Machine victory: IBM's "Watson" defeated the champions of champions on Jeopardy

IBM researchers in Israel, David Carmel and Dafna Sheinfeld, participated in the development of the system

At the end of a three-day duel, an IBM supercomputer known as Watson defeated the two champions of the American television game Jeopardy.
The game proved not only the ability of a computer to beat humans in the trovia game - but also the ability of machines to answer questions in a similar way to how humans express themselves: in complete sentences, and in eloquent speech. IBM explains that this achievement lays the foundation for a new form of communication between a person and a computer.
At the conclusion of the last part of the competition, which lasted over three evenings, Watson accumulated $77,147 for correct answers, compared to $21,600 for Brad Rutter and $24,000 for Ken Jennings.
Rutter and Jennings hold the winning record in terms of number of games and the record of financial winnings by humans in this game, which is very popular in the USA.
The first evening of the recording, on Monday this week, was characterized by a fair fight. Watson and Rutter finished the evening with $5,000 each, and Jennings with $2,000. On Tuesday, the IBM computer began to show muscle and create a gap at the top: it ended the evening with 35,734 dollars, compared to 10,400 for Rutter and 4,800 for Jennings.
In the last stage, on Wednesday, the IBM system continued to widen the gap, not only thanks to the vast knowledge base on which it relies, but also with the help of dedicated algorithms that were integrated into it, which allow it to make the best bet on correct answers. Thus, for example, in the daily double question - a special question that before its presentation the competitor is allowed to determine any bet amount himself - the computer decided to bet on $2,127, while evoking waves of laughter from the audience in the studio, who are used to watching competitors who punch rounder numbers.
The head of Watson's development project, Dr. David Procchi, explained after the game that even if such computer-generated numbers may appear to humans to be completely random, they are the result of a complex consideration of the computer's degree of confidence in the particular category of questions it is dealing with. At the same time, the computer evaluates its situation within the game, how far it lags behind other competitors or ahead of them, and how much money can still be earned or lost. In the end, numbers are obtained with maximum accuracy.
Watson acted incredibly smoothly in most cases - but not without making mistakes and failing here and there with errors that provoked laughter from the audience. So, for example, when asked about the name of an airport in New York named after a hero of the Second World War, the computer replied "Toronto" - the name of a city in Canada. In another case, the computer insisted on offering a wrong answer even when another competitor had already offered it, and after his answer was not accepted.
Researchers from Israel
Watson's development process lasted four years, and was carried out as a joint project of a number of IBM research laboratories around the world, including the IBM research laboratory in Haifa.
The researchers at the IBM research laboratory in Haifa focused on various aspects of advanced search capabilities, expanding the framework of computerized queries, and defining associative relationships that serve Watson's ability to jump between different contents and terms, in order to understand the context in which they are presented.

, who were partners in Watson's international development effort in IBM laboratories around the world,
And the leaders of the project in Israel stated that "Jeopardy is a great challenge." The competition presents a need for extensive knowledge, where each of the clues may include relevant or irrelevant knowledge for the final search object.
Humans are able to distinguish in the blink of an eye between what is relevant to them and what is not - but computers cannot know this. They must consider all the data in parallel, in order to reach an understanding."
The technology that led to victory
IBM's Watson supercomputer combines 2,880 IBM POWER 7 processor cores, and 15 terabytes of memory equivalent to the contents of about a million printed books, which are at the service of a special software application that allows the computer to communicate with the environment in a completely natural language. Watson understands the question presented to him in speech, analyzes it, produces the answer - and utters it in a synthesized voice. This computer, which continues the lineage of "Deep Blue" that beat world chess champion Garry Kasparov in 1997, is much more than a display of technological muscle in a televised game. This computing power can be harnessed for climate and weather forecasting, medical research, or other applications.
IBM explains that the design of the computer and its unique structure herald a new era in the field of information technology - where machines are able to learn, understand, analyze and respond to large amounts of data in real time. Analytical computing capabilities, such as those of Watson, are expected to bring about changes in the business world in particular and society in general.
Watson does not make decisions, and does not replace people. He helps them, and acts according to their instructions. It processes large amounts of data and presents the information required to make decisions as well as recommendations about those decisions. The technological breakthrough in the field of the interface between the computer and the people around it, and its ability to communicate in spoken language, make it possible to promote the use of computers in new and wider fields than ever before. So, for example, such a computer could help a doctor get the information required for a diagnosis, help citizens who need information about government services get it conveniently and easily, and provide information to tourists visiting various sites.
The challenge that Watson faced within the framework of television entertainment is twofold, and significantly greater than that faced by "Deep Blue", which was built for the purpose of a chess game, conducted within the framework of defined rules and in a single content world. Watson was able to cope with the game encompassing infinite content areas, and to be able to deal with different types of clues. Beyond that, he was able time and time again to understand the meaning of questions presented in spoken language, and to recognize concepts that may be implied from these questions in an associative manner.
The computing power required for these needs is provided by a cluster configuration that includes ten IBM Power servers, each of them the size of a large refrigerator, on which a complex language analysis algorithm operates - and an extremely large database.
On the way to each answer, the computer performed five key steps. First - analyzing the question: Watson breaks down the question and examines the relationship between the words that make them up. After that, it runs a search in the database: according to the rules of the game, the competing computer is not connected to the Internet, in order not to give it an advantage over the human players. The search is carried out on a huge database of data and documents collected at a later date from the Internet, stored in the disk array that serves Watson - and builds a sub-database of potential sources for the answer. A number of possible answers are defined by different algorithms that scan the documents located in the previous step. The possible answers are evaluated against each other with the help of algorithms that examine people's names, place names and geographical contexts, dates, and so on. The summary of the answer and the final decision occur less than three seconds after completing the question - otherwise the computer has no chance of beating the people standing in front of it.
The team of IBM scientists behind the development of Watson faced a research challenge that has not yet been fully answered: building a computerized system that can handle the human ability to answer questions presented to it in normal spoken language - with speed, accuracy and a level of confidence equivalent to that of humans. The Jeopardy game format presents the ultimate challenge in this field: the clues the contestant receives on the way to presenting the answer require the analysis of hidden meaning in the text, irony, verbal puzzles and other complex structures that the human brain knows how to deal with - while traditional computer systems are not prepared to handle these types of content and questions.
Jeopardy game rules require the players to formulate their answers in the format of questions - and not as a direct answer. Thus, for example, to a question such as "He heads the government with the largest number of ministers in the history of his country, having previously served as his country's ambassador to the United Nations, as minister of finance, as prime minister - and lost in previous elections" instead of the usual answer which would be "Benyamin Netanyahu" - the contestant should answer : "Who is Binyamin Netanyahu", and beat his opponents to the game in every round.
Watson, which operates, as mentioned, on a server based on IBM's POWER7 processors, is a research breakthrough in the field of computerized spoken response to spoken language questions. The server, which combines a large number of processors in parallel processing, is specially adapted to handle a large number of tasks that must be performed quickly in order to analyze the complex sentences in which the clues are formulated in the program - and to respond in the format required by the game's holes. The system combines a series of unique technologies to handle simultaneous tasks and real-time data analysis.
.
Much more serious goals
Although the press reports are expected to focus on the victory of the machine over the human in Jeopardy, IBM developed Watson for much more serious purposes. The company says that participating in the game is part of the process of developing new capabilities of computers that know how to analyze complex information, perform complicated analytical applications in the field of medicine - for example, and communicate directly in natural, spoken language. In the end, IBM predicts that it will be possible to run computer applications that communicate with the environment in natural language and not in lines of code or defined fields, and to analyze large amounts of unstructured information with them.