What are chatbots?
Chatbots are computer programs that use NLP to interacts with users in a human-like fashion. Interactions usually take place in a web interface or a chat program such as Facebook Messenger. Some chatbot systems run in programs such as slack or by SMS. The aim of a chatbot is to mimic the responses and reactions of a human being.
The term “chatbot2 originates from ChatterBots, TinyMuds, and the Turing Test: Entering the Loebner Prize Competition, which was published b Michael L. Mauldin in 1995. It is a diminutive form of the word “chatter-bots”.
Chatbots have existed in some form since the 1940s. However, it has only been in the last two decades that they have grown in popularity.
One of the first milestones in chatbot delivery was the publication in 1950 of Computing Machinery and Intelligence by Alan Turing. The ideas discussed within the papers formed the foundation of much of what we understand and how we think today about chatbots. Central to the paper is the now-famous Turing Test, which still remains the go-test assessment most know for AI.
The Turing Test is where a human sits down and types answers into a computer. The questions are answered. If a computer answers the question and the human cannot tell that they are speaking to a computer, then the machine has passed the test.
It took fourteen years for a program to pass a version of the Turing Test. That was in 1964 when Joseph Weizenbaum began working on ELIZA at MIT’s Artificial Intelligence Laboratory. ELIZA was released in 1966 and was written in Weizenbaum’s MAD-Slip programming language. In 1974, Weizenbaum published Computer Power and Human Reason: From Judgment to Calculation, in which he expounded on his thoughts about artificial intelligence.
Another computer program around that time that achieved some prominence was Kenneth Colby’s PARRY, which was published in 1972. PARRY was designed to mimic the actions of a human with paranoid schizophrenia. In its year of publication, PARRY took part in a demonstration at the International Conference on Computer Communications where it conversed with ELIZA. Eventually, PARRY would pass a version of The Turing Test.
Later chatbots to have passed The Turing Test include 1989’s Jabberwacky, which attempted to incorporate machine learning, and 1995’s ALICE.
How do chatbots work?
Early chatbots relied on very-basic structures in order to simulate a chat and consequently were neither fluid nor able to accurately simulate a human interaction.
Modern chatbots, however, rely on artificial intelligence and machine learning. This means that the bots can take what is being of them, extract from it the most-salient information, and search within an organisation’s data and archives to find the most-relevant answer.
Modern chatbots often rely on interfaces such as Facebook Messenger, Slack, or SMS.
Use cases
Good for routine questions, meaning that HR can better deploy resources to more-complex queries.
Chatbots are widely used in retail and other customer service roles, answering routine queries and providing answers. They are also used within large organisations to help employees navigate HR and corporate policies.
Recent research found that while 37 per cent of respondents were happy to use chatbots for getting quick answers in an emergency, 35 per cent said that they would use a chatbot to resolve a complaint or problem, or to get detailed answers or explanations. Among those surveyed, 35 per cent believed chatbots could answer complex questions, with 37 per cent thinking that they could give detailed, expert answers.
There are numerous benefits for chatbot implementation. The Harvard Business Review found that automating answers could be vital to retaining business: it found that if there was a five-minute delay in responding after a sales lead reached out, the odds of converting that lead decreased tenfold.
Deloitte estimate cost reductions of 15-90 per cent with the introduction of chatbots.