A rough definition of conversational AI is a program that attempts to appear human while interacting with a user. Computer programs designed to simulate an intelligent conversation, though appearing to be intelligently interpreting the human input prior to providing a response, in most cases, simply scan for keywords within the input and pull a reply with the most matching keywords or the most similar wording pattern from a local database.
The Turing Test
The Turing Test is a proposal for a test of a machine's capability to perform human-like conversation. Described by Professor Alan Turing in the 1950 paper "Computing machinery and intelligence," it is based on the "imitation game" and proceeds as follows: a human judge types a conversation with two other parties, one a human and the other a machine; if the judge cannot reliably tell which is which, then the machine is said to pass the test. It is assumed that both the human and the machine try to appear human. In order to keep the test setting simple and universal the conversation is usually limited to text-only.
The classic early attempts at conversational AI are ELIZA and PARRY.
ELIZA attempted to emulate a therapist by rephrasing the users statements into questions. Eliza worked by simple parsing and substitution of key words into hard coded phrases. Depending upon the initial entries by the user, the program could be instantly detected, or could continue through several interchanges.
PARRY attempted to simulate a paranoid schizophrenic. The program implemented a crude model of the behavior of a paranoid schizophrenic based on concepts, conceptualizations, and beliefs. It was a much more serious and advanced program than ELIZA.
A.L.I.C.E. uses a programming language AIML (Artificial Intelligence Markup Language) that is specific to its program. It is still based on pattern matching without any reasoning, the same technique used by Eliza.
Jabberwacky is designed to learn language and context through interaction with humans. There are no fixed rules or principles programmed into the system and it operates entirely through user interaction. The system stores all of the conversations and user comments and attempts to use this information to find the most appropriate response.
The results of these recent programs still appear fairly poor. A program has yet to pass the Turing Test and it seems reasonable to state that there is currently no general purpose conversational artificial intelligence.