What do you get with eBook?

Instant access to your Digital eBook purchase

Download this book in EPUB and PDF formats

Access this title in our online reader with advanced features

DRM FREE - Read whenever, wherever and however you want

Artificial Intelligence and Machine Learning Fundamentals

Chapter 1. Principles of Artificial Intelligence

Note

Learning Objectives

By the end of this lesson, you will be able to:

Describe the various fields of AI
Explain the main learning models used in AI
Explain why Python is a popular language for AI projects
Model the state space in AI for a given game

Note

In this lesson, you will learn about the purpose, fields, and applications of AI, coupled with a short summary of the features we'll use in Python.

Introduction

Before discussing different AI techniques and algorithms, we will look at the fundamentals of artificial intelligence and machine learning and go over a few basic definitions. Then, using engaging examples, we will move forward in the course. Real-world examples will be used to present the basic concepts of artificial intelligence in an easy-to-digest way.

If you want to be an expert at something, you need to be very good at the fundamentals. So, let's begin by understanding what artificial intelligence is:

Definition: Artificial Intelligence (AI) is a science that's used to construct intelligence using hardware and software solutions.

It is inspired by reverse engineering, for example, in the way that neurons work in the human brain. Our brain consists of small units called neurons, and networks of neurons called neural networks. Beyond neural networks, there are many other models in neuroscience that can be used to solve real-world problems in artificial intelligence.

Machine learning is a term that is often confused with artificial intelligence. It originates from the 1950s, and it was first defined by Arthur Lee Samuel in 1959.

Definition: Machine learning is a field of study concerned with giving computers the ability to learn without being explicitly programmed.

Tom Mitchell proposed a more mathematically precise definition of machine learning.

Definition: A computer program is said to learn from experience, E, with respect to a task, T, and a performance measure, P, if its performance on T, as measured by P, improves with experience E.

From these two definitions, we can conclude that machine learning is one way to achieve artificial intelligence. However, you can have artificial intelligence without machine learning. For instance, if you hardcode rules and decision trees, or you apply search techniques, you create an artificial intelligence agent, even though your approach has little to do with machine learning.

How does AI Solve Real World Problems?

Artificial intelligence automates human intelligence based on the way human brain processes information.

Whenever we solve a problem or interact with people, we go through a process. Whenever we limit the scope of a problem or interaction, this process can often be modeled and automated.

AI makes computers appear to think like humans.

Sometimes, it feels like AI knows what we need. Just think about the personalized coupons you receive after shopping online. By the end of this course, you will understand that to choose the most successful products, you need to be shown how to maximize your purchases – this is a relatively simple task. However, it is also so efficient, that we often think that computers "know" what we need.

AI is performed by computers that are executing low-level instructions.

Even though a solution may appear to be intelligent, we write code, just like with any other software solutions. Even if we are simulating neurons, simple machine code and computer hardware executes the "thinking" process.

Most AI applications have one primary objective. When we interact with an AI application, it seems human-like because it can restrict a problem domain to a primary objective. Therefore, we get a chance to break down complex processes and simulate intelligence with the help of low-level computer instructions.

AI may stimulate human senses and thinking processes for specialized fields.

We must simulate human senses and thoughts, and sometimes trick AI into believing that we are interacting with another human. In special cases, we can even enhance our own senses.

Similarly, when we interact with a chatbot, we expect the bot to understand us. We expect the chatbot or even a voice recognition system to provide a computer-human interface that fulfills our expectations. In order to meet these expectations, computers need to emulate the human thought processes.

Diversity of Disciplines

A self-driving car that couldn't sense that other cars were driving on the same highway would be incredibly dangerous. The AI agent needs to process and sense what is around it in order to drive the car. But that is itself is not enough. Without understanding the physics of moving objects, driving the car in a normal environment would be an almost impossible, not to mention deadly, task.

In order to create a usable AI solution, different disciplines are involved. For example:

Robotics: To move objects in space
Algorithm Theory: To construct efficient algorithms
Statistics: To derive useful results, predict the future, and analyze the past
Psychology: To model how the human brain works
Software Engineering: To create maintainable solutions that endure the test of time
Computer Science or Computer Programming: To implement our software solutions in practice
Mathematics: To perform complex mathematical operations
Control Theory: To create feed-forward and feedback systems
Information Theory: To represent, encode, decode, and compress information
Graph Theory: To model and optimize different points in space and to represent hierarchies
Physics: To model the real world
Computer Graphics and Image Processing to display and process images and movies

In this course, we will cover a few of these disciplines. Remember, focus is power, and we are now focusing on a high-level understanding of artificial intelligence.

