Introduction to AI Services

In the last few years, the field of artificial intelligence (AI) has undergone remarkable advancements, revolutionizing various domains and reshaping the way we think about and interact with technology. One particularly fascinating branch of AI that has gained significant attention recently is Generative AI. By enabling machines to exhibit creativity (or, more specifically, the appearance of creativity), Generative AI has opened up new frontiers in areas such as art, music, design, and storytelling, in addition to chat and human interaction.

Before we get too ahead of ourselves, let’s talk about some core concepts to help shed some light on how all this works.

What do all these AI terms mean? Generative AI, in particular, refers to a class of algorithms and models that can autonomously generate new and (somewhat) original content. Unlike traditional AI systems, which rely on pre-defined rules or explicit instructions, Generative AI systems are designed to learn from patterns and existing data to produce novel outputs. These systems leverage deep learning techniques, such as generative adversarial networks (GANs), variational autoencoders (VAEs), and recurrent neural networks (RNNs), to emulate the creative processes of the human mind.

As you’ll see, AI has a lexicon all its own. What do we mean when we say things such as generative adversarial networks and variational autoencoders? Let’s make a quick detour and define some of the terms that we’re going to use:

• Algorithm: An algorithm is a set of rules (typically expressed in a computer programming language) that are followed when solving problems.
• Neural network: When we talk about neural networks, we’re talking about computer systems and interactions that are modeled on our understanding of the human brain and nervous system. Like the human brain, the fundamental building blocks of artificial neural networks are referred to as neurons (nodes), each of which connects to other nodes. The connections have concepts of weight and bias, and when inputs reach certain thresholds, they flip on the next node in the chain. Imagine a neural network as layers of nodes arranged in grids, with each node connecting to multiple nodes on the adjacent layer, and each node’s output being used to influence the input in the adjacent layer’s nodes.
• GAN: A GAN is comprised of two neural networks that compete based on the same source data. GANs can create synthetic data that is unique but imitates the seed data.
• VAE: A VAE is an algorithm that has two functions. The first takes a complex data structure and then stores a more simplified version of it with some amount of randomness, while the second takes the simplified version and then generates a more complex output. Imagine the encode function as taking a high-resolution picture of a tree, downscaling it (so that it still looks like a tree but is missing some data and possibly looks blurry), and adding a few random pixels to it. When the decode function is activated, it retrieves the simplified data that’s stored and uses it to reconstitute a more high-resolution image of a tree. The new picture looks similar to the original, but partially due to the loss incurred through the simplification of original data and partially due to the insertion of some amount of randomness by the encoder, the new picture is also different.
• RNN: An RNN is a type of artificial neural network that can process sequential data by preserving information from previous steps.
• AI model: An AI model is a mathematical algorithm that mimics human intelligence, processing data to make predictions and generate outputs. It learns from training data to perform specific tasks such as image recognition or natural language processing.
• Large language model: A large language model is a type of AI model that is designed to understand and generate coherent and contextually relevant, human-like text. The popular ChatGPT is an example of a large language model.

There are many more complex concepts (including many more types of neural networks and AI models) behind deep learning and AI systems.

In addition to Generative AI, many types of AI models are currently in use today, such as those designed to do the following:

• Estimate shipping routes
• Predict traffic patterns and congestion
• Find weather anomalies
• Identify objects in pictures

Each of these different types of models depends on vast quantities of existing data and purpose-built algorithms, combined with training procedures to help the models “learn” how to predict or identify things.

Throughout this book, we’ll be using a variety of AI technologies – from prebuilt, purpose-oriented models to Generative AI. By the time you reach the final examples and exercises, I hope you’ll have some exciting ideas on how you can accelerate your team, organization, or even personal life with AI.