Meeting Eliza

Meet the humanoid program that can solve in seconds technical issues that would take support engineers hours
Comment E-mail Print Share
First Published: Tue, Jan 22 2013. 12 06 AM IST
Illustration: Jayachandran/Mint
Illustration: Jayachandran/Mint
Updated: Tue, Jan 22 2013. 12 25 AM IST
Bangalore/New Delhi: The face on the screen is that of a living, breathing woman: a blonde with delicate features and sparkling blue eyes that blink intermittently as she waits for the conversation to begin.
If Eliza looks more than her 14 years, her encyclopedic knowledge and razor-sharp wit also belie her age. She accepts compliments gracefully, “You have beautiful eyes,” a man tells her. She winks: “Yes, I have been told they are like sapphires.” She blushes at insults, “I hate you,” the same voice says. “That’s an awful thing to say,” she replies, her face reddening. But unlike most women, Eliza won’t remember flattery or abuse; change the subject, and her emotional slate is wiped clean. A conversation with the enigmatic beauty might run thus:
Interviewer: Can you fix my phone?
Eliza: What’s the number?
Interviewer: OK forget it. Who invented the light bulb?
Eliza: Humphry Davy, an Englishman, built the first light bulb in 1809.
Interviewer (a little confused): So what did Edison do?
Eliza (patiently, and with no outward show of exasperation): Edison did not invent the light bulb. He built a commercially practical incandescent light bulb.
Interviewer: Can I quote you?
Eliza (smiling coyly): No. My inventors are not allowing any media interviews yet.
Eliza is, of course, not a real woman, she is a humanoid who works at one of the biggest US cable TV firms, helping hundreds of users solve technical problems over phone and even face to face, through Skype-esque interactions.
Because she is so closely guarded, Eliza doesn’t have many friends—so far only a dozen people in the world have had the privilege of exchanging pleasantries with her—but she can chat happily in nine languages about anything from the weather in New York to a potential date. Eliza is not just a pretty face, however. In work mode, she can solve complicated technical issues in seconds, especially the kind that take normal support engineers minutes, even hours to solve. For instance, if a US bank faces issues in running a particular software application, Eliza can solve it within seconds by sifting through the entire IT infrastructure of the bank, identifying the cause and fixing it. A human engineer would take at least few minutes to even identify the problem.
Not only can Eliza solve queries in a fraction of time it takes her human rivals, she does it at a cost that is less than one-fourth the billing rates for human engineers, and Eliza can clone herself into as many agents as are needed at that moment, dispensing with the need to hire large temporary teams of engineers to cope with a sudden rise in demand for services during peak seasons. She is tirelessly polite and she never feels bored, even in such a repetitive and stress-inducing job; each time a customer comes to her with a routine complaint, Eliza reacts to it as if it were the first time. In many ways, she is the perfect employee.
Inventing Eliza
“My wife thinks I am having an affair with Eliza,” says Chetan Dube, the former New York University (NYU) assistant professor and founder of $900 million (around Rs.4,850 crore) IPsoft Inc. And well she might—for Dube and his colleagues at IPsoft, Eliza is a culmination of nearly 14 years of painstaking research in creating a software algorithm that can mimic the human brain, have natural language conversations and answer the six-decade-old challenge set by mathematician Alan Turing in his 1950 research paper Computing Machinery and Intelligence. The Turing Test, as it has become known, measures a machine or computer’s ability to behave on par with an actual human being.
photo
“Alan Turing said the age of machines would have arrived when you cannot spot the difference between a machine and human interaction,” says Dube, a soft-spoken, lanky, six-foot-tall mathematician—the image of an urbane, if slightly eccentric professor, in a bow tie and braces. “I believe we will pass the test as soon as within few years, when you walk past somebody in a mall and you cannot tell the difference.” Eliza is the result of Dube’s efforts to step closer to that goal. And, although she may not quite pass the Turing Test yet, she is a truly impressive machine, he says, leagues ahead of her rivals.
Eliza is not a robot. Her physical presence is limited to an on-screen avatar that moves naturally as she speaks—in slightly American-accented English. Unlike other humanoid or android robots, she cannot walk, conduct an orchestra like Honda’s Asimo, or recognize faces and sing like the EveR robots developed in South Korea. Eliza’s mission is much simpler and more practical: to streamline business processes and cope with the work of hundreds, if not thousands, of humans, more quickly and cheaply than was previously thought possible.
