Historians of artificial intelligence never talk about AI’s progress in the 1960s without a reference to Eliza, the first virtual personality. Eliza was a text-chat program written in 1966 by MIT AI expert Joseph Weizenbaum to parody a Rogerian psychotherapist, largely by turning every statement by the “patient” back into a question. If you tell Eliza “I am feeling blue today,” it’s apt to respond, “Do you enjoy feeling blue today?” To modern users, the pattern is obvious, and the illusion of talking to a real person drops away almost instantly. (See for yourself here or here.) Yet many people who used Eliza when the program was new were convinced, at least temporarily, that it was a real person.
Now there’s a Web-based service that, in essence, lets you set up your own Eliza and train it to mimic your own personality. No one will be fooled into thinking it’s you, but MyCyberTwin, launched earlier this month, does a decent job of acting as your stand-in or virtual public-relations agent when you’re not reachable. If you embed your cybertwin in your blog, website, or MySpace profile, visitors can learn about you through an open-ended conversation. You can program your cybertwin with as much factual information and as much of your personality as you like. If you think visitors to your blog might ask “What are you doing Saturday night?”, you can train it to respond “Going to see Harry Potter with friends. Why don’t you join us?”
MyCyberTwin is free, up to a point. About 10,500 people have signed up for the service, which is a venture of a Sydney, Australia, company called RelevanceNow and is still in its beta-testing phase. Of course, the concept only goes so far. Like Eliza, your cybertwin has no real intelligence at its core, and it must resort to lame conversational gambits if you haven’t provided it with a canned answer to your visitor’s specific question. Helpfully, however, the MyCyberTwin site contains extensive tools to help you anticipate those questions, such as personality tests and quizzes about your views on sex, politics, and religion. Those tools are all free too. RelevanceNow plans to make money by charging heavy users, such as businesses, by the conversation if their cybertwins have more than 500 visitors per month; by licensing MyCyberTwin to social-networking sites, which might integrate it into their offerings; and possibly, in the future, through targeted Web advertising.
“We wanted to build software clones of humans that learn about you and effectively function on your behalf,” says Liesl Capper, cofounder and CEO of RelevanceNow. “The problem with creating a chat AI is that it’s very laborious, trying to think of variations on what people will say and then creating responses. Building one has always been a labor of love that takes months, if not years. What we have built is the ability for people to make a cybertwin really quickly.”
“Quickly” is a relative term. If you opt to take advantage of MyCyberTwin’s entire collection of training tools, you could easily spend a day or more programming your cybertwin. The reward, however, is a more convincing simulacrum than Eliza or most of the other consumer-level “chatterbots” that have been written since Eliza’s day. Because a cybertwin can be armed with limitless information about you, it’s much more intimate and engaging to converse with than many competing chatterbots, such as the instant-messaging-based bots from Spleak, avatar-based chatterbots from Verbots, or Microsoft’s virtual search assistant, Ms. Dewey.
Of course, academic and corporate AI experts have built more-advanced chatterbots in hopes of one day passing the so-called Turing test by convincing human judges that a machine is human. Since 1991, the annual Loebner Prize competition has offered $25,000 to the programmer of the first chatterbot that passes this test in a text-only conversation; so far, the prize has gone unclaimed. But Jabberwacky, the winner of the smaller $2,000 prize for the most human-seeming chatterbot in 2005 and 2006, is capable of deep and sometimes bizarre conversations that make a cybertwin sound rather vacant. Designed by Rollo Carpenter of the U.K. AI company Icogno, Jabberwacky stores every conversation it has ever had and uses pattern-matching techniques to generate contextually appropriate responses in a new conversation.
For $30 per year, you can get a Jabberwacky chatterbot that mimics your own personality. But it can’t be embedded into another website, and it must be trained through lengthy, repetitious conversation. Starting off with MyCyberTwin is far easier: you choose a base personality, such as “warm-hearted, intellectual” or “cheeky, down to earth,” and answer about 30 questions about yourself drawn from the Myers-Briggs Type Indicator, a widely used personality-assessment system. After that, however, there are optional “classroom” modules that ask a total of 425 fairly deep questions in 18 subject areas, such as family, humor, philosophy, and politics.
Writing a thoughtful answer to a question such as “What is the meaning of life?” is bound to take at least a few minutes; now multiply that by several hundred. Indeed, the classroom questions are so probing and thorough that I doubt many MyCyberTwin users will put in the work. I made an attempt with my own cybertwin. The basic setup process was simple and easy. But I spent about three hours on the classroom questions and only completed three subject areas.
Once you’re finished training your cybertwin, you can give people the link to your personal page at MyCyberTwin.com, or you can make your cybertwin appear on another Web page by pasting the provided code into that page’s source HTML. If your cybertwin turns out to be extremely popular and you pass the limit of 500 conversations per month, you can buy more credits for a price that RelevanceNow hasn’t yet announced.
Also, be careful what you tell another cybertwin. Although the MyCyberTwin site does not make it obvious, the service actually saves a transcript of every chat session for the perusal of the cybertwin’s owner. That’s mainly so that visitors can leave information such as phone numbers, e-mail addresses, and requests for more information. But an unsuspecting user who gets into a racy conversation with a cybertwin on the assumption that there’s no one at home might be in for some embarrassment.
MyCyberTwin can be a bit slow, taking up to 10 seconds to “think” before it responds to a visitor’s question. Nonetheless, creating a MyCyberTwin chatterbot can give your online admirers much deeper, quicker access to your personality and background than dry autobiographical statements or even months of blog entries.
“People are spending a lot of time putting their personalities online,” Capper notes, whether through their MySpace pages, blogs, or avatars in virtual worlds such as Second Life. “It may only be for an audience of 10 people, but it’s important to them, and it’s a taste of things to come. This way, you can have your friends chat to you when you’re sleeping. It’s about engaging with people and answering their questions without having to go through a hundred e-mails.”
A quick guide to the most important AI law you’ve never heard of
The European Union is planning new legislation aimed at curbing the worst harms associated with artificial intelligence.
It will soon be easy for self-driving cars to hide in plain sight. We shouldn’t let them.
If they ever hit our roads for real, other drivers need to know exactly what they are.
This is the first image of the black hole at the center of our galaxy
The stunning image was made possible by linking eight existing radio observatories across the globe.
The gene-edited pig heart given to a dying patient was infected with a pig virus
The first transplant of a genetically-modified pig heart into a human may have ended prematurely because of a well-known—and avoidable—risk.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.