There are two kinds of people working with bots today: Those who inevitably allude to ELIZA, the 1964 computer program, and those who’ve never heard of it. But ELIZA’s story contains a valuable lesson for modern bot marketers, even if artificial intelligence concepts bore them stiff.
ELIZA was one of the first natural language processing programs, written in 1964 by researcher Joseph Weizenbaum at the Massachusetts Institute of Technology’s Artificial Intelligence Laboratory. Weizenbaum’s motive was actually contrarian: He wanted to show how shallow and superficial communications between man and machine actually were. The program’s name isn’t an acronym — it’s named after Eliza Doolittle from the play Pygmalion (or the better known musical My Fair Lady), a lower-class woman who was taught to successfully mimic the outward traits of upper-class Brits.
Weizenbaum, clearly being ironic, programmed ELIZA to mimic techniques of Rogerian psychotherapy, in which a therapist doesn’t offer insights but rather keeps the person talking to examine their own thoughts, feelings, and beliefs.
Tell ELIZA that you are angry, and it will respond with one of a limited set of sentences that include your own words, such as, “How long have you been angry?” or “Do you enjoy being angry?”
The trick is to keep you talking. ELIZA makes no attempt to parse your context, to offer any solutions, or even to remember what you are talking about. Yet many people who try it succumb to what’s known as the ELIZA effect: They mentally attribute human behavior to a machine which has no thoughts or emotions of its own.
As a result, some people find themselves having long discussions with ELIZA. Weizenbaum claimed that his secretary sometimes balked at his interruptions of her personal chats with the program. He wrote, “Extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.”
One person’s delusional thinking is another person’s marketing dream of customer engagement. Today’s bots, says Instigate.ai co-founder Marc Canter, aren’t much more sophisticated than ELIZA: “That’s how most of today’s chatbots work — giant tables of words and phrases which are matched to user’s input.” But the goal of many bot adopters is simply to become a major part of people’s lives. A place where they spend long periods of time daily. It’s a powerful way to build a consumer’s relationship with a brand.
For that purpose, it’s not necessary so much to have deep algorithms that deduce chatters’ logic and emotions, so much as it is to have an understanding of what makes other human beings tick. What responses will prompt them to keep talking, to open up personally to a bot, and to keep coming back?
There’s a long-running (since 1964!) joke among some programmers that no computing platform is complete until someone implements ELIZA on it. (See GitHub for support of that theory.) When I worked at Splunk, we seriously considered building ELIZA into the network analysis tool’s help system, so that beleaguered systems administrators could have someone to whom they could complain freely.
There’s a similar potential for marketers seeking to make their brand a bigger part of people’s lives. An ELIZA-like bot could easily be customized to poke at the personal thoughts of those who come to talk about a specific person, product, or service, keeping them talking about something for which they already have an attachment in order to deepen that attachment: “Does Taylor’s music help when you are angry?”