Fields and Applications of Artificial Intelligence

Now that we know what Artificial Intelligence is, let's move on and investigate different fields in which AI is applied.

Simulation of Human Behavior

Humans have five basic senses simply divided into visual, auditory, kinesthetic, olfactory, and gustatory. However, for the purposes of understanding how to create intelligent machines, we can separate disciplines as follows:

Listening and speaking
Understanding language
Remembering things
Thinking
Seeing
Moving

A few of these are out of scope for us, because the purpose of this course is to understand the fundamentals. In order to move a robot arm, for instance, we would have to study complex university-level math to understand what's going on.

Listening and Speaking

Using speech recognition system, AI can collect the information. Using speech synthesis, it can turn internal data into understandable sounds. Speech recognition and speech synthesis techniques deal with the recognition and construction of sounds humans emit or that humans can understand.

Imagine you are on a trip to a country where you don't speak the local language. You can speak into the microphone of your phone, expect it to "understand" what you say, and then translate it into the other language. The same can happen in reverse with the locals speaking and AI translating the sounds into a language you understand. Speech recognition and speech synthesis make this possible.

Note

An example of speech synthesis is Google Translate. You can navigate to https://translate.google.com/ and make the translator speak words in a non-English language by clicking the loudspeaker button below the translated word.

Understanding Language

We can understand natural language by processing it. This field is called natural language processing, or NLP for short.

When it comes to natural language processing, we tend to learn languages based on statistical learning.

Remembering Things

We need to represent things we know about the world. This is where creating knowledge bases and hierarchical representations called ontologies comes into play. Ontologies categorize things and ideas in our world and contain relations between these categories.

Thinking

Our AI system has to be an expert in a certain domain by using an expert system. An expert system can be based on mathematical logic in a deterministic way, as well as in a fuzzy, non-deterministic way.

The knowledge base of an expert system is represented using different techniques. As the problem domain grows, we create hierarchical ontologies.

We can replicate this structure by modeling the network on the building blocks of the brain. These building blocks are called neurons, and the network itself is called a neural network.

There is another key term you need to connect to neural networks: deep learning. Deep learning is deep because it goes beyond pattern recognition and categorization. Learning is imprinted into the neural structure of the network. One special deep learning task, for instance, is object recognition using computer vision.

Seeing

We have to interact with the real world through our senses. We have only touched upon auditory senses so far, in regard to speech recognition and synthesis. What if we had to see things? Then, we would have to create computer vision techniques to learn about our environment. After all, recognizing faces is useful, and most humans are experts at that.

Computer vision depends on image processing. Although image processing is not directly an AI discipline, it is a required discipline for AI.

Moving

Moving and touching are natural to us humans, but they are very complex tasks for computers. Moving is handled by robotics. This is a very math-heavy topic.

Robotics is based on control theory, where you create a feedback loop and control the movement of your object based on the feedback gathered. Interestingly enough, control theory has applications in other fields that have absolutely nothing to do with moving objects in space. This is because the feedback loops required are similar to those modeled in economics.

Simulating Intelligence – The Turing Test

Alan Turing, the inventor of the Turing machine, an abstract concept that's used in algorithm theory, suggested a way to test intelligence. This test is referred to as the Turing test in AI literature.

Using a text interface, an interrogator chats to a human and a chatbot. The job of the chatbot is to mislead the interrogator to the extent that they cannot tell whether the computer is human or not.

What disciplines do we need to pass the Turing test?

First of all, we need to understand a spoken language to know what the interrogator is saying. We do this by using Natural Language Processing ( NLP ). We also have to respond.

We need to be an expert on things that the human mind tends to be interested in. We need to build an Expert System of humanity, involving the taxonomy of objects and abstract thoughts in our world, as well as historical events and even emotions.

Passing the Turing test is very hard. Current predictions suggest we won't be able to create a system good enough to pass the Turing test until the late 2020's. Pushing this even further, if this is not enough, we can advance to the Total Turing Test, which also includes movement and vision.

Description

Machine learning and neural networks are pillars on which you can build intelligent applications. Artificial Intelligence and Machine Learning Fundamentals begins by introducing you to Python and discussing AI search algorithms. You will cover in-depth mathematical topics, such as regression and classification, illustrated by Python examples. As you make your way through the book, you will progress to advanced AI techniques and concepts, and work on real-life datasets to form decision trees and clusters. You will be introduced to neural networks, a powerful tool based on Moore's law. By the end of this book, you will be confident when it comes to building your own AI applications with your newly acquired skills!

What you will learn