything. And that's just in self-driving
cars, so I think we all kind of have a good sense of how AI and computer vision shapes the driving
and transportation industry. >> Well, self-driving cars are certainly the future. And there's tremendous
interest right now in self-driving vehicles. In part because of their potential to really change the way
our society works and operates. I'm very excited about being able to get into a self-driving car and
read or sit on the phone on the way to work. Instead of having to pilot through Toronto traffic.
Play video starting at 2 minutes 58 seconds and follow transcript
2:58
So I think they represent a really exciting step forward, but there's still lots to do. We still have lots of
interesting challenges to solve in the self-driving space. Before we have really robust and safe cars
that are able to drive themselves 100% of the time autonomously on our roads.
Play video starting at 3 minutes 18 seconds and follow transcript
3:18
>> We've just launched our own self-driving car specialization on Coursera. And we'd be really happy
to see students in this specialization also come and learn more about self-driving. It's a wonderful
starting point, it gives you a really nice perspective on the different components of the self-driving
software stack and how it actually works. So everywhere from how it perceives the environment, how
it makes decisions and plans its way through that environment. To how it controls the vehicle and
makes sure it executes those plans safely. So you'll get a nice broad sweep of all of those things from
that specialization. And from there you then want to become really good and really deep in one
particular area, if you want to work in this domain. Because again, there's so many layers behind this.
There's so much foundational knowledge you need to start contributing that you can't go wrong. If
you find something interesting, just go after it. And I am sure there'll be companies that'll need you for
this.

Computer Vision
IBM Research creates innovative tools and resources to help unleash the power of AI. In this hands on lab, you will
learn about IBM’s Adversarial Robustness Toolbox, and use it to mitigate simulated attacks by hackers.
Follow these steps to explore the demo:

1. Access the demo here: Your AI model might be telling you this is not a cat.

2. In the Try it out section, click the image of the Siamese cat.

Figure 1 - Select an image

3. In the Simulate Attack section, ensure that no attack is selected, and that all the sliders are to the far left,
indicating that all attacks and mitigation strategies are turned off.

What does Watson identify the image as, and at what confidence level? E.g. Siamese cat 92%

4. In the Simulate Attack section, under Adversarial noise type, select Fast Gradient Method. The strength

slider will move to low.

Figure 2 - Select an attack and level

What does Watson identify the image as now, and at what confidence level?

3. In the Defend attack section, move the Gaussian Noise slider to low.

Figure 3 - Mitigate the attack

4. What does Watson identify the image as now, and at what confidence level? Did the image recognition improve?

Figure 4 - View the results

Note that you can use the slider on the image to see the original and modified images. 

5. Move the Gaussian Noise slider to medium, and then to high. For each level, note what Watson identifies the
image as, and at what confidence level. Did the image recognition improve?

6. Move the Gaussian Noise slider to None.

7. In the Defend attack section, move the Spatial Smoothing slider to low. What does Watson identify the image
as now, and at what confidence level? Did the image recognition improve?

8. Move the Spatial Smoothing slider to medium, and then to high. For each level, note what Watson identifies
the image as, and at what confidence level. Did the image recognition improve?

9. Move the Spatial Smoothing slider to None.

10. In the Defend attack section, move the Feature Squeezing slider to low. What does Watson identify the image
as now, and at what confidence level? Did the image recognition improve?

11. Move the Feature Squeezing slider to medium, and then to high. For each level, note what Watson identifies
the image as, and at what confidence level. Did the image recognition improve?

12. Which of the three defenses would you use to defend against a Fast Gradient Attack?

Optional:

If you have time, use the same techniques to explore the other methods of attack (Projected Gradient Descent and
C&W Attack) and evaluate which method of defense works best for each. If you want, try a different image.

Use the Discussion Forum to talk about the attacks and mitigation strategies with your fellow students.

Lesson Summary

In this lesson, you have learned:

 Natural Language Processing (NLP) is a subset of artificial intelligence that enables computers to
understand the meaning of human language, including the intent and context of use.
 Speech-to-text enables machines to convert speech to text by identifying common patterns in the different
pronunciations of a word, mapping new voice samples to corresponding words.
 Speech Synthesis enables machines to create natural sounding voice models, including the voice of
particular individuals.
 Computer Vision enables machines to identify and differentiate objects in images the same way humans do.
 Self-driving cars is an application of AI that can utilize NLP, speech, and most importantly, computer vision.

To learn more about Natural Language Processing, read this article:

 A beginner's guide to Natural Language Processing

Issues and Concerns around AI

Can you talk about issues and concerns surrounding AI? >> So I would say the number one key
challenge is privacy.

Play video starting at 21 seconds and follow transcript0:21

It's been a challenge getting through to patients, to care providers, to hospital administrators, that
their information is safe.

Play video starting at 30 seconds and follow transcript0:30

So what we've had to do is prove that we're actually anonymizing the data, we're de-identifying it
from the patient themselves and then blending it into the system. So that's something that we've
had to educate our users on to kind of get through that barrier. Because it is information that we
need. We do need their information in order to provide the better health outcomes. >> Yeah, so, any
new technology produces a bit of angst in all of us. Change is inevitable, but change is scary,
especially when that scare factor is enhanced and propagated by popular movies and social media.
So it's only natural. I think at the same time, we view this ethical dilemma as being not something that
it currently presents, but it's something that we're looking in the future to be more prevalent. I think it's
something we need to start caring about today and not wait until the year 2050. So some of people
have described AI as the last invention humankind is ever going to make. And it's basically
driving us towards this point where artificial intelligence will eclipse biological intelligence.
And at that point in time, humans will become irrelevant. >> And in terms of ethics when it comes to
machine learning technology, this is another one of those sort of gray areas. Where we still have to
sort of catch up and think about exactly the law and stuff that goes behind machine learning or AI.
And we have to do this with a fundamental assumption that machine learning technology's not
human. It's just another computer. Like, for example, I have an application called Make Your Own
Mozart that enables you to generate your own pieces of music based off of the style of music from
Mozart, Beethoven, their piano sonatas. Now the thing is, people ask me, well, who would own the IP
of the music generated by this neural network? And the thing is, it's a very difficult question to
answer, but if you really think about it, it's actually pretty simple at its core. Now, the machine learning
algorithm is just another algorithm. So when it generates a piece of music, for example, whoever
owns the algorithm, whoever owns the rights to use the algorithm is the one that owns the piece of
music it generated. So, for example, let's just say I was a programmer and I programmed this for
Company XYZ and now XYZ has the rights to use this application, it's their music once they generate
it with a neural network. So there are many different concerns to work through here. This was just
one of them. >> As humans, we have a history of creating technology that can be destructive for our
human race. In July 16th 1945, in the New Mexico desert, we have exploded a nuclear device. And
we've been struggling in containing that technology ever since. We're always living in the fear of us
wiping out the civilization with that device. As humans, we've made an imprint on our planet and our
environment. We we have a first hand in deforestation. We've dried up entire seas. And these are all
things that have consequences that can be fundamentally destructive in terminating our society as
we know it. Where we're now looking for plan b of relocating the human civilization to Mars. So
artificial intelligence is no different. Artificial Intelligence can be used for nefarious reasons and can
possibly be our last great invention, or it can be one of our greatest inventions. It's how we apply it.
>> It's not like when you see in the movies AI is going to be some big, here's some monster that
takes us over, it's just a tool and it's how we use it.

