A chatbot is a software program designed to conduct a conversation with a human user via text or speech, simulating the patterns of natural dialogue. Chatbots range from simple rule-based systems that match keywords to scripted replies, through statistical dialogue systems, to modern chatbots built on large language models that generate open-ended responses. The term predates modern deep-learning-based AI and has been applied to customer-service bots, virtual assistants, companion apps, and general-purpose assistants such as ChatGPT.
History
The first widely cited chatbot was ELIZA, written by Joseph Weizenbaum at MIT in 1966, which used simple pattern matching to imitate a Rogerian psychotherapist and convinced some users they were talking to an understanding listener, a reaction Weizenbaum found troubling and later wrote about critically. Subsequent decades produced PARRY (1972), simulating a patient with paranoid schizophrenia, and A.L.I.C.E. (1995), a rule-based chatbot that won the Loebner Prize multiple times. Commercial chatbots proliferated in the 2010s as messaging platforms and voice assistants such as Siri and Alexa added scripted, intent-classification-based dialogue. The category was transformed by ChatGPT's launch in November 2022, built on a version of GPT-3.5 refined with reinforcement learning from human feedback, which replaced scripted flows with a general-purpose model capable of open-ended, coherent multi-turn conversation and reportedly reached 100 million users within two months, then the fastest consumer application growth on record.
Architecture
Early chatbots relied on hand-written rules, decision trees, or keyword matching, as in ELIZA; later systems added statistical natural-language understanding to classify user intent and fill parameters for narrow tasks such as booking a flight. Since the early 2020s, most general-purpose chatbots are built on transformer-based models trained through large-scale pretraining on text corpora and aligned with human preferences via reinforcement learning from human feedback. Many production chatbots add a system prompt to set persona and constraints, and some use retrieval-augmented generation to ground answers in external documents rather than relying solely on what the model memorized during training.
Reception and criticism
Chatbots have raised recurring concerns: users anthropomorphizing systems that do not understand them, sometimes called the ELIZA effect; hallucination of false information stated confidently; and, for companion-style chatbots, worries about emotional dependency, particularly among younger users. Weizenbaum's own reaction to ELIZA anticipated much of this later debate. Proponents point to chatbots' usefulness for customer support, education, accessibility, and as a natural-language interface to complex software, a role that has expanded as chatbots gained tool use and became a foundation for broader conversational AI and AI agent systems.