Transcripts
1. Course Description: Hello there, Welcome to the beginning of
your first journey. In this course, we will
guide you through the past, present, and future of
artificial intelligence. We will also have a short glance at different fields
in which AI can be applied and show
some applications that might be
interesting to you. However, we will also go
through the definition of AI and define what a neural
network is step-by-step. Show how it works and explain
the differences between AI, machine learning
and deep learning. There will also be one short multiple choice
tests to see whether you've grasped the concept of neural networks
and AI in general. Without further ado, we wish you the best luck and
success on your journey. Stay motivated.
2. What is AI: Deep blue, which the
world champion in chess, Garry Kasparov, AlphaGo destroys components
in a game of Go. The first AI costs are coming. A new era for machine
learning has begun. If you have heard or read
such or similar things, you probably think, wow, I is the next big thing. But what exactly is
the next big thing? What exactly is
artificial intelligence? And why a term such as deep
learning, machine learning, or neural networks repeatedly associated with the
AI breakthroughs. To cross the first
obstacle on your journey, Let's get started by defining the different
areas in general. After we define each area, we will then continue to have a closer look at
the differences. The definitions. If we go by the definition of one
of the founding fathers of artificial intelligence
than it is the science and engineering of making
intelligent computer programs. It is related to the
similar task of using computers to understand
human intelligence. But a does not have to confine itself to methods that are
biologically observable. It's the science and
engineering of making intelligent computer programs,
machine-learning go. The other side is
an application of artificial intelligence
that includes algorithms that past data learn
from that data and then apply what they have learned to make
informed decisions. Then we have deep learning, which is defined as a subfield of machine learning data structures algorithms in layers to create an
artificial neural network that can learn and make
intelligent decisions on its own. Did you notice something? In all three definitions, the word intelligent
was mentioned. That's right, intelligence. But what exactly
is intelligence? How is it defined or measured? When exactly is someone or
something truly intelligent? Since it's a rather
often discussed topic, we defined intelligence
as for now as a general mental capability that involves the
ability to reason, solve problems,
think abstractly, comprehend complex ideas,
and learn from experiences. It reflects a broader
and deeper capability for comprehending
our surroundings. Catching on making
sense of things, or figuring out what to do. Now that we know the definitions
of AI, machine learning, deep learning, and intelligence, we can look at how they
differ from each other.
3. AI and the 3 faces: Hey, I enter three phases. Imagine the free areas as
three concentric rings with, I think the largest ring and
deep learning the smallest. Each area is simply a subset of the previous bigger area. S. A short overview, we can
say for now that I need to explicitly programmed and can
do only one task at a time. Machine learning systems have
the ability to learn and improve from experience without being
explicitly programmed. Deep learning or the other side, uses neural networks to analyze
different structures and patterns and therefore work
similar to the human brain. You will hear more about
that in a few minutes. Now that we are done
with the short overview, let's have a closer look at
each of the three areas. We will start our journey
with the field of AI. This field itself can again be differentiated into three
different thoughts. Artificial narrow intelligence,
also called weak. Weak, strong at
certain activities but cannot surpass
humans in general. Although these machines,
if you're clever, The only have a limited
range of capabilities. Which is why this kind of artificial intelligence is
referred to as weak AI. Narrow AI just replicate human behavior based on the limited set of
factors and actions. E.g. an AI program that to win games edges will most likely
fail to play the game of Go. Artificial general
intelligence, aka strong AI. At this point, A IS systems
are becoming more human-like. Such an AI system could make its own decisions without
human interaction, solve complex logical tasks that require abstract thinking, but also have emotions
at one point. However, when considering that the human brain is the model for creating such general
intelligence, it's not surprising
that achieving a strong AI is an
immense challenge. Artificial Superintelligence,
AKA super AI. If we ever arrive at this point than one
thing is for sure such a robot or
being would not only outperform humans
in multiple tasks, but it would instead
be ahead of humans in almost every thinkable area, such as intelligence, wisdom, social skills, creativity,
and many more. Well, if this causes some fear that machines will
overrun us one day, don't worry, we're still far from even reaching
the secondary iPhone. Currently no strong or
super AI is known to exist, and it will probably still
take decades to arrive there.
4. Machine Learning and the fundamentals: Machine-learning
and fundamentals. Diving deeper into
the next layer, we arrive at machine learning. Machine learning is a subset of AI and focuses on learning how to solve specific tasks without being
explicitly programmed. Instead of just executing a list of automatic
instructions, machine learning models
improved through experience and the
use of statistics. For this, they need three
components to work. Number one, datasets. Before applying machine
learning models to any task, they need to be trained
on a collection of samples, also called data-set. Usually, this is one of the most time-consuming
steps in machine learning, since most datasets require multiple thousands of samples, which takes a lot of time
and effort to create. One of the most well-known
datasets would be e.g. the iris flower dataset. This multivariate
dataset consists of three different
flower species, each consisting of 50 samples. Each sample has four
features which describe e.g. the petal length or
the petal width. Since this dataset is open to anyone and rather
easy to handle, it's often recommended to AI Guinness when starting
the first day I project. Number two, features,
usually features, uh, pieces of data that
describe the samples. Let's e.g. stay at the
Iris flower dataset. In this iris flower dataset, there are four features
which described the flowers. Petal length, petal width, sepal length, and sepal width. Depending on your model
and the features, it can make a big
difference on how your model performance during
the training and testing. Have a look at the
following graphics. Here we plotted the correlation
map of the four features. This allows us to see what features are
correlated with each other and what features are best for separating the dataset. Good choice might be the
petal length and petal width. Why those two? You might ask? Let's have a closer look. In the graphics, we see the correlations with
three different colors. These three colors represent, in this case the flower species. Now our task is to look
at the map and decide which features best separate the dots in different colors. E.g. the first picture
in the second row does a pretty good job at separating the yellow dots from
the other ones, but it completely fails to separate the pink ones
from the purple ones. However, if we look at the
third picture in the last row, we can see that all three colors are almost perfectly separated. The features that were used were petal length
and petal width. If you're still curious
about the graphics, just pause the video
for a few seconds and have a look at the
other rows and columns. However, we will now move
on to the next point. Algorithms. An algorithm can be imagined as a list of instructions which will be executed step by step
to solve a specific task. However, in machine learning, It's often the case that multiple different
algorithms can be used in combination with statistical methods to solve the same task or to get
a better performance. One could also just combine multiple algorithms and play
around with the settings. Now that we know what components are needed for machine learning, Let's have a look at deep
learning in the next video.
5. Deep Learning _ Example-Use-Case: Deep learning and
neural networks. Do you remember how
important it was in the case of machine learning
to select good features? In the case of deep learning, it's not necessary anymore. Instead, the model collects
the features itself and the proofs with the help of so-called neural networks. Since deep learning was inspired by the
structure of our brains, deep learning algorithms use complex multi-layer
neural networks, abstract previously
unknown patterns in the data to come
to a solution. Still no clue what the
neural network is. Usually when explaining how
neural networks work exactly. It would involve
some mathematics. But since this is the
Introduction to AI course, we will explain it in
a rather simple way. Neural networks consist of
the following note layers. An input layer, one or
more hidden layers, and then output layer. Each node are also called
artificial neuron, connects to another one and has an associated weight
and threshold. If the output of
any individual node is above the specified
threshold value, that node is activated sending data to the next
layer of the network. Otherwise, no data is passed along to the next
layer of the network. Now to be able to train
a neural network, we would need data, a lot of data actually, only then can we truly improve the accuracy of
the model over time. But once these learning
algorithms are fine-tuned, They will allow us
to classify and cluster data in a
very short time. Now that we've gone
through everything, how about the small example? Example use case. Suppose you own a
small business that specializes in sorting fruits
into different categories. In the sorting plant, the
fruits are all mixed up. It is necessary to separate the fruits and package
them into caught, brought through traced before delivering them to supermarkets. Among the fruits that
