April 19, 2014

EMU professor says Jeopardy/IBM experiment moves computers one step closer to 'holy grail' of artificial intelligence

by Ward Mullens, Published February 14, 2011

Bookmark and Share

YPSILANTI - QUESTION: "Not the chocolate coated candy in a wrapper, but a rapper from Detroit."

ANSWER: Who is Eminem?

This type of pop-culture infused wordplay has made the long-running game show "Jeopardy" popular.

It is the same language ambiguity that has long challenged the development of artificial intelligence - until now.

 Jeopardy will air three episodes (Feb. 14-16) that pits an IBM built "super computer" named Watson against two former Jeopardy champions.

"Correctly formulating questions to given answers (the Jeopardy game) involves many linguistic gymnastics for a computer," said Bill Sverdlik, professor of computer science at Eastern Michigan University. "Jeopardy answer/questions are subject to ambiguities, multiple meanings and even cross language problems."

Sverdlik, whose area of expertise is artificial intelligence, will be watching the event.

The IBM Challenge is much more than a computer playing a game, Sverdlik said.

"Many of the previous computer/human challenges involved complete knowledge games," Sverdlik said. "Chess and checkers are two examples of such games. Complete knowledge means that both adversaries have total knowledge of the current state of the game, unlike the card game bridge or poker, where cards are hidden from players. Winning a complete knowledge game involves many strategies, but perhaps the most important one is the ability to generate positions several moves into the future. This is easy for a computer to do.

WATSON represents a true advance in the area of natural language processing (NLP), the ability of a machine to interact with humans in a natural language like English or Spanish," said Sverdlik.  "We've all had the experience of talking to computers on an automated phone line and some automobiles now have systems that accept voice commands. But these systems have very limited vocabulary and only can process a very small set of commands. WATSON brings us one step closer to a natural language interface with a machine."

And that knowledge can lead to a better understanding of how humans process information.

"WATSON is attempting to automate a cognitive process; doing so hopefully helps us better understand our own cognitive processes," said Sverdlik. "This is the Holy Grail of Artificial Intelligence (AI) research: developing computer algorithms that help us understand what it is that we humans do. Some previous success stories in the AI community include computer vision, object manipulation and path planning. These areas, while well developed already, are still experiencing wonderful new results. WATSON represents a major leap in yet another cognitive process."

Even if you don't know anything about AI, this experiment, aside from its entertainment value, does provide a peek into the future.

"Any machine/human competition grabs at the imagination. Robbie the Robot, HAL, I Robot - they raise the possibilities of machines endowed with human qualities and  human frailties and challenge us to question what exactly separates us from a machine. It's fun to think about," said Sverdlik.

"This is a coming technology. Someday soon, we all will be conversing with a computer on matters more significant than dialing a phone number or changing the radio station."

 

###

 

Ward Mullens

Contact: