Watson Wins Jeopardy!—6 Artificial Intelligence Milestones

Games Score Serious Advances

Photograph from Science & Society Picture Library, Getty Images

In 1951 British computer scientist Christopher Strachey wrote one of the first machine game-learning programs for checkers and ran it on the Ferranti Mark I computer (pictured). Since then, games have helped computer scientists to test their machines and programs.

Six years later, for example, IBM researcher Arthur Samuel created a similar program that was good enough to actually beat a human checkers expert—beginning the man-vs.-machine tradition so publicly continued with Watson on Jeopardy!

But despite Watson's publicity, there's a shift within the AI research community away from "prototype spectacles" like the Watson Jeopardy! match, said Dan Klein, a computer scientist at the University of California, Berkeley.

"In recent decades AI has become successful enough that that's not where the victories are," Klein said. "The victories are in these commercial products that are successful and impact life."

For example, every time you use a search engine or a spam filter, or use the motion-detection abilities of your Microsoft Kinect, Klein said, "that's AI."

(Video: Jeopardy! host Alex Trebek hosts the championship round of the most recent National Geographic Bee.)

Published February 17, 2011

Watson's Ancestral "Brain" Born

Image by David M. Phillips, Visuals Unlimited/Getty Images

Inspired by research in brain science, MIT researchers Belmont Farley and Wesley Clark, using the Memory Test Computer, created one of the first successful simulation of networked brain cells, or neurons (pictured), in 1954—enabling the machine to "think" and "learn" by recognizing patterns.

"The short history of AI is that in the 1950s, computers could do these low-level things like multiplying and sorting, and they could do them better than humans," Klein said.

"So researchers thought the next step would be high-level things like chess or writing poetry or translating one language to another"—or, presumably, trouncing humans on TV game shows like Jeopardy!

(Related: "Artificial Intelligence to Boost Space-Probe Efficiency.")

Published February 17, 2011

Machines Find Their Voices

Photograph by Roy Stevens, Time & Life Pictures/Getty Images

In 1966, MIT Joseph Weizenbaum created one of the earliest natural-language processing programs, called ELIZA.

ELIZA was capable of participating in simple conversations with humans. For example:

ELIZA: How do you do. Please state your problem.
HUMAN: I spend too much time with computers.
ELIZA: Why do you mention computers?
HUMAN:  You are one.
ELIZA: Does it please you to believe I am one?
HUMAN: Yes.
ELIZA: You seem to be quite positive.
HUMAN:  I suppose I am.

The main purpose of ELIZA wasn't to demonstrate artificial intelligence, according to artificial intelligence expert and Stanford University computer science professor emeritus Nils Nilsson.

"Weizenbaum didn't really write [ELIZA] to answer questions," Nilsson said. "He just wanted to show how easily humans could be fooled into thinking that the program was intelligent."

Today, the Watson computer’s success at feigning intelligence stems largely from its a natural-language processing program, DeepQA—the computer’s “most useful software,” according to PC World. DeepQA enables Watson to understand typical Jeopardy! questions and respond, well, naturally.

Richard Doherty, research director at the Envisioneering Group technology consulting firm, told Computerworld that "this is the largest [computing] advancement in decades. This isn't an iPad. To reach [a computer] conversationally and have it respond with knowledgeable answers is a sea change in computing."

(See why some think artificial intelligence may offer humans electronic immortality.)

Source of ELIZA transcript: The Quest for Artificial Intelligenceby Nils J. Nilsson.

Published February 17, 2011

Computers Learn to Crawl

Photograph by Ralph Crane, Time & Life Pictures/Getty Images

The late Charlie Rosen, artificial intelligence pioneer and computer scientist at SRI International, poses for a picture with Shakey the Robot in the late 1960s.

Like a primitive, vacuumless Roomba, Shakey "was the first robot program that could perceive its environment using vision and touch sensors, make a plan to achieve a goal, execute the plan, monitor its execution, and recover from execution errors," Stanford's Nilsson told National Geographic News.

The robot's moniker came from its tendency to shake when the machine came to an abrupt stop. Shakey had an on-board television camera, a laser range finder for sensing distance from walls and objects, and whisker-like detectors for identifying obstacles.

According to Nilsson, the search algorithm used by Shakey is still used today to find directions by programs such as Google Maps.

Published February 17, 2011

IBM’s Deep Blue Defeats Kasparov

Photograph by Adam Nadel, AP

Chess grand master Gary Kasparov (left) plots against IBM's Deep Blue chess computer while Chung-Jen Tan, manager of the Deep Blue project, looks on in 1997. Deep Blue eventually beat Kasparov in the six-game rematch—the Russian mastermind had beaten the computer a year earlier.

In the early years of  artificial intelligence research, computer scientists aimed to replicate human intelligence with their machines, but the definition of AI has gradually changed over the years as the field has grown and split into various subspecialties.

Nowadays some researchers define artificial intelligence as systems that efficiently perform tasks assigned by humans. According to this view, humans today are surrounded by AI. We use it every time we conduct an Web search, use spam filters, or activate the face-recognition features in our digital cameras.

Scientists have realized "there's no magic bullet for human-level intelligence," UC Berkeley's Klein said.

"We've learned that we won't get there in one simple step. … Instead, intelligence is a long-term goal, and researchers are pushing forward on many aspects of it" by solving increasingly harder problems all the time.

Stanford's Nilsson added that the initial dream of artificial intelligence research—to build a machine that can do the cognitive things that humans do—is very much alive.

"That was the original definition" of AI,  Nilsson said. "And it remains the goal for a lot of people."

Published February 17, 2011