need to be sorted out, bananas, apples, and oranges. Now that we know the task, let's go through each
of the three areas. Ai approach. In AI, you'd now have used an
AI based algorithm that makes use of decision logic
within a rule-based system. An example would be, if the object is an apple, then transport to the right. If the object is a banana, then transport to the left. However, the AI-based
system success is dependent on the fruit
being accurately labeled by the fruit pickers and having a scanning mechanism in place to tell the algorithm of
what the fruit is. A machine learning approach. Machine learning based algorithm is now proposed for improving the AI based approach to fruit sorting when labels
are not available. For machine learning to work, the description of what each fruit looks
like, it's needed. This is called
feature extraction. This is done by creating
a blueprint based on the features and attributes
unique to each fruit. The algorithm is trained
using features such as size, color, shape, and so on
to classify the fruits. Moving on to the next approach, we arrive at deep learning by removing the need to define
how each fruit looks like. A deep-learning based algorithm could be used to
solve any fruit. A major advantage of the deep
learning model is that it does not require features to classify the fruits correctly. With lots of fruit images, the model can build up a pattern of what each fruit looks like. These multiple layers of
neural networks will be used to process the images in
the deep learning model. Then each network layer will define specific
features of the images, like the shape of the fruits, the size, the color, and so on. However, for the model
to achieve good results, it will require significant
computational power and vast amounts of data. Now that you know somewhat
the differences between AI, machine learning
and deep learning, let's have a look at the
history of AI in the next part.
6. History of AI: History of AI from the
past to the present. Moving towards the first winter. Welcome to the history of AI. After hearing and reading many articles of
successes in the eye, many people might assume that it is a relatively new field, but this is not the case. It has a longer past
and you might think, Let's take a seat and
have a talk about the awesome history
and success stories. Today we hear a lot about new achievements
in the field of artificial intelligence,
automation and robotics. But it, you know
that the idea of intelligent machines already
existed in ancient times. Do you know the story of
Carlos, the bronchi giant? According to move,
Carlos is described as a giant bronze men created by the Greek god of invention
and blacksmithing. Zeus, the king of Greek gods, assigned him the
task of defending the island of Crete
from attackers. While we haven't created a giant robots or anything
like that in the recent past, we still have had a lot
of interesting things. Let's start with
Asimov's three laws. Asimov's laws were first
described by Isaac Asimov as the basic rules of robotic service and should be followed by any
type of robot. Asimov's rules are
stated as following. First, a robot shall
not knowingly injure a human being or through inaction allow a
human to be harmed. Second, a robot must
obey orders given to it. They are human unless such an order would conflict
with rule number one. For a robot must protect
its existence as long as that protection
does not conflict with rule number
one or number two. Moving forward in time, we meet Alan Turing with
the so-called Turing test. He tried to formulate in 1950 how one could
determine whether a computer or model could have the same ability to
think as humans. The test uses a
simple question and answer process between
a human questioner and to anonymous
respondents who are not visible to the
question of the free, non predetermined questions
are asked by people without any visual or
auditory contact with the interviewer
using an input tools such as a keyboard or a screen. If at the end of the
test, the human question, I cannot determine
from the questions which of the two
respondents is the machine. The intelligence of the machine can be defined as human-like. Only six years later, the famous Dartmouth
Conference took place. To Dartmouth Conference is
considered the birth of artificial intelligence as
an academic discipline. It was requested, planned, and carried out by John
McCarthy, Marvin Minsky, Latin in Rochester and
Claude Shannon under the full name dot MOV
summer research project on artificial intelligence. It took place in the
summer of 1956 from June 18th to August 16th at Dartmouth College
in New Hampshire. Topics such as automatic
computers, neural networks, abstraction or randomness and
creativity were discussed. And as it turned out,
after just a few years, practically all participants
in the conference had become internationally
renowned experts in the field of artificial
intelligence. Many other innovations followed
the Dartmouth Conference, such as the first
chat bot eliza, which was supposed to take over the task of psychotherapists. However, as promising
as these projects were, researchers finally concluded
