Annals of Artificial Intelligence: Man vs. Machine, IBM's Watson.

[Prometheus]

IBM's Supercomputer 'Watson' Wins Jeopardy Practice Round

By Ian Paul, PCWorld

 

It will be man vs. machine part 2 on Jeopardy next month, and if Thursday's friendly match up was a sign of things to come, the humans are in trouble. Former Jeopardy champions Ken Jennings and Brad Rutter lost by $1000 to IBM's Watson DeepQA-based supercomputer during a three-category Jeopardy practice round Thursday night. The trio will officially square off for a $1 million grand prize during two Jeopardy matches that will air February 14-16.

 

Who is Watson?

 

Watson is IBM's latest supercomputer based on the company's DeepQA software, which combines natural language processing, machine learning and information retrieval. The device is packing 15 terabytes RAM, about 2,880 processor cores that can perform 80 trillion operations a second, and is the size of 10 refrigerators according to Wired. Watson will have to rely on its self-contained databases for answers, and won't be hooked up to the Internet during the Jeopardy challenge.

 

On stage, a computer monitor will be the only part of Watson people will see, and just like his competitors Watson will have to trigger a buzzer before it can answer a question. When Alex Trebek calls on Watson it will answer in a computer generated voice that is eerily reminiscent of HAL from 2001: A Space Odyssey.

 

Many experts are saying the challenges Watson had to overcome to play Jeopardy are far more complex than the challenges IBM's chess-playing computer, Deep Blue, faced when it defeated chess grandmaster Gary Kasparov in 1997.

 

In Chess, there are only so many possible moves you can make to respond to your opponent, and all your moves are defined within a clear set of rules.

 

Jeopardy, on the other hand, will require Watson to handle a much bigger challenge: decoding human language with all its nuances, implied meanings and colloquial expressions. As David Ferrucci, IBM's lead manager for Watson, recently told IDG News, "Natural language processing is so difficult because of the many different ways the same information can be expressed."

 

That's why the mere fact that Watson is able to compete in Jeopardy--let alone win--is considered a significant milestone for artificial intelligence. If you want to learn more, check out IDG News' great write up on the challenges Watson faces next month.

 

Also, check out this Engadget video from Thursday night's practice round:

 

[Prometheus]

Is there something unique about human intelligence or, for that matter, consciousness? Or do such phenomena merely represent a series of complex computations and calculations? Aren't humans, after all, very complex machines-- machines, as science tells us, with no inner ghosts or souls? No one seriously believes that there is more to our minds than our brains, which is a physical thing (machine) itself. What does it mean to 'think', or to have 'intelligence'? Perhaps the Turing test is ever elusive. Nomads, I ween, remember the reactions that greeted the loss of chess grandmaster Kasparov to IBM's Deep Blue. However, as complex and cognitively strenuous as chess seems, it exhibits nowhere near the complexity required to apprehend and analyze the intricacies of language. Should machine outsmart and outperform Man next week, it will be another milestone in AI technology. HAL 9000- some paranoid observers remark - isn't far behind. That's a tad alarmist and histrionic in my view. In any event, I think there's still a long way to go, but my hope is our intuitions of uniqueness, our anthropocentric arrogance, will suffer another blow.

[Somalina]

 

Almost beat Watson, just needed more time. Darn it!

[Somalina]

 

Watson in action!

[Prometheus]

Ha! Well, the computer made some obvious blunders, but the game is far from over. It's estimated that IBM spent close to 2 billion on this project. But I take such marketing-freindly rumors with a grain of salt. In any event, I've always found the analogy of mindless algorithms and mindless neurons ineluctably seductive. With enough mindless algorithms, you can produce seemingly intelligent behavior. If your mind comprises billions of neurons, and if each neuron is a little more than a mindless robot, how does it produce higher-level functions such as intelligence. For those who are interested, click on this link to follow a lively debate between a neuroscientist, computer scientist, and a philosopher. The authors pretty much cover the well-known controversies about AI, consciousness, and brain science.

[Kulmiye]

its actually worthy of a project as it shows a remarkable human progress, in addition it shows what the human mind is cable of with technology.

[Cara.]

Did they have to make his voice sound vaguely HAL9000-ish, or am I being paranoid?

[Prometheus]

^ Lol. You're not being paranoid. The voice simulator was not chosen at random.

[Prometheus]

John Seabrook writes " Is Watson a great breakthrough in science—a Sputnik moment—or an elaborate parlor trick? When I asked Steven Pinker this question, he responded":

 

I don’t rule out the possibility that some components of Watson could both provide insight into human cognition and lay the groundwork for more sophisticated artificial intelligence applications, such as natural language processing (the fancy term for understanding human languages like English, as opposed to computer languages). On the other hand, when a system is designed to meet a highly specific challenge like playing Jeopardy, and one where the reputations of the designers are on the line, there will be enormous pressure to tailor the system to succeeding at that challenge by any means whatsoever, including kludges that are specific to the rather peculiar requirements of the game of Jeopardy.

Pinker went on:

 

The real problem is that we may never know. It will depend on whether the I.B.M. team divulges the methods in technical publications or keeps them as trade secrets. In the golden years of A.I. (1960s and 1970s), there was a lot of back-and-forth between academia and industry. Labs at Xerox, I.B.M., B.B. & N., and A.T. & T. were among the best research departments in the world, and people and ideas flowed in and out of them. Then A.T. & T. lost the free money from its telephone monopoly, and the other companies realized that their openness was just helping their competitors (e.g., the Macintosh GUI, which was basically stolen from Xerox PARC), and they forced their scientists to work on applied projects and kept the details out of the public domain. The result is that A.I. has become disengaged from cognitive science, the old A.I./philosophy gurus like Minsky, Papert, Simon, and Schank have not been replaced, and questions like yours may be impossible to answer, if I.B.M., as seems likely, will keep the specs secret. That is, we won’t be able to know how much of the program’s success to attribute to humanlike or superhuman intelligence, and how much to Jeopardy-specific hacks.

Marvin Minsky, sometimes called the father of AI had this to say:

 

Watson program may turn out to be a major advance, because unlike most previous AI projects, it does not depend mainly on a single technique, such as reinforcement learning [learning via reward and punishment], or simulated evolution ... but tries to combine multiple methods,"

.

 

Source: New Yorker and National Geographic