AI and Ethical Concerns

What are some of the ethical issues and concerns around AI? Ethics issues is another big question.
There's certainly many capabilities in automation and in AI that are emerging that are going to lead to
all abilities to track humans endlessly to identify actions and behaviors way beyond what's been
possible before, and so we need to have a keen eye on what's allowed and what isn't allowed and
how these technologies can be used in impact society. There are always ways to work around this,
because technology fundamentally since the very beginning of human history. There have been
technologies used for good purposes bad purposes. No matter how much of a good intent you put
behind using something for good, it's going to be used and misconstrued in some way to be used for
bad. Like fire. It's great for cooking food, but it's also great for burning people's houses down. But at
the same time, you can use it to chase away the people who were burning people's houses with. So
that's a good analogy for what we're doing with machine-learning. It'll be used for bad things by
humans, but we will use machine learning to counter that. But it won't itself turn into this horrible thing
for machines or for humans. The key point from my viewpoint is is to think of ethics in Artificial
Intelligence not as something that somebody else was going to do in the future, is something when
you felt like it today. So for example, today, police forces in countries like China and soon in Russia,
are starting to use glasses, do image recognition, and face recognition, identify wanted person,
person with warrants, for example, in a crowd and be able to better enforce law and maybe arrest a
person or investigate this further. That's a technology that's in use today. It can be used for great
good of reducing the crime. Simplifying the job of the law enforcement or it can be used for
nefarious reasons for a dictatorial government to enforce their will on the people, on arresting
and suppressing dissent, on suppressing democracy, and other. It's really is. Ethics is not a
technological problem, ethics is a human problem. So that's something that all of us need to care
about. Recognizing this leading companies in the space of Artificial Intelligence have banded
together to create consortia. So companies like Apple, IBM, Amazon, Microsoft, DeepMind,
Google, all got together and they're all advocating best ways off applying ethics to this technology. It's
interesting, particularly in self-driving cars, there's a lot of ethical questions that are emerging that
people are very worried about. The classic one is the trolley problem. If the car has to decide
which accident to cause. If it has to pick between running into a sign and hurting passengers in the
vehicle, or running into pedestrians on the side of the road, but potentially saving the passengers.
How do we make those decisions? So it's a really interesting area and lots of open questions there,
and it's unclear how we can properly define regulations so that different companies that are
producing these vehicles all behave in consistent ethical manner that meets our expectations.
They're also many concerns when it comes to for example, self-driving cars. For example, there's a
self-driving car and it has the option to injure another person on the road or the person in the car.
What would it choose? Or maybe a pedestrian. Now, the thing is, it might seem like a very difficult
decision as to who would be to blame for the choice that the car makes, but again, we face these
issues throughout history. For example, if you take a look a few 100 years ago when humans would
drive horse carriages. Let's just say there was a horse carriage traveling. There was a pedestrian in
the middle, weird noise, or did something that the horse didn't like, and the horse trampled a
pedestrian. Who do you blame? You can't blame the carriage driver, because he didn't have any
control of the situation. The horse can't be punished. What's the point of that? I can't understand. It
doesn't have that long-term memory planning ability, and the pedestrian, they can't be blamed,
because they were just doing what they were doing in public, and that's not illegal. So it's very difficult
to work around these issues, but again, looking into the past see into the future, we will be able to
work around them.

AI and Bias

So we hear about concerns and issues with AI. By us being an example, can you talk a little bit about
that? So to start off with the bias problem, this is a very famous problem within the field of machine
learning. To give an example, let's go back to the Microsoft Surface facial recognition. So Microsoft
Surface facial recognition is great system that works very nicely. But there's a little problem. It has a
very difficult time understanding and recognizing the faces of people in certain demographics. The
reason for this is because well, there simply wasn't enough representative data of those
people's faces in the training dataset that Microsoft used. This is really something that's very
difficult to get around. We certainly have to be cognizant with AI systems of things like systematic
bias and ethical issues related to AI. We need to be sure that the data that we feed to our AI
systems does not have or does not contain bias. Or that we are able to adjust for that bias, so
that we make sure we're not misrepresenting the population as a whole or preferring certain groups
over others. I think this is still a very challenging issue that we'll need to work through as these
AI systems are developed. First of all, machine learning technology is inherently biased.
Machine learning works off of the fundamental assumption of bias. That's how machine learning
technology functions. It's biasing certain input data points to map them to other output data
points. That's just fundamentally how they work. That's the mathematics behind machine learning.
There is also the option of directly modifying the data we feed in through techniques like data
augmentation to enable less biased data. But there is a problem. You see as humans we
consciously know not to apply certain kinds of bias. But the thing is, subconsciously, we end up
applying such certain kinds of bias that we really can't even control. On a massive scale with
hundreds of thousands of humans, this bias really shows up in data. So we moving it from data isn't a
good option either. You need a combined approach at solving the bias problem. There are numerous
researchers working on solving this problem. I'm sure there will be huge breakthroughs in this field
very very soon. But currently, this is an active area of research.