that the real-world is just far too complex to be
processed there such models, which led to the cancellations
of important findings at the beginning of the first
AI winter in the 1960s. Preparing for the second winter. After the effects of the first
AI winter began to fade, a new age of EI began. This time, much
greater emphasis was placed on developing
commercial items. Furthermore, significant
conferences such as the Association for
the Advancement of Artificial Intelligence, began in the early 1880s and saw a tremendous
surge in ticket sales. Ai technology has
peaked a curiosity of both the general public and
government authorities. Expert systems were crucial to the commercialization of AI. This systems were
created by developing if ten rule sets and have been used in a variety
of applications. Including financial planning, medical diagnostics,
geological investigation, and microelectronic
secret is n. However, since the models and
techniques we're still very limited and could not solve
more complex problems. The second winter came
just a few years later. The present progress will
slow after the second winter, but major breakthroughs came
only a few years later. Among other things, it
was possible to defeat the den chess world champion Garry Kasparov with
the help of Deep Blue. Deep Blue was a supercomputer developed by IBM
specifically for playing chess and was best
known for being the first AI program to ever
win a chess match against the reigning world champion
after losing the first six k match against Garry Kasparov in 1996 and receiving
a massive upgrade, Deep Blue was able to beat the world champion in May 1997. A few years later, AlphaGo beat the world champion in the game of golf
with four to one. It might not sound like a big
milestone, but it truly is. Alphago differs greatly
from earlier AI projects. To calculate its
chances of winning, it used neural
networks rather than probability techniques that were hard coded by human programmers. In addition to
games that AlphaGo plays against itself
and other players, AlphaGo also excesses and analyzes the complete
Internet go library, including all games, players,
stats, and literature. One, setup, it examines
the optimal strategy to solve the game of golf without the assistance of the
development team. Alphago estimates enormous
amounts of probability for many moves in the future using neural networks and
Monte Carlo tree search, which you will learn more
about in another course. Now that we're at the
end of the history, it's time to go
back to the future.
7. Different AI-Fields: Future applications. There are so many
theories about what impact ai will have
on us in the future. And since there are so
many possibilities, Let's just have a look at three examples that
consume become a reality. Number one, fully smart
and autonomous cities. The concept of fully smart
and autonomous cities is an exciting possibility
for the future of AI. With the advancements
in technology, we can see homes and apartments becoming smarter with
voice recognition systems, fingerprint sensors, and more. If this trend continues, we could soon see entire cities becoming fully autonomous. In these cities, everything
from garbage disposal to public transport could be operated without
human intervention. Just imagine wasting
disposal trucks driving themselves to
designated areas for collection or public transport systems that automatically reroute based on traffic and passenger demand. One of the potential benefits of autonomous cities
is the reduction in traffic pollution and
accidents caused by human error. This could lead to a cleaner and safer environment for residents. Additionally,
autonomous cities could also reduce the cost of public services and
enhance the efficiency. Number two, a, i, discovering new
technologies and laws of physics. That's right. It has already been possible to predict some physical
processes on a small scale with the help of AI or even to create new
mathematical theories. Scientists from Osaka
University and COBie e.g. have succeeded in extracting Hamiltonian equations
using neural networks. That's a short info.
Hamiltonian mechanics is based on Lagrangian and
Newtonian mechanics. Without going into
too much detail. In physics, Hamiltonian
mechanics is the theory of how energy changes from kinetic energy to
potential energy. And tech again over time, it's used to describe systems like a pendulum or
a bouncing ball. However, its strength
is demonstrated in more complex systems like celestial mechanics
or planetary orbits. Number three, ai
in law and order. I'm sure you have
often heard that the legal system
is struggling with too many tasks to help
the legal system out. And AI created through cooperation with lawyers,
judges, developers, and other groups of
people can be used in smaller court cases
such as damage claims. It can also save valuable time when structuring
and preparing files. Nevertheless, there are also moral and ethical
questions in this regard. However, since we don't have the time to go through
those questions, we will learn more about
them in another course.
