The first ‘successful’ chatterbot from 1966
ELIZA introduced the world to the chatbot a full 50 years ago. What can be learnt from ELIZA, and how does this breakthrough help us understand today’s chatbot potential?
Joseph Weisenbaum was said by MIT to have taken a meaningful step towards AI when in 1966, he coded ELIZA.
It was a program that demonstrated both the possibilities and limitations of person-to-machine communications.
ELIZA became famous playing the part of a psychiatrist. The program’s DOCTOR script imitated Rogerian psychoanalysis: non-directional questions allowed users the freedom to speak, and when things were unclear, what the user said would be mirrored in the form of a question.
The program amused many people, and even managed to briefly convince some that they were talking to a human. But DOCTOR had no idea what was being said. The program would respond to keywords, assign values to those keywords, and then assess an output for input values. It was a hard-coded illusion, and would often be dispelled after only a couple exchanges.
It showed just how incredibly dynamic and complex a satisfying chatbot would have to be.
Since then, conversational computing has been a series of accomplishments, but along a certain timeline the technology inevitably disappoints the user. Each breakthrough in chatbot programming buys another couple of exchanges, and it’s a slow slog forward in terms of constructing complicated, and useful conversations.
Some interesting tricks have been programmed to deal with a chatbot’s inability to understand or answer some questions. For instance, when Siri or Cortana give a little sass, this is not human-like intelligence, but the exact opposite. It is a deflection away from having to deal with something Siri is not programmed to deal with. In essence, it is a snarky way of saying “try again.”
But in 2016, with the rise of chatbots on Kik, Slack, Facebook Messenger and other platforms, we are starting to enjoy talking to the bots more and more. Bit by bit the technology is approaching a satisfying level, and might just need one, big, sophisticated breakthrough — much like ELIZA was for 1966, and Siri was for 2011.
So what can we learn from ELIZA, a bot from 1966?
You are only as good as your tools. This is timeless wisdom. The DOCTOR script that ELIZA ran was ingenious because it made full use of the programming potential of 1966. It did this by choosing the perfect ‘character’ to emulate. A chatterbot psychologist asks questions, prompting a response. So the patient does almost all the work in the conversation. The chatterbot simply encourages them to continue talking.
And what does this tell us about chatbot development in 2016?
We need new chatbot ‘characters,’ ones that hit that sweet spot, where the type of conversation you would have with this ‘character,’ (like DOCTOR) befits the technology of today. Without a chatbot that mirrors a real-world reference point, it will be difficult to truly ascertain the limitations and possibilities of AI, and build on it.
Is it time to update ELIZA?
Maybe the character we need is another, newer and more complex DOCTOR. It would serve to clarify how far we’ve come in chatbot technology over the past 50 years, and in what ways we need to improve. And in the age of open source collaboration, with enough programmers capable of contributing to a chatbot script like that of ELIZA’s DOCTOR, we could truly ascertain and redefine the limits of person-to-machine communications through our collective understanding of what a ‘character’ should talk and act like.
In this vein of collaborative programming, Mark Zuckerberg is considering giving away the code he created for a JARVIS-like personal assistant for his home. We think such an approach would be the quickest way forward. Programming a satisfying chatbot is extremely complicated and strenuous, and getting it just right is certainly something no one will do alone. Not even the Zuck.