A robot has been teaching grad students for 5 months... and NONE of them realized

  • Jill Watson has been a TA at Georgia Institute of Technology for months
  • She is in fact a robot powered by IBM's Watson system 
  • But students said her emails seemed casual, personal, and to-the-point 
  • Researchers said TAs spend too long responding to email questions 

By Mia De Graaf For Dailymail.com and Cheyenne Macdonald For Dailymail.com

Published: | Updated:

There are some human attributes robots could never replace - or at least that's what you might hope.

But one university has brought that into question by replacing one of their teaching assistants with a machine.

And none of the students realized.

Scroll down for video 

Students at the Georgia Institute of Technology had no idea one of their TAs, Jill Watson, was a robot

Students at the Georgia Institute of Technology had no idea one of their TAs, Jill Watson, was a robot

DID STUDENTS SPOT THE BOT? 

Student Tyson Bailey began to wonder if Jill was a computer and posted his suspicions on Piazza.

'We were taking an AI course, so I had to imagine that it was possible there might be an AI lurking around,' said Bailey, who lives in Albuquerque, New Mexico. 

'Then again, I asked Dr. Goel if he was a computer in one of my first email interactions with him. 

'I think it's a great idea and hope that they continue to improve it.'

Jill Watson, an IBM-designed bot, has been helping graduate students at Georgia Institute of Technology solve problems with their design projects since January.

Responding to questions over email and posted on forums, Jill had a casual, colloquial tone, and was able to offer nuanced and accurate responses within minutes.

Her replies included 'yep!' and 'we'd love to'.

The students had no idea until they were told - and many were shocked.

Tyson Bailey began to wonder if Jill was a computer and posted his suspicions on Piazza.

'We were taking an AI course, so I had to imagine that it was possible there might be an AI lurking around,' said Bailey, who lives in Albuquerque, New Mexico. 

'Then again, I asked Dr. Goel if he was a computer in one of my first email interactions with him. I think it's a great idea and hope that they continue to improve it.'

'It seemed very much like a normal conversation with a human being,' Jennifer Gavin, one of the students, told the Wall Street Journal.

Another student, Petr Bela, told the newspaper: 'Just when I wanted to nominate Jill Watson as an outstanding TA.'

The bot was named Jill Watson after the IBM Watson analytics system that all her responses come from - essentially her brain.

She was trained by Georgia Tech researchers before being thrown into the mix with nine other TAs.

Some students were suspicious at how swiftly she responded. 

And once she used the word 'design' instead of 'project'. 

But none actually suspected she was a bot.

In fact, some looked her up online and found LinkedIn and Facebook accounts that could correspond with their prompt TA.

And they said many of the TAs are sharp, impersonal, and quick to respond anyway.

The experiment sent shockwaves through the tech industry, with many questioning the ethics of covertly inserting artificial intelligence into real-life situations.

But Ashok Goel, the computer science professor who designed the project, insisted it was a worthwhile experiment - and Jill performed a necessary task.

HOW THE SAME BOT BEAT HUMANS AT JEOPARDY

In February 2011, Watson appeared alongside two other contestants to compete for the cash prize.

During the show, clues are given to contestants that 'require analysis and understanding of subtle meaning, irony, riddles and other language complexities' that humans can perform naturally but computers, traditionally, do not.

Watson had to be programmed to make decisions and conclusions in this way by a team of experts at IBM.

Watson was given clues as electronic texts, as they were also asked to the human contestants.

Former JEOPARDY! Champion contestants during the final day of sparring sessions against Watson, at IBM's TJ Watson Research Center, Yorktown Heights, NY.

Former JEOPARDY! Champion contestants during the final day of sparring sessions against Watson, at IBM's TJ Watson Research Center, Yorktown Heights, NY.

The bot then parsed the clues into different keywords and sentence fragments in order to find 'statistically related phrases'.

The more algorithms that find the same answer independently the more likely Watson was to be correct.

Once Watson had a number of potential solutions, it was able to check against its database to make sure the answers made sense.

Watson then evaluated the response and determined whether to virtually press the buzzer.

The bot would then speak with an electronic voice synthesized from recordings that actor Jeff Woodman made for an IBM text-to-speech program in 2004.

Students posted upwards of 10,000 messages on forums, taking up professors' time with routine responses.

'The world is full of online classes, and they're plagued with low retention rates,' Goel said. 

'One of the main reasons many students drop out is because they don't receive enough teaching support. 

'We created Jill as a way to provide faster answers and feedback.' 

'One of the secrets of online classes is that the number of questions increases if you have more students, but the number of questions doesn't really go up,' Goel said. 

'Students tend to ask the same questions over and over again.'

Watson's cognitive computing system can be run on a single Power 750 server using Linux, which turns it from the size of a master bedroom to the size of four pizza boxes.

The bot has helped medical research teams diagnose illnesses in patients.

The bot was named Jill Watson after the IBM Watson analytics system that all her responses come from

The bot was named Jill Watson after the IBM Watson analytics system that all her responses come from

And in 2013 it took on the role of customer service manager.

Companies were be able to sign up to IBM's service and its customers could then ring a helpline and complain or get help from the Question Answering (QA) machine.

According to Goel, Jill Watson's capabilities were far more nuanced than that. She answers only to questions that she is 97 per cent certain of the answer to. 

One of the first high-profile tests of Watson's capabilities came during a game of Jeopardy, the televised quiz show. 

In February 2011, Watson appeared alongside two other contestants to compete for the cash prize.

During the show, clues are given to contestants that 'require analysis and understanding of subtle meaning, irony, riddles and other language complexities' that humans can perform naturally but computers, traditionally, do not.

IS YOUR JOB UNDER THREAT FROM ARTIFICIAL INTELLIGENCE? 

Claims made by an expert in artificial intelligence predict that in less than five years, office jobs will disappear completely to the point where machines will replace humans.

The idea that robots will one day be able to do all low-skilled jobs is not new, but Andrew Anderson from artificial intelligence company, Celaton, said the pace of advance is much faster than originally thought.

AI, for example, can carry out labour intensive clerical tasks quickly and automatically, while the latest models are also capable of making decisions traditionally made by humans.

'The fact that a machine can not only carry out these tasks, but constantly learn how to do it better and faster, means clerical workers are no longer needed in the vast quantities they once were,' Mr Anderson said.

For example, a machine can recognize duplicate insurance claims by knowing it has seen a phone number or an address before. 

Watson had to be programmed to make decisions and conclusions in this way by a team of experts at IBM.

Watson was given clues as electronic texts, as they were also asked to the human contestants.

The bot then parsed the clues into different keywords and sentence fragments in order to find 'statistically related phrases'.

The more algorithms that find the same answer independently the more likely Watson was to be correct.

Once Watson had a number of potential solutions, it was able to check against its database to make sure the answers made sense.

Watson then evaluated the response and determined whether to virtually press the buzzer.

The bot would then speak with an electronic voice synthesized from recordings that actor Jeff Woodman made for an IBM text-to-speech program in 2004.

 

Robots

Most watched News videos