Название: AI Literacy Fundamentals
Автор: Ben Jones
Издательство: Data Literacy Press
Год: 2024
Страниц: 240
Язык: английский
Формат: epub (true)
Размер: 10.1 MB

Feeling overwhelmed by AI? It's not you—it's the breakneck speed of technological progress. To quickly get into the AI conversation, you need a clear and simple foundation of knowledge to build on. This book is a friendly primer on the basic concepts of AI, how it's already snuck into our daily lives, and what we need to know to prepare for the future.

Ben Jones, an expert at breaking down technical concepts from teaching thousands of people the basics of data literacy, lays out everything you need to know to join the AI conversation, from the history of AI to the Deep Learning revolution happening today. This technology is here to stay. Time for you to pull a seat up to the table.

Machine Learning (ML) is an umbrella term describing the study and use of different types of statistical algorithms that we can apply to data sets in order to learn from the patterns in those data sets how to accomplish specific tasks. When we apply a Machine Learning algorithm to a data set to learn its patterns, we say that we are training a model.

The data set we use for training purposes could be any group of items, in digital form, from the real world: a bunch of image files, e-books or other digital documents, audio files of people talking, spreadsheets of historical stock prices – the list of possibilities is endless. In Machine Learning, the data sets are typically very large. The more data, the better. We’re not talking about a dozen digital photographs here; we’re talking about thousands or even millions of them. Of course the training data set does not include all of the possible items; it’s a sample, not an entire population. The point is that the model will learn from the sample data how to work with items that aren’t in the sample.

The algorithm is the set of instructions that the program follows in the training process. It’s like a recipe of sorts, a step-by-step procedure to learn from the training data. There are different kinds of Machine Learning algorithms, from neural networks to decision trees to various clustering algorithms, and we’ll consider some of them in this book.

Deep Learning is a special branch of Machine Learning that involves a specific family of artificial neural networks: deep neural networks (DNN). DNNs have changed the world of AI, and in so doing they have changed the world we live in. Any neural network that contains two or more hidden layers is considered a deep neural network. Unsurprisingly, then, a neural network with a single hidden layer is sometimes referred to as a shallow neural network. Furthermore, the number of hidden layers corresponds to the depth of the deep neural network.

To wrap up our own study of deep learning and neural networks, I’d like to instead consider a few different types of deep neural networks that have led to breakthrough capabilities in the fields of computer vision and natural language processing. The first type is the convolutional neural network (CNN), sometimes called the ConvNet. This type of deep neural network has produced significant breakthroughs in computer vision – giving AI models the ability to recognize objects and to analyze photographs and videos. CNNs are the reason that you and I can deposit a handwritten check into our bank account with a photograph we take using our smartphone, and they’re central to advances in self-driving cars, medical imaging, and many other applications.

Generative adversarial networks (GANs), introduced by Ian Goodfellow and his colleagues in 2014, consist of pairs of neural networks coupled together in training. One of the two models in the pair is the generator, which tries to generate an output (it could be words, images, audio, or video) that’s indistinguishable from real data. Think of an image of a face that looks so real that you can’t tell it doesn’t actually belong to a real person. The other neural network in the pair is the discriminator. Its job is to determine whether a given input is a real sample from its training data, or a fake that has been created by the generator. These two neural networks are trained together and compete against one another, and are therefore in an “adversarial” relationship.

The Transformer and Large Language Models (LLMs): A large language model (LLM) is a particular type of deep neural network that has been trained on massive amounts of text data to process and generate human language. State-of-the-art LLMs have billions of parameters – the weights and biases of its many neurons – that have been adjusted and fine-tuned during training to enable these models to produce outputs that resemble human speech. Today’s advanced LLMs, such as OpenAI’s GPT-4, Meta’s LLaMA 2, and Mistral AI’s Mixtral-8x7B, can perform a wide array of human language tasks with speed and relatively high accuracy. Due to their limitations as well as the many imperfections and biases inherent in their training data, they’re by no means perfect. But there’s no denying that they’re incredibly capable and powerful. One reason for their recent surge in capabilities is the pivot to a technology known as the Transformer.

Praise for AI Literacy Fundamentals:

"I can’t think of a better written and more thoroughly researched introduction to the fundamental concepts of AI Literacy than Ben’s wonderful book. I cannot recommend it enough. Read. Be inspired. Be ready for our changing world." - James Wilson, author of Artificial Negligence

"A highly enjoyable and easy read covering both the history and evolution of AI, as well as several sections with a lot more depth. It explains the connection to data, and various bias in the popular AI tools of today. The book explains the pros and cons in the last section, and there is a really good glossary as well." - Sebastian Anthony, SVP Analytics and Reporting, BofA

Contents: