Dr. Watson – as the supercomputer from IBM is known – seems to be a very special computer, not withstanding that it contains 90 servers with the computing power of 2,880 computers. Watson – the computer is named after the founder of IBM Thomas J. Watson – is a computer system that understands natural human language (in this case English), analyses words and context, processes information quickly and accurately and answers questions in a natural language. What it is capable of doing was demonstrated by it in February this year, most notably in the U.S. television quiz show “Jeopardy“. Anyone who wants to succeed in this cult show has to have a profound general knowledge, very quick reactions, to be capable of associative thinking, to have a fair measure of reliable “gut feeling” and a very good feel for language – and of course, to be capable of learning. This quiz is for a computer – even more than a chess game – the ultimate challenge, because the instructions given in the game include puns, hidden meanings, irony, puzzles, and similar semantic subtleties in which the human brain has been until now far superior to computers. To serve as a reminder: back in 1997 an IBM chess computer (Deep Blue) managed to defeat the then world chess champion Garry Kasparov. But even chess at the highest level is for a computer less challenging than this game called “Jeopardy.”
Dr. Watson beats Jeopardy champions
That he was raised for the challenge of “Jeopardy” Dr. Watson has impressively shown, as he at the beginning of this year competed against two champions – and beat both. In order to compete in this game against leading human candidates, a computer must be able to answer about 70 percent of the questions with an accuracy of over 80 percent within a maximum three seconds. Dr Watson works with the so-called “DeepQA” technology from IBM. It enables a new form of analysis in which several thousand tasks are processed simultaneously in seconds in order to correctly answer questions in natural language. According to the company, Watson makes use of over 200 million pages of knowledge in human-language form (equivalent to approximately one million books having been read), including medical knowledge. The Watson-Project at IBM is managed by researcher Dr. David Ferrucci.
Watson does Medicine
In collaboration with IBM and Nuance, the leading provider of software for voice recognition, the scientists at the University of Maryland (School of Medicine) in Baltimore and the Columbia University Medical Center (New York) now want to use Dr. Watson in medicine. “The body of medical knowledge doubles presently within a few years,” states Janet Dillione, head of health care markets in Nuance, in the journal Technology Review (TR). She believes that no human brain is capable of even having an approximation of all this information at hand. Unlike other health IT systems, Watson is also the first computer that can comprehensibly process medical information in the form of voice recordings, notes and articles, Dillione told TR. IBM has, according to the American Medical News, now entered into partnerships with eight universities in order to feed the supercomputer with medical information and to find out how doctors could best handle the system.
No doctor-substitute, but an assistant
According to radiologist and nuclear medicine professor Eliot L. Siegel who heads the project at the School of Medicine, Dr. Watson could lead to a renaissance of artificial intelligence in medicine. The supercomputer is certainly no substitute for doctors, but could as a kind of assistant not only gather huge amounts of data but organise and prepare it according to relevance, as well as, for instance, make several suggestions about a diagnosis. Diagnosis, therapy and safety of a treatment could improve and costs could be reduced. Dr. Watson could therefore be a valuable assistant who, so to speak, “sits” alongside the doctor and opposite the patient and delivers information in real time. In a specific treatment situation, Dr. Watson could not only provide the latest study data on any drug therapy, but also individual information about laboratory results such as actual genomic data. Dr. Watson could permit, says Siegel, a truly “personalized medicine“. The nephrologist Professor Herbert Chase of Columbia University New York, who collaborated on the development of the supercomputer, sees it the same way. No differently from Siegel, his colleague Dr. Nancy Knight looks on the super computer as a potentially valuable assistants to doctors, but not as a replacement. “The doctor will always be the one who ulimately decides. We do not want any kind of “HAL” in medicine”. (HAL was the fictional super computer in the Kubrick film “2001: A Space Odyssey” which became very neurotic and turned off the life support systems of three astronauts when he learned, through observation of lip movements, that the astronauts were considering switching him off).
Skirmishes between specialists foreseeable
According to information from IBM and Nuance, a commercially viable version of Watson could be available within about one and a half to two years. Before full integration into medicine however, according to Siegel’s assessment, several years will pass. First, the two companies intend to find out which medical data Watson actually needs. The next task will then be to prepare the data properly. “This will be extremely difficult, as medical texts often contain ambiguous acronyms, abbreviations, tables and bulleted lists and grammatically unusual phrasings”, explains Stephane Meystre, bioinformatician at the University of Utah, to TR. It should be just as difficult for Watson to follow doctor-patient conversations. Speed is not the main problem in this case, but rather accuracy. Whoever wants to use Dr. Watson will, therefore, have to be trained, assistant professor Rohit J. Kate, who works as an expert in speech processing at the University of Wisconsin in Milwaukee, cites. Before Watson is able to chat with specialist doctors and nurses, at least ten years will have passed by, estimates Kate. Ferrucci also stresses: “Our work for Jeopardy sets us in the right direction, but we have not arrived yet. Intelligent conversations with machines, as can be watched in an episode of Star Trek, are therefore still part of the future.”
To err is not only human
Not all experts however are enthusiastic about Dr. Watson. “Solving tricky quiz questions in Jeopardy is not the same as solving medical problems,” says Professor Edward Shortliffe, president of the American Medical Informatics Assn. Sceptical of course are also potential competitors of IBM such as Jason Maude, CEO and cofounder of the company Isabel Healthcare which he developed about ten years ago, “Isabel” being an expert system for diagnosis. “What does Watson do that does not already exist?” is the kind of question posed by Maude. Moreover: even Dr. Watson is not unbeatable. On the first day of March, the computer lost a game of Jeopardy against the nuclear physicist Rush Holt who had already won the quiz a total of five times.