Java Cookbook, 1st Edition

Chapter 13. Machine Learning/Artificial Intelligence

13.0 Introduction

The notion of “artificial intelligence” was envisioned in science fiction for some years before the first software developments in that direction. Today we still have AI in fiction, like Jarvis, Iron Man’s helmet AI. But we also have AI in real life, coming to popular attention starting with ChatGPT in the early 2020s. It’s actually been in the works for three quarters of a century.

There are two main approaches to AI: machine learning (ML) and coding. Coding toward AI was one of the factors that led to the creation of the Lisp language in the late 1950s. The goal of coding AI was defined as Artificial General Intelligence, or AGI: an AI that could pass the Turing Test of interacting exactly as a human would. Of course, “intelligence” has many different definitions…​

Machine learning (ML), based on neural nets, has been around since at least the 1980s. ML doesn’t necessarily aim for AGI, just to perform classification and generation tasks better—or at least faster—than humans. Unlike coded solutions that can be understood by humans, neural nets are based on training. Training data is fed into the AI, which ingests and stores it, for use when interacting with a user. Over time developers connected several neural nets in a row, resulting in the term deep learning.

This training process typically consumes massive amounts of computing power, so only small datasets can be trained on a typical mid-2020s desktop ...