AI: Ethics, Bias, and Trust

AI is extremely powerful and can be used in ways that negatively affect society. To develop and use
AI systems responsibly, AI developers must consider the ethical issues inherent in AI. They must
have a realistic view of their systems and their capabilities and be aware of the different forms of bias
potentially present in their systems. With this awareness, developers can avoid unintentionally
creating AI systems that have a negative rather than positive impact. Even outside the realm of
science fiction some conversations about the ethics of AI suppose that AI will become sentient, that is
become self-aware, develop its own moral code and turn against its inventors. Nick Bostrom of
Oxford University says in the very long term machine super intelligence might eventually match
human intelligence and that when that happens there may be risks to human society .
Professor Stephen Hawking expressed reservations about the future of AI by saying "the rise of
powerful AI will either be the best or the worst thing to happen to humanity we do not yet know which
". Elon Musk went even further by saying "AI is more dangerous than nuclear weapons". But there
are also ethical issues concerning AI that must be addressed in the medium-term, short-term and
now. In developing AI systems experts must guard against introducing bias whether gender, racial,
social or any other form of bias. Early AI systems were prone to bias, an example being image
recognition software reflecting the unconscious bias of its developers. Due to poorly selected training
data it identified scenes showing kitchens laundry and shops with women and scenes showing sports
coaching and shooting with men. AI powered facial recognition systems provide another
example of an unintentionally bias system. In one case a system proved to be far more effective
at recognizing individuals with lighter toned skin than individuals with darker toned skin. Again, this
was discovered to be because the training data provided to the AI system was not so varied featuring
more individuals with lighter toned skin than those with darker toned skin. These types of bias can be
extremely dangerous when used in real life setting. For example, there are AI powered risk
assessment systems used in courts that help predict the probability of a person reoffending and
hence provide guidelines for sentencing or granting parole based on the calculated risk of recidivism.
There is concern that these systems may be biased against people of color. Developers of AI
systems can guard against introducing bias by providing effective training data and
performing regular tests and audits to ensure the system is performing as expected. Another
ethical issue that has arisen is a person entitled to know when they are speaking to a human being
and when they are speaking to a bot. As many bots are now indistinguishable from human beings.
Especially during short conversations a lack of trust in AI systems can be exacerbated by lack of
transparency. The sudden discovery that one is not speaking to a human being but rather a bot can
be unsettling and can make the transaction feel unequal. Trust is key in developing useful,
successful AI systems. For developers there are four aspects of AI that help people perceive it as
trustworthy. Transparency - people should be aware when they are interacting with an AI system
and understand what their expectations for the interaction should be; Accountability - developers
should create AI systems with algorithmic accountability so that any unexpected results can be traced
and undone if required; Privacy - personal information should always be protected; Lack of bias -
developers should use representative training data to avoid bias and regular audits to detect any bias
creeping in. You, as a consumer of services provided by AI should take an informed and balanced
view. Remember that every news article, every photo or video shared on social media and every
advert has been created by someone and their motivations may be different from yours.

Detect the Bias

Biased data may lead to biased predictions, and AI models are only as fair as their training data. It’s important to be
aware of the possibility of bias, so that you can mitigate against it.

In this hands on lab, you will see how different datasets affect the predictions of an AI model. Rate the fairness of
the outcomes, and see if other people agree. Then learn why reducing bias in AI models is so important.

Note: Detect the Bias is a live online demo, so you may see different models and datasets than those pictured
below.

Follow these steps to explore the demo:

1. Access the demo here: Detect the Bias
2. On the Detect the Bias page, read the introductory text, and then click Get Started.

3. In the Consent message box, click I agree, and then click Next.

4. Page 1 of 3 describes the data the model uses, and how it might affect the defendant. Read the text, and then
click Next.

5. Page 2 of 3 tells you that the data has now been preprocessed to make the decision more fair. Read the text, and
then click Next.

6. Page 3 of 3 presents a rules-based model, and the outcome for a particular group. Read the page and then
click FAIR or UNFAIR based on your evaluation of the model.

7. Click Submit Answer.
The Round 1 page shows the decision tree the model used to make the decision.

8. Read the page and then click FAIR or UNFAIR based on your evaluation of the model.

9. Click Submit Answer.

10. The message box confirms your choice, and tells you how many people agreed with you. Click Next.

The Round 2 page shows the weighted model with the relative weights applied to key attributes that the model
used to make this decision.

11. Read the page and then click FAIR or UNFAIR based on your evaluation of the model.
12. Click Submit Answer.

13. The message box confirms your choice, and tells you how many people agreed with you. Click Next.

The Round 3 page shows the decision tree the model used to make this decision.

14. Read the page and then click FAIR or UNFAIR based on your evaluation of the model.
15. Click Submit Answer.

16. The message box confirms your choice, and tells you how many people agreed with you.Click Next.

The Round 4 page shows the decision tree the model used to make this decision.

17. Read the page and then click FAIR or UNFAIR based on your evaluation of the model.
18. Click Submit Answer.

19. The message box confirms your choice, and tells you how many people agreed with you. Click Next.

The final page summarizes the results.

20. If you wish to provide your age and gender, do so. It is not required.

21. Click Submit Answer.

[Optional]: On the Learn More page, click the link to get more information about IBM’s research on Fairness in
Models.

22. Click Exit.
Use the Discussion Forum to talk about the models with your fellow students.

 Which model did you feel was most fair?

 What attribute seems to have most effect on re-offending?

