From ELIZA to Character Chat: The History of Conversational Programs

Before the chatbot became a product

The history of conversational programs begins before the familiar web chat window. In the 1960s, researchers were exploring how computers could process language and respond to typed instructions. Early time-sharing systems allowed multiple users to interact with a central computer, creating opportunities for messaging, discussion, and experiments in human-computer communication. The machine was not simply a calculator anymore. It could participate in an exchange that resembled dialogue.

Joseph Weizenbaum's ELIZA became a defining example. Developed at the Massachusetts Institute of Technology between 1964 and 1966, the program was described in his January 1966 paper, ELIZA: A Computer Program for the Study of Natural Language Communication Between Man and Machine, published in Communications of the ACM. Its existence established a recognizable foundation for later chatbot history, although conversational software had several different technical and cultural paths ahead.

ELIZA and the appearance of understanding

ELIZA analyzed typed sentences using keyword recognition, decomposition rules, and response templates. Its best-known script, DOCTOR, imitated a Rogerian psychotherapist. When a user mentioned a family member, for example, the program could select a related response pattern and turn the statement into a question. It could produce an exchange that felt responsive without maintaining a human-like understanding of the subject.

The distinction mattered. Weizenbaum had designed the system to investigate language interaction, but some users attributed more understanding to it than its mechanisms justified. In Computer Power and Human Reason, published in 1976, he examined the assumptions people made about computers and their apparent intelligence. The phenomenon later associated with ELIZA's reception became known as the ELIZA effect: the tendency to perceive understanding or intention in a system that produces convincing conversational signals.

From isolated experiments to networked communities

Conversational programs developed alongside a separate history: the growth of online communities. Bulletin board systems appeared in the late 1970s, giving users ways to post messages, exchange information, and participate in shared interests. Usenet, which emerged in 1979, organized discussions into newsgroups. Internet Relay Chat followed in 1988, making real-time group conversation a familiar online activity. These systems created social environments in which automated participants could become part of ordinary interaction.

The early internet was not a single unified community. Bulletin boards, university networks, mailing lists, and commercial services each developed their own conventions. Members learned where to find specialist discussions, how to identify regular contributors, and which spaces welcomed newcomers. As chat rooms and message boards expanded, automated accounts could perform useful work: announcing information, responding to commands, maintaining channels, and helping enforce local rules.

Chat-room bots acquire social roles

A chat-room bot did not need to imitate a person to become familiar. Its value often came from reliability and a recognizable role. On IRC, bots could manage channel access, maintain lists, provide information, and respond to commands. Some were operated by community members, while others became part of the infrastructure that kept a channel functioning. Their presence made automation visible as a participant in a shared environment.

The difference between an operational bot and a conversational character was not always sharp. A bot that answered questions about channel rules might also develop a familiar name and recognizable style. A game bot could provide an ongoing activity for regular visitors. In each case, members learned its capabilities through repeated encounters. The community supplied much of the meaning: the same automated response could be useful in one channel and irritating in another.

Scripted personalities and the rise of character interaction

As online communication moved into web-based services, conversational entertainment became more accessible. Scripted virtual characters appeared in websites, games, and interactive environments. Some relied on fixed dialogue trees, while others used pattern matching or collections of prepared responses. These systems could offer a consistent personality because their behavior was shaped by a limited set of rules and authored conversational paths.

This approach extended the central idea behind ELIZA. A character did not need a sophisticated model of the world to produce an identifiable conversational style. Its personality could emerge through recurring phrases, preferred subjects, and predictable reactions. The limitation was equally clear: an unexpected question might lead to repetition, a generic answer, or a conversation that stopped making sense. Developers had to decide how much freedom to allow without losing the character's recognizable identity.

Large language models change the conversational range

Modern language models changed the mechanics of character interaction. Instead of selecting only from a relatively narrow collection of scripted responses, these systems can generate new text based on conversational context and learned language patterns. This allows longer exchanges and a wider range of possible replies. Character chat has consequently developed into a distinct form of online entertainment and interaction, with systems presenting fictional personalities through text and, in some services, additional media.

TheFriend is one present-day example of this category: an adults-only platform offering text conversations with more than 90 fictional AI characters, alongside narrated audio stories. Such services illustrate how the conversational character has moved beyond the occasional channel bot or fixed dialogue tree into a dedicated environment organized around individual fictional identities.

The change should not be confused with the disappearance of earlier limitations. Generated conversation can be inconsistent, lose track of details, or produce confident statements that are not grounded in reliable information. A recognizable character experience is an interface and editorial design choice, not evidence of human understanding. The historical continuity lies in the social form of the interaction, while the underlying technology has changed substantially.

The older internet remains visible in today's chat

Modern conversational systems inherit habits established in earlier online communities. Users still enter a space, learn its conventions, recognize recurring personalities, and decide whether an interaction deserves continued attention. Chat-room bots introduced the expectation that automated accounts could be useful members of a digital gathering. Scripted characters extended that expectation into entertainment. Generative systems have widened the range of responses without removing the social expectations that surround them.

For community managers, the distinction between a bot and a character matters. A moderation bot should be judged by reliability, transparency, and the authority it has been given. An entertainment character is judged more by conversational consistency and the expectations established by its presentation. Both need clear boundaries. When an automated account participates in a shared community, members should be able to recognize its status and understand what it is authorized to do.

The next step in reading this history is to preserve the artifacts that show how each generation worked: ELIZA's original paper and surviving scripts, IRC bot documentation, archived message-board discussions, and examples of early scripted characters. Begin with Weizenbaum's 1966 Communications of the ACM paper, then place those early conversational rules beside a preserved chat-room bot manual. The comparison makes the distance between a keyword-triggered reply and a modern generated conversation concrete.

Questions readers ask