Chatbots Before The Hype

Chatbots are receiving renewed attention in 2016. Messaging platforms are opening APIs, companies are experimenting with automated assistants, and services such as Facebook Messenger are becoming places where software can talk to users directly.

The idea itself is not new. ELIZA demonstrated a simple form of conversation in the 1960s, and rule-based chat systems have existed for decades. What feels different now is the environment around them. People already spend a large part of the day inside messaging applications, so a conversational interface can appear where users already are.

Most current chatbots are not intelligent in the science-fiction sense. They usually follow designed flows, recognize limited intents, search databases, or connect to existing services. A good chatbot for booking a ticket does not need to understand every possible sentence. It needs to understand enough of the small domain to be useful.

This creates an interesting interface problem. Buttons and menus show users what choices are available. Conversation is more open. A user can type almost anything, which creates expectations the software may not be able to meet. When a bot fails to understand a simple sentence, the experience can become frustrating very quickly.

Natural-language processing is improving, but I think successful bots will need careful limits. It may be better to be clearly useful in one narrow task than to pretend to be a general digital person.

Chatbots are interesting because they reduce the visible interface. Instead of learning where a command lives inside a menu, the user can ask for something. Whether this becomes a major computing interface or only another temporary trend is still unclear.

For now, the best chatbot is probably not the one that sounds most human. It is the one that knows what it can do and does not waste the user’s time.