Jobs and AI

Please talk to us about some of the concerns and issues surrounding AI. Yeah, absolutely. There's a
lot of growing concern around AI replacing jobs, and taking over some of the more automatable tasks
or manual tasks. Now if you think about the economics of things, it's very clear. We wouldn't think
twice about regretting that the refrigerator was invented, or that the bicycle was invented, or that the
car was invented. You nor I would probably think of creating a time machine to go back to prevent the
invention of the car from being ever invented in the first place. Just because it has transformed our
lives in ways that we could never have imagined, in the same sense had potentially AI could
potentially transform, and has already begun to transform our lives in very meaningful ways. Through
the process, there are going to be inevitably some jobs that are going to be automated, or replace
like ice pickers who were trying to acquire a large blocks of ice cubes to bring it to towns, in homes so
they can refrigerate the food, those jobs are replaced by the refrigerator, and also horse-drawn
carriages, they were replaced by bicycles and cars as well. So technology is always going to
transform society. My views is that we've been through many revolutions where technology has had a
profound impact on the type of jobs that humans need to perform. We no longer have horse and
buggy drivers. There's a lot of changes that have had big impacts on the way people make their
livings, and so AI is going to be no different. It's going to have a very large impact on automation, and
automatable tasks. So anything that's repeatable even in the office setting, repeatable tasks, the
document generation in the industrial setting, repeatable manufacturing, repeatable assembly, these
are the jobs that are going to evaporate, but there'll be replaced with a huge set of new opportunities,
where we need people that understand how these technologies work, and they can maximize their
efficiency in their applications. So that's a very interesting question and it's one that we're asked very
frequently. In my area, we're often asked, will robots take over? Are they going to take away jobs
from a large number of people? I think our answer is or at least my answer at the moment is, no.
Robots are very likely in fact to generate new jobs. So in the areas of robot maintenance,
manufacturing, design, all new positions I think will become available. So I think the fear of robots
taking away a huge number of jobs is perhaps somewhat unfounded at this stage. What do we do
with those people today in which the jobs are vulnerable to AI? Well, I think there's also some
responsibility for lots of especially large corporations that are heavily investing AI to help come up
with some sort of solution that can help retrain these people who are perfectly capable of learning
new skills as well, of that I'm confident, and help them develop new skills to work in other jobs. I think
that's something that is required for large corporations part of their responsibility to build to help
reduce the negative impact of AI. So yeah, there's a lot of growing concern, and I think it's both a
responsibility of corporations to be able to accommodate the changes in the industry. It's also the
workers who have to be cognizant that some jobs may be prone to be automated, and to never stop
learning, to always try to look towards new skill.

Employment and AI

A major ethical consideration for developers of AI systems is their effect on jobs. Some AI experts
including Nick Bostrom believe that there will be a significant economic impact of AI. Many people are
already concerned about the impact of AI on existing jobs and worry that it will change the entire
employment landscape. Let's look at how AI can affect jobs, where jobs may be at risk, and where
jobs are being created. AI can be used most effectively for jobs that are primarily repetitive or
rules-based. This category of job could include; bank tellers, sales people, call center operatives,
and drivers. People in these jobs may see demand for their skills fall, and may need to retrain in
other areas. For example, there have been tremendous advances in autonomous vehicle technology
in the last few years and as self-driving car has become more and more of a reality, the jobs of many
millions of drivers around the world are at risk, some of whom have found new sources of income in
the recent years by driving through the likes of; Uber, Lyft, Ola and DD. People converse daily with AI
Assistant bots. Bots answer calls at call centers, answer questions on websites, perform tasks on
mobile phones, and many other things besides. Bots can be used to perform simple tasks for many
workers, freeing those people up to attend to more complex tasks or replacing them entirely,
depending on the implementation. Alternatively, bots can be integrated into an organization's existing
systems providing help and support to people, and making their jobs easier. But along with improving
communications and reducing support costs for organizations, they can also eliminate many call
center and help desk jobs. There is great concern among economists that developing countries are at
even greater risk from AI encroaching into the job market with low-wage jobs quickly being lost.
Although this may be broadly true, there are some ways that a developing country with many people
looking for work can participate in the development of AI, and gain some benefits from the
technology. For example, workers can prepare images that will become training data for self-driving
cars by identifying parts of the images. This work does not require high tech infrastructure or highly
trained workers, but it can be used by some of the most sophisticated AI systems available today.
According to a report by the World Economic Forum, it is expected that 75 million jobs will be
displaced by AI, robotics, and automation in the next three years. However, AI also brings with it new
opportunities and job roles that have not existed in the past. In fact, the very same report also
indicates that AI will foster net new job growth, resulting in a 133 million new jobs in the next few
years. That's almost 60 million more jobs than will be lost. So with this changing job landscape, there
will undoubtedly be a shift in society as people cease to work in certain professions but need to find
other fulfilling sources of income. A key part of this is up-skilling and re-skilling initiatives. Companies
like IBM are working with policymakers, academic institutions, and training providers to help students
and workers acquire new skills that will emerge in an AI-enabled economy. As AI permeates our daily
lives, new ways of working will be discovered. Already AI is augmenting jobs rather than
eliminating them. Bringing new jobs to less developed countries, and bringing investors and small
business together regardless of location. Although the employment landscape will certainly change, it
may well be for the better.
AI working for good

In the BBC’s Machine Minds feature, Richard Gray writes about the many benefits of AI in today’s world. The article
is summarized below, and you can read the whole article here: Treating cancer, stopping violence… How AI
protects us.

There are many applications of AI that are beneficial to society, helping to protect us from disease, from crime, from
hunger, and from ourselves.

In the health field, AI systems are making impacts in controlling the spread of infectious diseases like Dengue fever,
yellow fever, and Zika, by predicting outbreaks. The Artificial Intelligence in Medical Epidemiology (Aime) system
uses over 270 variables to predict the next Dengue fever outbreak, and has an 88% accuracy rate up to three
months in advance. (Aime)

Early detection is crucial in the successful treatment of many cancers, sight loss, and other health problems. AI is
having an impact here too. IBM Watson systems are being trained to identify tumors and help diagnose breast,
lung, prostrate, and other cancers. (IBM)

Google DeepMind is working with the National Health Service (NHS) in the UK to train AI systems to interpret eye
scans. (DeepMind, Vox, Forbes)

Violent crime is a seemingly insoluble issue, but again, AI is having an impact in two major areas: gun violence and
knife crime. In the US, the Shotspotter system is being used to detect the sound of gunshots and alert authorities
quickly. (Shotspotter)

In the UK, violent knife crime is a rapidly growing problem. Police Forces across the UK are exploring the use of an
AI system called National Data Analytics Solution (NDAS). This system focuses on identifying people already
known to the police who may be more likely to commit knife crime. The intention is to prioritize getting appropriate
help and support for those people, but some people are interpreting this as predicting a crime before it happens,
making the plan very contentious. (PublicTechnology.net)
In agriculture, keeping crops healthy and free from disease is a never-ending challenge. In areas at risk of famine,
growers must be able to accurately identify multiple crop diseases with similar appearances and different
treatments. In Uganda, the Mcrops project combines the use of photographs taken on cheap smartphones and
computer vision to help farmers keep their crops healthy. (Mcrops)

Maximizing our efficient use of energy is critical to reducing the cost and impact of generating power. AI systems
are being used here too, for managing increasingly complex electricity grids, locating damaged cables, and even
helping to reduce the demand that devices make. (The conversation)

Principles for ethical AI

At the World Economic Forum in 2017, IBM CEO Ginni Rometty spoke about the three guiding principles that IBM
follows to ensure that the AI and cognitive systems it develops are ethical, trustworthy, and socially responsible.