8. Future Applications: Different AI fields
and overview. Now that we have gone through the history of
artificial intelligence, let's have a look at how it's
used in different fields. Since I is a very
complex and broad field, it is difficult to keep
an overview or even impossible to list all
areas that make it up. To help you out,
you will first get an overview of the
most important areas. Machine learning,
knowledge representation, planning, neural
networks, or e.g. robotics, computer vision, NLP, searching, and many more are all important
sub periods of AI. One crucial sub area of AI
is knowledge representation, which involves
representing information about the world in a format that the
computer system can use to perform complex tasks, such as diagnosing
medical conditions or engaging in natural
language conversations. Nlp, on the other hand, enables computers to understand and interpret human language. While computer
vision is vital in enabling machines to
perceive the environment. Each sub area is essential and plays a unique role in the
development of the eye. While it is impossible to cover all these exciting sub
arrows in this video, let's focus on some
examples of how AI is currently being used in various industries
and applications. Examples, searching exoplanets. Did you know that in
the last decade alone, over 1 million stars
have been observed to find out if they are
home to exoplanets. In short, exoplanets are
planets that orbit other stars. So far, the search has
been largely manual, but through the use of AI and
especially deep learning, the process can be
automated and quantified. Just imagine, instead
of 100 planets a year, you suddenly find
thousands of new planets. In this context, a group of astronomers from the
University of Geneva, burn and NCC our
planet, Switzerland, teamed up with a company
called this high-tech to use artificial intelligence for identifying planets in pictures. They wanted to find exoplanets that were previously
undetectable. So they trained a
computer program to predict how planets
interact with each other. By using this new technique, the scientists were able to improve the search
for exoplanets and make discoveries that they wouldn't have been
able to find other ways. Ai in drug discovery. Various pharmaceutical companies
such as fire, Moderna, and others are already using AI significantly shorten the
research process for new drugs. The best example of this
is the development of the COVID vaccine by the
pharmaceutical company Moderna. With the help of data from
the sars COVID virus, a predecessor of
the coronavirus, and two combination with AI
at especially deep-learning, the company has
managed to provide the vaccine in a
very short time. However, AI is not only used in a search for the
right vaccine composition, but also in part to
create drugs and test them for side-effects
in simulations. Which not only saves
time and money, but also reduces the number
of animal experiments. Creates art. That's right. Pi creates images, videos, backgrounds,
and artwork. With new emerging AI players such as stable diffusion, dolly, or medullary, the
creation of images, videos or odd is easier
than ever before. Just have a look at
this short video here about the Assembly. President Trump is a total
and complete the picture. Now. You see, I would
never say these things, at least not in a
public address, but that was pretty scary. How about creating
fake faces instead? Although AI isn't
perfect edits right now, imagine how it will be in
the upcoming 15 to 20 years. There is also the
possibility of combining two images to create a
completely new work. E.g. let's just take the
picture of Mona Lisa, but let's try it this
time in fact hostile. Or instead, how about a combination of the screen
and the picture of Obama? With a better understanding of the current applications of AI, we can now turn our attention to the exciting possibilities for the future of this technology. Let's explore some of the potential
future applications of AI in the next chapter.
9. What we have learned so far: What we have learned so far. Arriving at the end, Let's just rethink what
we have learned so far. We look together at the terms
of differences between AI, machine learning
and deep learning. Then we were able to catch a glimpse of the past of EI and surprisingly found that AI is an older research area
than previously thought. We have heard about azimuths, laws and the Turing test. Back to the future, we
learned which areas EI consists of and where it
is already used today. In the last chapter, we
were able to speculate as to how AI could develop
as it currently stands. Now that you have
a solid foundation of knowledge on the eye, you're all set to dive into the rest of the
causes with ease. So keep up the great
work and stay motivated.
Astra Learning, Learn AI with ease!
