Talking machines have fascinated people for longer than computers have been common. A short tour of the history helps explain why today’s chatbots look the way they do, and it is a theme that many general books on the subject open with.
The question that started it
In 1950 the mathematician Alan Turing published a paper that asked whether machines can think. He proposed replacing that hard question with a game, now usually called the Turing test, in which a judge converses in writing with a person and a machine and tries to tell which is which. The test remains a cultural reference point, though researchers debate how useful it is as a measure of intelligence.
ELIZA, 1966
Joseph Weizenbaum at MIT built ELIZA, often described as the first well-known chatbot. Its best-known script imitated a psychotherapist who reflects statements back to the patient. The program used simple pattern matching, yet many users felt it understood them. Weizenbaum was troubled by how readily people attributed understanding to it, and he later wrote cautiously about the issue. The effect has since been named after the program.
Rules and scripts
Through the following decades, chatbots stayed largely rule-based. Developers wrote patterns and responses by hand. In the 1990s the program ALICE, created by Richard Wallace, used a large set of such patterns written in a markup language and won several chatbot competitions. Text games and early customer-service systems relied on similar ideas.
Chat on the web
The growth of the web and instant messaging brought bots to a wider audience in the 2000s. Companies put simple assistants on their sites to answer common questions, and messaging platforms opened up to developers around the middle of the 2010s, which led to a wave of business bots.
Voice assistants
Voice changed expectations. Siri arrived on the iPhone in 2011, followed by assistants from other large technology firms and smart speakers in homes. These systems combined speech recognition, language understanding and a growing list of “skills”.
Neural language models
Progress in machine learning, and the 2017 Transformer architecture in particular, made it possible to train large models on huge amounts of text. When ChatGPT was released at the end of 2022, it brought fluent, open-ended conversation to a mass audience almost overnight. Since then, businesses have been exploring how to combine such models with company data, safeguards and clear limits.
Why the past still matters for readers
Many introductory books reuse these milestones as their organising story, so recognising them makes it easier to place new ideas. It also guards against hype. Each wave of progress, from scripted bots to voice assistants to language models, came with strong promises, and each found real limits. Knowing the pattern helps you weigh claims in any guide with a calm, critical eye.
Lessons from the history
Three themes recur. People are quick to read intelligence into fluent text. Hand-built rules offer control but are costly to maintain. Learned models are flexible but need testing and oversight. Compare approaches in rule-based and AI chatbots, or go back to the home page.