Purpose
AI systems developed by IBM are designed to augment human intelligence, not to replace it. IBM refers to these
systems as cognitive, rather than AI. Cognitive systems will not become conscious, or gain independent
agency, but will always remain under the control of humans. Cognitive systems will be embedded in systems
used to enhance human capabilities. Cognitive systems must be built with people in the industry.

"We say cognitive, not AI, because we are augmenting intelligence," Rometty said. "For most of our businesses and
companies, it will not be man or machine... it will be a symbiotic relationship. Our purpose is to augment and really
be in service of what humans do."

Transparency
Cognitive systems must be transparent to be fully accepted as a normal part of people’s everyday life.
Transparency is required to gain public trust and confidence in AI judgments and decisions, so that cognitive
systems can be used to their full potential.

For IBM, this has three parts:

 People must be aware when they come into contact with AI and for what purposes it is used.
 People must be aware of the major sources of data in use.
 IBM clients always own their own business models, intellectual property, and data. Cognitive systems
augment the client’s years of industry experience and domain specific knowledge.

IBM recognizes that AI systems must be built with people in the industry. "These systems will be most effective
when trained with domain knowledge in an industry context," Rometty said.
Skills
There are two sides to the AI story; the cognitive systems, and the humans who use them. The  human side of the
story must also be supported.

IBM will work to help people gain the knowledge needed to engage with AI systems safely and securely. IBM will
also  help people develop the skills necessary to perform the new kinds of jobs that AI creates (much like this
course you’re taking now).

Lesson Summary

In this lesson, you have learned:

 AI raises ethical concerns because of the risks of bias and lack of transparency.
 AI professionals can mitigate against those risks.
 Mitigation strategies include providing effective training data, performing regular audits to ensure that AI
systems are performing as expected, and adapting algorithms to adjust for bias.
 Many jobs will be lost to AI, and that the most vulnerable jobs will be those with rules-based, repeatable
tasks, like call center workers and drivers.
 AI will generate new jobs, and new types of work.
 AI is already being used to benefit humankind in many fields, including healthcare, crime prevention,
agriculture, and power generation, among others.
 Gaining public trust is key to allowing AI to reach its full potential.
 IBM follows three principles to develop ethical AI – Purpose, Transparency, and Skills.

To learn more about ethical AI, visit:

 BBC Future - Machine Minds

 IBM Think Blog - Transparency and Trust in the Cognitive Era
 IBM Think 2019 - Trust and Ethics in Technology

The evolution and future of AI

The original AI researchers were very interested in games because they were extremely complex.
Huge numbers of possible positions and gains were available, yet they're simple in a certain way.
They're simple in that the moves are well-defined, the goals are well-defined. So you don't have to
solve everything all at once. With chess in particular, in the work on Deep Blue at IBM, what became
apparent, what computers could do on our problem like that was bringing a massive amount of
compute resource to do deeper searches, to investigate more options of moves in chess than was
previously possible. Watson defeating jeopardy. So this was another crossover point, in the
development of AI and cognitive computing. That the questions that IBM was able to answer with
jeopardy were questions that weren't simply looking up in the database, and finding the answer
somewhere. Rather it required information retrieval over lots of different information resources. Then
the combining of these together using machine learning that could arrive at answers that went
beyond what was simply written somewhere. Now, our technology is so much better and so much
more advanced that we're really ready to move on and to tackle much more challenging problems
that have this ill-defined or messy nature. Every industry from oil and gas, to healthcare, to media
and entertainment, to retail are just being swamped by a tsunami of unstructured data. That can be
multimedia, can be images, it can be video, it can be text. It's really the ability to understand that data
that is becoming critical. One of the most valuable applications of cognitive computing is in the health
domain. Medical providers, physicians, nurses, assistants face enormous challenges, leveraging all
of the available information that's out there. The medical literature increases by about 700,000
articles every year. There's already millions of journal articles out there, and today's imaging
technologies produce very rich amount of information. In fact, a particular scan might have 5,000
images in it. You combine the image analysis with natural language understanding, and text analysis,
leveraging the medical literature, leveraging the patient's medical history, the physician has got a lot
more information and knowledge at their disposal to help them make the best diagnosis possible.
Clearly, there is intersection of what the computer can do and what people are able to do. That gives
you something that's better than each of them individually. What is going to be truly interesting is to
see what is the best way for them to have really symbiotic type of interaction, taking advantage of
each other's strengths to collectively solve a problem? Watson, it looks at another aspect of
intelligence, and a much more difficult aspect of intelligence, that is language. You have to be able to
interpret the questions and come up with the right answers no matter what the topic. So I think the
ideal scenario for AI in the modern world is not to try and develop a system that completely
autonomously handled every aspect of a problem, but have a collaboration between machines
doing what they do best and people doing what they do best. I can imagine that combination will
do better than either one by themselves. We're constantly here looking for what's the next grand
challenge problem we can take on? That's not just around the corner or a year away, but it's going to
take a multi-year effort. When we get there, we'll have something that's valuable for the world.

What’s next for AI?

In Q&A sessions, IBM gathered the opinions of AI visionaries to consider what the future might hold for AI.

Click each name to discover more, and then return to the course to continue.

Kevin Kelly, Tech Author and co-founder of Wired

That's a good question. People expect AI to be human-like. I've been stressing

the fact that there are different types of thinking and the chief benefit of AI is that
it does not think like humans. As an example: Birds flap their wings to fly, but to
make humans fly, we had to invent a different type of flying—one that did not
occur in nature. And so, similarly, through AI, we’re going to invent many new
types of thinking that don't exist biologically and that are not like human thinking.
Therefore, this intelligence does not replace human thinking, but augments it.
Kevin is optimistic about the effect of AI on jobs, and believes that a key role of AI will be to transform big data into
something businesses can use.