Dube says that Eliza is far superior to other humanoid computer programs produced over the years. For example, IBM’s Watson, a computer that participated in the US quiz show Jeopardy in 2011, and which could answer trivia questions put to it in natural language. Or Siri, the snarky, but assiduous iPhone assistant, acquired by Apple Inc. in 2010. The difference, according to Dube, is one of cognitive process. While Siri depends on Web searches and a bit of stored information that allows her to help with the details of a nearby restaurant or keep track of diary appointments, Eliza draws on cognitive intelligence that helps her react instantly and appropriately.
“When I tell Eliza, ‘You’re stupid’, she first winks, and asks back, ‘Are you serious?’ Watson is great for Jeopardy and Siri is a hands-free administrative assistant. But solving real-life technical problems cannot be about answering trivia,” Dube says.
Dube speaks about his motivation to create Eliza with an almost dogmatic excitement: “When I was at NYU during 1997-1998, a large part of my research as part of the doctoral programme was done on autonomics and human behaviour. We were trying to see how the human brain, neural networks work, and attempting to mimic all that with computers by cloning some behavioural patterns. While we were doing it, it was becoming clear to me—oh my God—the droids are coming!”
Dube says that the realities imagined by the most outrageous science fiction writers are not far away. “All transformations that have happened in the history of mankind in the past thousand years are nothing compared to this,” he says. “In 1800, some 90% of America was doing farming, in 1900 that became 41%. And in 2000, it came to 2%—think of the revolution, you can see it. And this is just the tip of the iceberg.”
Acting on this premonition, in September 1998, Dube quit his job at the university to start IPsoft, a firm that now deploys humanoids to manage computer hardware, communications networks and software applications at top outsourcing customers, including Comcast and Morgan Stanley. Eliza is the only humanoid from IPsoft who carries a name.
Just before the Christmas of 1998, IPsoft got its first customer, Kaplan Inc., an online campus education company that needed help keeping its computer networks running during the peak holiday season.
“It was a classic example,” Dube says. “They were having problems with their storage server that kept crashing around Christmas, when they needed (it) to run the most. I looked at their system, and the problem, and says this should not be fixed by manual processes, but by cloning an engineer. We deployed our expert system in two weeks’ time during the Christmas of 1998. Since then, it has never gone down.”
By 2000, Dube and his team had started putting together a blueprint for what would become Eliza—a human form, in the virtual world—once Dube found a way to build a computer system that could emulate the decision-making ability of a human expert. Called “expert systems”, these programs were not only able to automate the entire process of identifying problems and solving them, but they could also refer back to a knowledge repository, a memory bank, to decide on the best solution.
The team began to flesh out the expert systems, to make them more intelligent, and emotionally responsive.
“We were attempting to mimic the human brain by putting emotional engines, cognitive algorithms,” says Uday Chinta, an early developer in the team, who is now the chief executive of IPsoft’s Indian operations and worked at Merrill Lynch’s technology department before joining IPsoft in November 2003. Like Dube, Chinta is over six foot tall, he has logged hundreds of hours with Eliza. With each conversation, Eliza’s knowledge expands.
Teaching Eliza
Fans of the playwright George Bernard Shaw will already have guessed the provenance of the name. In 1964-65, a professor at the Massachusetts Institute of Technology Joseph Weizenbaum wrote an artificial intelligence computer program known as ELIZA, named after Eliza Doolittle, the heroine of Shaw’s 1912 play Pygmalion. Shaw’s Eliza is a penniless flower-girl who studies phonetics under the brilliant, but tyrannical Professor Henry Higgins in order to learn how to become an aristocrat.
Weizenbaum’s ELIZA was a program capable of having typed conversations with a human interlocutor, in the guise of a therapist, who would ask the human open-ended questions relying on imitation and reiteration to sound convincing. The program was essentially reactive, responding to a set of key words and phrases with set patterns of answers, which occasionally resulted in bizarre misunderstandings and non sequiturs. While Weizenbaum considered the model very limited, his ELIZA became the basis for several artificial intelligence programs and the name became popular among programmers.