Mark Sagar, Oscar-winning AI engineer

Mark combines his knowledge of computer graphics, human physiology and artificial intelligence to develop
emotionally responsive avatars for businesses.

Chieko Asakawa, Accessibility research pioneer & IBM Fellow

One of the world’s premier accessibility researchers, Chieko is excited about the progress of Ai in helping blind
people experience and explore the world.

Yoshua Bengio, Preeminent deep learning researcher

Describe as a founding father of deep learning, Yoshua sees unsupervised learning and natural language
processing as being the areas in which AI will develop very quickly.

Margaret Boden, Veteran AI & cognitive scientist

Margaret is interested in how AI will help us understand human creativity, and champions a multidisciplinary
approach to AI.

Future with AI

What does AI have in store for us in the future? >> My crystal ball is a little cloudy, so I don't know if I
have a prediction that I would bet money on. What I do know is, what we have seen is the way AI has
progressed, it starts fairly slowly, but then it gains steam exponentially. One good example is,
DeepMind put the system together that won in the game of Go. It's a 2,500-year-old game, which
won against a human opponent. But what's most amazing is while the first system was able to out do
2,500 years of human history in the game, the second generation of that system was able to outplay
the first generation of the system in less than one year. And it only took about 40 hours of training for
it to be able to achieve that level of proficiency in that game. And it won 100 games out of 100. So
what we do know is the pace of which this technology is accelerating is just breathtaking. And it's not
something we can predict very well. So if there was one prediction I were to make it's, it's going to get
faster, it's going to get better, it's going to get cheaper, and it is going to happen very, very rapidly
within a very short period of time. >> It's going to be an interesting world. It's going to evolve very
rapidly. Because these technologies not only can perform tasks that we've never seen automated
before, but learn as they do them and continue to improve. So what we'll see is we'll deploy systems
and every year they'll get incrementally better at the tasks they're trying to do. So we won't have to
drive our own cars anymore. Hopefully, we won't have to put away our own dishes. There'll be a lot of
simple things in our lives that become automated and become eliminated from our daily task list. It's a
revolution that we've been through before. The first washing machines, the first dishwashers, etc. All
these things that helped us, enriched our lives and simplified the way we live and increased our
comfort. And I think we'll see many more such evolutions as we watch the AI world unfold. >>
Particularly in healthcare, I think we'll see faster recovery times, we'll see better patient outcomes,
we'll see people spending less money and time in hospitals and in various care centers. AI has been
game changing for healthcare. There's so much information that we can take out of a particular
system and apply it to another. >> We can build all sorts of models that can be hugely beneficial to
long-term care. So that's something that I'm very excited about and seeing that evolve. >> Looking
into the crystal ball, I'm always apprehensive to make long-term predictions, but my hope is that we
are able to indeed deploy these collaborative robotic systems, including self-driving cars, to make
people's live better. That's the idea, is to use AI and robotics technology to improve the quality of
life for the whole spectrum of society. So I would be looking forward to a future in which AI plays a
central role in freeing us from the dull, the mundane, the dirty, the dangerous jobs, and hopefully
providing us with more time to spend with our families. Better health, potentially, better healthcare
through data analysis. And so I think if envision the future that I hope for will come to pass in a
number of years, we'll be really leveraging these technologies to make our lives better and to free
ourselves from dull, dirty, dangerous work.

What will our society look like when AI is everywhere?

Writing for the Smithsonian magazine, Stephan Talty and Jules Julien paint a fiction picture of how life might be in a
society where AI is almost everywhere. The full article can be read here: What Will Our Society Look Like When
Artificial Intelligence Is Everywhere?

AI has come a long way since the seminal “Dartmouth workshop” at Dartmouth college in June 1956. A new
discipline was being discussed, so new that “People didn’t agree on what it was, how to do it or even what to call it,”
said Grace Solomonoff, widow of one of the workshop attendees. In recent decades, AI has developed to be
present in our everyday lives, to an extent that has caused some to wonder: how far can this go, and what will
happen then?

Read the article and then return to continue the course.

Lesson Summary

In this lesson, you have learned:

 AI experts have varied views of the long-term future of AI.

 Understanding and generating natural language is likely to be the next big growth area for AI, along with
vision systems for the blind.

Optional:

If you want to learn more about the future of AI, watch:

 IBM Think 2019 - Chairman's Address: Building Cognitive Enterprises

 IBM Think 2019 – The Future of Infrastructure
The AI Ladder - The Journey for Adopting AI Successfully

In a world of dramatic digital transformation, companies are looking to AI to really help them shape
the future of work. AI can predict and inform future outcomes. It enables people to do higher value
work and businesses to imagine new models. It can automate decisions, processes, experiences but
AI is not magic. The truth is there is no AI without IA or information architecture, but many
organizations can't start because 80% of their data is locked in silos and not business-ready. So, how
do you turn your aspirations into outcomes? Through a prescriptive set of steps we call the ladder AI
and it starts with modernizing all of your data on a single platform that runs on any cloud. Then on the
ladder itself, there are four steps

Play video starting at 1 minute 8 seconds and follow transcript1:08

First, collect the data to make it simple and accessible

Play video starting at 1 minute 13 seconds and follow transcript1:13

Really think about the models you need to train Second, organize your data to create a business-
ready analytics foundation for those AI models.

Play video starting at 1 minute 24 seconds and follow transcript1:24

Third, analyze your data both for trust and transparency Because there's no use in applying and
scaling AI if you can't explain the outcome, detect bias or prove its accuracy

Play video starting at 1 minute 38 seconds and follow transcript1:38

Fourth, once you can really trust your data and the AI that you deploy Then you can realize its full
value Inside of the apps and processes that control your everyday work. In other words, The last step
is infuse or you begin to operationalize AI throughout your entire business We help thousands of
enterprises put AI to work by unlocking the value of their data and this AI and multi-cloud world By
delivering the right set of skills for their people and by building trust and transparency into AI. That's
the ladder to AI in a nutshell. Let's start climbing.

Advice for a career in AI

What advice would you give to someone who wants to learn AI, or perhaps even get into a career in
AI? >> So I believe artificial intelligence is going to permeate every sphere of human endeavor. So
the first thing I would say is I would congratulate the person who've chosen to study and apply AI as
having made the right choice. The second part I would say is because the technology is moving so
fast, because we have not yet discovered what we're going to discover, I think it's really important to
keep an open mind. It's important to keep an open mind and not get too attached to any particular
technology, any particular technique, any particular implementation. But to really think forward and
think outside of the box, and think about what may, think about the art of the possible, think about
what may come. And the number one advice I would give is apply what you learn. Don't make it
academic, make it practice. >> The first thing I would say is take a look at a great Coursera course,
like this one, and grasp the basics of AI. Get a fundamental lay of the land and understand what is
involved in various types of AI systems. And perhaps find the niche that you're most interested in.
And then, of course, look at academic programs that can help you to learn more about that specific
area. And for students who are coming up through high school, for example, I would say definitely
spend time focusing on math and science. Certainly, all disciplines are valuable, but your knowledge
of mathematics and science will certainly pay off when you're working on AI systems. Because there
are tons of opportunities and you'll need mathematics to really fully understand how the AI systems
operate. So take your mathematics and your science seriously. And then look at Coursera and then
beyond to other educational programs that can give you the basis you need to really jump into the
industry. >> The field of AI has changed a lot in the last five years. No longer require a PhD in some
kind of advanced mathematics or know some obscure programming language. You just have to know
how to use the software APIs and understand the problem. For example, you can use Watson, and
you just have to understand the problem and understand how the API works. But you still require
some advance knowledge if you would like to build your own algorithm. >> So artificial intelligence
machine learning, this is a very important field. It is the future of technology. Because it essentially
opens up this whole new world of interaction with computers. A world that so far, we barely even
knew existed. And it enables us to interact with computers in an implicit way. But what I would say is
that while machine learning technology is important, it's not something that you can learn in isolation.
Machine learning technology is not its own standalone subject. It's another algorithm in the toolbox of
a plethora of algorithms that programmers will use in their applications, albeit a much more intelligent
or a much more powerful algorithm than most. But still, it's another algorithm. So before you learn
machine learning technology, it's very important to understand the actual programming and the
technology that goes behind machine learning technology. Specifically because machine learning
actually requires this kind of next generation of programming, hardware acceleration, so much more,
that regular programming and regular algorithms don't necessarily require. So I'd really recommend
making sure that, first of all, you are passionate about technology itself. If you are, continue, learn
about programming, learn about coding, learn how to actually speak the computer's language. Learn
about the computational thinking behind code. And then go ahead and learn about machine learning
from the very, very basics. Start off with an API like IBM Watson to help you get an idea of what it's
capable of. Then move on to more advanced, custom techniques and the math behind them. >> So
AI is a fascinating field. But it's built on a huge number of foundational domains or foundational fields
of study. So you really need to know your mathematics, you really need to know your probability
statistics, your optimization. And you have to be able to program. You have to be able to take
advantage of the tools that are out there to train these networks and understand how they work. So
AI is really a broad field that requires a lot of specialists and a lot of specialization in a lot of areas. So
one of the best things you can do is get started quickly. Start playing with some simple tasks. Try to
identify digits or try to find cats on Internet pictures, right, things like this. These are wonderful
challenges that can get you going in understanding what you need to learn about that field. >>
Getting into AI now is like getting into anything to do with Internet, 20, 30 years ago. It's kind of the
Renascence of software right now. It's possible now, it wasn't possible ten years ago. We have the
computing technology. We have the computing power. We have the knowledge. And this is a field
that will only grow. So I think you need to get in yesterday

Hotbeds of AI Innovation
My latest poll of top cities for AI, which is primarily myself just looking at what's coming through my
own inbox in terms of opportunities, it's fairly widespread. Actually there are a lot of opportunities in a
lot of different places. So there's of course the San Francisco Bay area, which continues to be a
hotbed for startups in AI and robotics, and I think will continue to develop, there's Boston in
the United States which also has a large number of startups, there's Seattle. Then here in
Canada, we've been doing a great job in both Toronto and Montreal, of actually developing quite a
robust and thriving AI ecosystem that I think is going to also continue to develop and grow. So I think
we're quite excited about that. Then farther afield there are a number of locations in Europe, Asia,
some opportunities in China certainly, China and India as well, absolutely. So if you're located
somewhere next to a major city center, I think in any of these areas, you're well-positioned to to
launch your AI career. Toronto is exploding. It's been amazing to watch and be a part of. We have
Geoffrey Hinton obviously one of the three pillars of the Turing Award for major contributions in deep
learning, and the whole Vector Institute and ecosystem has come up around the wonderful progress
that came out of the University of Toronto over the last few decades. So what we're seeing is
wonderful incubators and startups in the area, investing heavily in AI technology and capabilities,
we're seeing large companies bringing in research labs and putting those right next to the university,
and we're seeing a wonderful amount of funding coming in for research specifically in AI in Toronto.
So I think we're competitive with any of the best places to learn and work on AI, in the world. I think
Canada is actually doing very well. We've done an excellent job overall in terms of centralizing some
really fantastic AI talents. So in Toronto and Montreal in particular and also in Vancouver, but
Vancouver, Edmonton, a number of different cities in Canada, have really become very strong AI
centers. They continue to grow and we're attracting a lot of great AI talents, and I believe that will just
continue. So Canada is shaping up to be a great place to come and do AI research and work in AI.