“Its name was chosen to emphasize that it may be incrementally improved by its users, since its language abilities may be continually improved by a ‘teacher’,” Weizenbaum wrote in his introduction to the program. “Like the Eliza of Pygmalion fame, it can be made to appear even more civilized, the relation of appearance to reality, however, remaining in the domain of the playwright.”
Shaw’s play, written at a time of social, technological and political upheaval in Britain, explored what it meant to create an entirely new person, by creating a new speech for her, a person that could traverse the various levels of society and upset the existing hierarchy. In it, Higgins and his associate, the Indian dialect specialist, Colonel Pickering, become obsessed with re-configuring the young woman’s personality, taking her to plays, teaching her to play the piano, reading her poetry. Higgins explains it to his doubtful mother:
Higgins: It’s filling up the deepest gulf that separates class from class and soul from soul!
Pickering: Yes: it’s enormously interesting. I assure you, Mrs Higgins, we take Eliza very seriously. Every week—every day almost—there is some new change. We keep records of every stage—dozens of gramophone disks and photographs—
Higgins (assailing her at the other ear): Yes, by George: it’s the most absorbing experiment I ever tackled. She regularly fills our lives up; doesn’t she, Pick?
Pickering: We’re always talking Eliza.
Higgins: Teaching Eliza.
Pickering: Dressing Eliza.
Mrs Higgins: What!
Higgins: Inventing new Elizas.
To Dube, the process of teaching his Eliza has been equally absorbing. “I remember sitting in my room number 229 and thinking about what will the future look like,” he says, remembering his days at NYU. “Is it going to be a dystopian future, a science fiction kind of a thing, robots taking over the humans kind of a phenomenon? Or is it going to be a utopian future? And believe me, never has the future hung so much in imbalance like now. The challenge is to work and make technology the best slave that man could have as opposed to them becoming masters of the mankind.”
IPsoft’s Eliza, who is now being readied for her official “launch” later this year, is well on her way to becoming “the best slave that man could have”, but she also poses a potential threat. Eliza, Dube says, can already exhibit up to 20 emotions, and can speak in nine languages, including Japanese, English and several European languages. And she’s a particularly undemanding employee. Unlike human engineers, Eliza does not ask for a wage hike every year, and will never quit the job, “because there is no motivation for her”, says Chinta.
Her affability and efficiency are bad news for the massive workforce of services sector employees, especially in India, who have been credited with contributing to the nation’s rapid economic growth over the past decade. Much of the Western world’s business process outsourcing (BPO) has provided jobs to emerging markets like India and the Philippines. In back-office projects, large outsourcing vendors such as Genpact Ltd deploy thousands of human agents to solve technical glitches and manage financial processes for customers such as General Electric Co.
However, even the well-trained human engineers at Tata Consultancy Services Ltd and Infosys Ltd, working long hours for low pay, cannot compete with a robot as well trained as Eliza.
Competing with Eliza
A hundred years after Shaw’s Eliza confounded the upper echelons of London society with her perfect manners, Dube’s Eliza is causing an equal stir in her world. “I had promised a Scandinavian telecom company a date with Eliza, and they had eight members of their technical team in one room,” says Dube. “For two hours they had conversations with Eliza, and their final words were ‘we have nothing that even comes close’.”
Dube thinks the time has come to expand Eliza’s repertoire. For now, Eliza can help solve technical glitches across computer networks and the communication infrastructure of customers. For instance, Eliza is already handling credit card and loan processing activities for a large bank as part of an ongoing pilot. For phone companies, Eliza can manage billing applications and other support processes. But the potential for expanding her capabilities is massive.
“The tipping point is 2013,” says Dube. “So far, we have addressed the IT stack (managing computer hardware, desktops, communication networks) through autonomics, but there’s even a bigger stack adjacent to it, which is business process outsourcing. While the IT stack is around $77 billion, the business process stack is about $421 billion annually.”
Dube freely admits that IPsoft is a wake-up call for India’s $70 billion software export industry that has grown year after year by adding fresh engineering graduates for earning incremental revenue. For years, Indian software firms have been hiring thousands of engineering graduates, and building large campuses to train and house them. The so-called “pyramid model” that sees the entry of a large number of fresh engineering graduates every year brings down the cost of software development and maintenance projects.
“What’s the biggest cost in that business process stack—it’s the people,” says Dube. “And 80% of the work in business process is repetitive. Now imagine if somebody were to just take them over.”