Tanmay’s Advice to Learn AI

So can you talk about advice for people who want to get into AI, especially the teenagers like you or
other students were fascinated by technology and AI. So really what I believe, first of all, is I mean
I've seen a lot of misconceptions about AI really wherever I go. Whenever people ask me, "How can I
get into the field of machine learning or AI?" There are lots of misconceptions, they feel like machine
learning has its own independent, isolated subject, and if I learn it, I can go ahead and implement
machine learning algorithms with these things. But really if you think about it, machine learning is just
another algorithm in the toolbox of algorithms when it comes to programming. Albeit in some cases it
can be much more powerful than other algorithms, and then on a very fundamental level, it is another
algorithm. So before you get into the field of machine learning, before you get into the field of AI, it's
very important that you understand how to code. Since machine learning is relatively complex
technology, having a very advanced knowledge of how computers work; how exactly coding works.
Then even sometimes a back-ends behind compilers, really does help you on your journey of
learning how to code machine learning algorithms. The reason I say this is because machine learning
isn't just a regular algorithm that you implement, its not just path finding or search or something of
that sort. It's special because it requires intense hardware acceleration. It requires you to understand
at least a little bit of the calculus that goes behind backpropagation. All of this is necessary to
understand the fundamental workings of machine learning and AI. So really to summarize what I'd
recommend is first of all, you should be passionate about technology itself. If you're not passionate
about technology, machine learning technology, it really isn't something that you should go and work
towards. Then from there, once you know that you're passionate about it, you want to do it. Then go
ahead and start off by learning concepts in the field of technology. Learn how to code in languages
like Python, Julia as a new language, SWIFT because of the new SWIFT for TensorFlow project.
Then from there, go ahead and learn a little bit at least of the actual math behind neural networks. I
mean, I remember when I stumbled upon Watson, that was the first time I ever heard of machine
learning. When I went from Watson to custom neural networks, the first thing I did is I actually drew a
small neural network on paper, and in the backpropagation, manually to understand how exactly
weights are updated, how a loss values work, all these sorts of things. Once you have that
fundamental understanding, then you go ahead to implement those neural networks from scratch and
your languages. Then once you have a good idea of how it works, once you practice. Once you learn
by example, then you're finally ready to go ahead and use different libraries, use different toolkits to
enable you to rapidly prototype and build applications in the field of machine learning. So that's a very
very high level view point, that I'd say to get into machine learning, but of course, there are better
ways and more intuitive ways to get into machine learning as well. For example, Watson, if you take
a look at how it provides its APIs on the cloud, you don't need to understand any of the workings
behind machine learning to use Watson, and at the same time, you can still leverage all those
powerful capabilities, the language translator. Lets you use practically state of the art, neural machine
translation techniques, and even train your own models without having to understand a single clue
behind what goes behind the models. So being able to start off with a toolkit like Watson, getting a
good idea of both programming and machine learning at once, what machine learning is and isn't
capable of. Then getting to the custom math, is another great way.

Polong’s Advice for a Job in AI

So what advice would you give to someone who's looking to get into a career in AI. Yes absolutely. I
had a bit of an unusual route becoming a data scientist or an AI specialists myself. I have a
background in psychology so when I was studying, I was basically thinking of continuing to do
academic research and not going to industry. But at some point in time, I had a realization that I didn't
really want to become a professor. I didn't really want to teach or do research in an academic setting.
I started looking for jobs and it was quite difficult because what we see today is sometimes what you'll
learn in university doesn't translate directly into what companies want. So how do you bridge that gap
exactly? For me, when I was at the time still in Tokyo, Japan, I didn't have a lot of resources around
me or English speaking resources at the time where I could develop my skills. So I looked towards
online courses and online courses helped me practice python, practice R, my data analysis skills
through programming and I was able to develop a foundation in the skill set that was highly sought
after by many different companies. Then from there, the question became, okay, now I have these
skills, these newly developed skills through these courses that I've taken online, how do I translate
that to something that I can show and demonstrate to potential employers? So then the next problem
I had to deal with was, well, how do I demonstrate that some way? So I created a portfolio of different
online assignment exercises projects to demonstrate that I could do the kinds of things that they were
hoping that I could do to show my competency. I also when I came back to Toronto, Canada, I
started doing meet ups to teach at these meet ups, teach R, teach Python, teach data science.
Again, further demonstrate my competencies so that people could really see that I could do what I
was claiming that I could do. Which was a little bit difficult because people would just look at my
resume and you see, oh, you have a background psychology, what does that have to do with data
science or AI? So it was hard in that sense that had to overcome that burden of proof.

Play video starting at 2 minutes 25 seconds and follow transcript2:25

The final piece of advice that I would give is connect with fellow local AI specialists. Whether they are
data scientists, specialists at particular field of AI, machine learning researchers, data engineers, just
to get a better sense of what kind of AI specialists you might be interested in becoming because
today there's so many different kinds of AI specialists. You could be very technical. You could be
doing research on the latest state-of-the-art algorithms or you could be doing something a bit more
business-oriented, like trying to optimize revenue, maybe some reporting as well. Trying to find ways
to personalize a particular service to each and every kind of user that you might have. There's
different aspects of AI and I think it's really important to have some clarity into what kind of role and
what kind of job you want. The first step to do that is to be able to reach out and ask questions to
people who are already in the field. It's a great way to get a sense and also maybe even form some
connections that might be meaningful in your career as well. Those are the four things. Number one
is, acquire skills wherever you can. Always be open to learning new things and new skills. Number
two is demonstrate your capabilities through projects, exercises. Something that you can share
online. Even blog posts and articles. Number three, if you can teach at meet ups or maybe even
more ideally conferences so you can demonstrate your competencies in public. Then number four is
going to be try to connect with local experts or local specialists who are already in the field so you
can get a better sense of what that career kind of road map might look like for you.

Lesson Summary

In this lesson, you have learned:

 How to approach a career in AI.

 Where there are centers of AI innovation.
 Steps you can take to learn AI.

IBM Watson Visual Recognition (VR) is a service that uses deep learning algorithms to identify objects and other
content in an image. In this hands-on lab, you will use Watson VR to upload and classify images.

Conclusion and Next Steps

You started this course in Module 1 by hearing different definitions of AI. You then explored different use cases and
applications of AI.

In Module 2, you got a basic understanding of how AI works, familiarized yourself with terminology like Machine
Learning, Supervised Learning, Unsupervised Learning, Training Data, Deep Learning, and Neural Networks. You
also got to know about some of the key fields of application of AI including Natural Language Processing, Speech,
Computer Vision, and Self Driving Cars.

In Module 3, you explored several issues and ethical considerations surrounding AI, such as Bias, impact on
employment, etc.

In Module 4, you heard about how AI may evolve and impact our future. You also heard from several experts about
learning and starting a career in AI. And finally you demonstrated AI in action by getting Watson to classify images.

This is the end of this Introduction to AI course, but certainly not the end of the journey for learning about AI.

This course is part of the following programs:

1. AI Foundations for Everyone Specialization

2. IBM Applied AI Professional Certificate

As a next step to learn more and develop your AI expertize, we encourage you to explore other courses in these
programs such as:

Getting Started with AI using IBM Watson, or

Building AI Powered Chatbots Without Programming

We hope you enjoyed taking this course.

Good luck!