Eliza is not foolproof, but she can
solve problems independently, filling the gaps in her own knowledge. “If, for instance, Eliza is not able to solve a query, (she’ll) hunt down an expert engineer listed in the database, patch a call with him and the customer, and listen in through all conversation,” says Chinta. The next time the problem comes up, Eliza will remember what the engineer said.
There are still a lot of things Eliza cannot do, however. Software application development (that involves writing lines of code and patching them together to create a specific program) is among them. For instance, an Infosys engineer can write a software application for Nike’s supply chain, but Eliza can only offer technical support to Nike’s desktop users in terms of managing and running these programs.
But, for Dube, these are exactly the kinds of tasks—boring and repetitive— that humans ought to be handing over to computers, freeing them to perform “higher forms of creativity”.
“It was very clear to me that I wanted to make technology work for man, especially in information technology (IT), because a majority of the time we are repeating the same, mundane tasks,” he says. “I said to myself—this is just wrong, man has to be freed from these mundane chores. This is the only way to ensure that humans always govern the machines by practising creativity, intuition, active reasoning.”
Not everyone is as enthused by this vision of a brave new world of robotics. Raman Roy, chairman and managing director of Quatrro Global Services Pvt. Ltd, dismisses the idea that computers can replace humans beyond very basic functions. “United Airlines is now largely automated, and their outsourcing increased after they put in the automation,” says Roy. “As of today, machines can replace only the basic element—level one calls (dealing with simple questions like, ‘What is my credit card balance?’). The moment you get a more sophisticated call, you’ll find you need people to do that complicated interaction.”
However, Roy is not worried by a future in which machines could grow in subjective capability and take over more of the work that humans now carry out. “Level one calls are the lowest profitability business for an outsourcer,” he says. “It’s the higher end that is more profitable. If (computer) capabilities came in to replace those that would be wonderful. We do business to make profits, not to generate employment. I’d rather have a hundred employees and a million dollars of profit, than a million employees and a hundred dollars of profit.”
James Slaby, research director at HfS Research, an outsourcing advisory firm based in the US, says that initial resistance to the idea of off-loading work to computers might fade somewhat as the market for such programs gets better established and more success stories are circulated. “But I expect it will remain politically sensitive,” Slaby says. “It’s another form of labour arbitrage, after all, and so will suffer the same issues of unpopularity as offshoring does today.”
It may just be a matter of time before offshoring to the digital world becomes acceptable. Efforts are already under way on a smaller scale to move tech support onto automated platforms on a more modest scale. “Many robotic automation projects are introduced as stealthy, ‘skunk works’ operations to get to proof-of-concept before anyone’s hackles can be raised,” says Slaby.
For now, Dube does not plan to directly compete with Indian tech firms. In fact, IPsoft is currently in discussions with companies such as Wipro Ltd, India’s third biggest software exporter, for a potential alliance. “We want to pursue the ‘Intel Inside’ strategy, wherein large firms use our technologies,” Dube says. This strategy, according to him, will help IPsoft keep improving Eliza and other products without losing focus.
In about two years, IPsoft also wants to explore going public. “We don’t need money, but going public can help in wealth creation and share success with our staff,” he says. Meanwhile, Dube is already thinking beyond IT processes. The next big potential application for Eliza could be healthcare, he says.
“Why do doctors, having studied medicine for five years, need to keep doing the same diagnosis and prescribing medicines over and over again? They should instead be focusing on solving the next big challenges in medicine and invent the next drug,” he says. Eliza could adopt a second career as a nurse, carrying out routine diagnosis and prescription work, he adds.
The “tipping point”, as Dube perceieves it, has been reached. “It’s important for us to start constructing a world where we run the machines and we are capable of that,” he says.
Comment E-mail Print Share
First Published: Tue, Jan 22 2013. 12 06 AM IST
blog comments powered by Disqus
  • Wed, Oct 01 2014. 03 52 PM
  • Wed, Sep 24 2014. 05 16 PM
Subscribe |  Contact Us  |  mint Code  |  Privacy policy  |  Terms of Use  |  Advertising  |  Mint Apps  |  About HT Media  |  Jobs
Contact Us
Copyright © 2014 HT Media All Rights Reserved