IBM's Watson program went to study at honey. institute

Original author: Eliza Strickland
  • Transfer
This AI program has already mastered the game “Jeopardy!”. Now she will begin to study cancer.

In the final of the Jeopardy quiz, where the best players fought against the IBM Watson AI program , one of the participants, admitting defeat, along with the answer to the question attributed: “I wholeheartedly welcome our new computer rulers”

Now even doctors speak out in a similar way. “I would like to shake Watson's hand,” says Mark Chris , an oncologist at Sloan Kettering Cancer Centerin New York. He enthusiastically talks about the day at the end of 2013 when Watson, who is now his student, will graduate from the full course of training and will be ready to help doctors in the oncology center with the right diagnosis and determine the appropriate treatment courses.

For Watson, this will be a pretty good career growth, which, however, scientists from IBM foresaw from the very beginning. They hope that Watson, an AI program with exceptional natural language processing capabilities, will become absolutely essential in the healthcare industry. For the first time, Watson has demonstrated its capabilities in the game "Jeopardy"(an analogue of "My game"), in matters of which puns and puns are actively used. To solve each question, Watson needed to understand difficult English, understand complex formulations, and then search over 200 million pages of text.

After passing the analogue honey. Watson Institute will be able to understand the patient’s medical history, examine test results, search the medical literature, and provide treatment recommendations. To make the task easier to control, Watson at this stage limited the study of cancer only. Watson is currently exploring lung cancer and breast cancer, but will soon start exploring other types of cancer.

Chris, a lung cancer specialist, is collaborating with IBM on the first iteration. He admits that the project is just an experiment at the intersection of medicine and technology, but he believes that a real tool will be created from the results of this experiment. Chris notes that now, in many cases of oncology, it is not easy for doctors to determine which of the chemotherapy drugs they have available would be most effective. “Sometimes everything is quite obvious: if a certain genetic change has occurred, then a drug should be prescribed that aims to correct this change. However, for the overwhelming majority of patients, there is no such direct relationship between their physical condition and what treatment they should prescribe. Today we have many different medicines,

WATSON BRAIN: IBM Research Center servers in Yorktown Heights, NY, dedicated to the Watson project. Photo: IBM

Doctors are in difficulty, but Watson appears on the scene. He can look through thousands of similar cases of oncology, compare the results of treatment, review the latest achievements, information about which is scattered in hundreds of medical journals, and then give his recommendations for treatment. Chris explains that the goal in this case is to reproduce the decision-making process carried out by the oncologist from Sloan-Kettering. “Suppose there is an oncologist in the small town, and now he suddenly gets access to all medical journals and can take advantage of the knowledge and experience of the best Sloan-Kettering specialists,” says Chris. He emphasizes that Watson will never replace a human doctor, however she can give advice and she has her first-class advisory voice. “For doctors, this is a great tool,

In the battle with the participants of the quiz show, Watson is located between the two most successful players “Jeopardy!” for all the time the existence of the game. Her animated avatar depicted the IBM Intelligent Planet emblem, whose rays of light often shone green, indicating that Watson is on the road to victory.

The program easily managed to figure out the most difficult issues. So, for example, to a question on the topic “Search for literary characters,” which sounded like “A criminal is wanted, convicted of 12 years of eating warriors of King Khrodgar. The case was assigned to Officer Beowulf. ”Watson answered with her mechanical computer voice:“ Grandel. ” At the bottom of the screen, viewers were shown the first three results of the search for the answer to the question, along with the degree of confidence of the program in each of the answers. When Watson called the beast from the epic Anglo-Saxon poem Beowulf, which devoured the people of the king, she had 97 percent confidence in the correct answer.

IBM Research Team Watson Won't Win Jeopardy! only by a comprehensive database. The program also needed to learn how to interpret intricate clues. Like a child, the program needed to learn to understand. But IBM did not have time to explain everything to the computer program, so they had to use sophisticated machine learning methods so that Watson could quickly gain speed. Tens of thousands of question-answer pairs from past Jeopardy! Games have been added to the program so that Watson can formulate its own rules for displaying the correct answers. The program was then tested using new questions. In the case where the answer was correct, Watson noted which of the algorithms available to her,

Martin Cohn , chief medical consultant for medical care at IBM Research, said the process will be similar in Sloan Kettering. “The program will be given information about various cases of the disease and the principles of treatment, and it will have to give recommendations,” he says. As in the game “Jeopardy!” Watson will need to give out a ranked list of likely solutions and display its level of confidence in each of them. “Then one of the oncologists will say,“ Yes, Watson’s sentence sounds reasonable, ”or vice versa,“ Watson’s suggestion is completely nonsense, ”says Cohn. In this way, Watson will be trained and a degree of confidence in its answers will be established.

According to Ari Caroline, Director of Strategic Initiatives and Quantitative Analysis at Sloan-Kettering, who oversees the Watson machine learning process at the Cancer Center, the Sloan-Kettering team is currently introducing Watson case studies with all the necessary information to develop a treatment plan. At the next stage, Watson will receive examples of such cases of diseases in which there is not enough information, and Watson will need to note exactly what it lacks.

CHAMPION: Watson utterly defeated his human opponents in two games “Jeopardy!” aired in February 2011 Photo: Seth Wenig / Associated Press

“Oddly enough, Watson can request information from the user,” Caroline says. “Watson can say: 'I can give an answer right now, but I’ll be sure of it only by 30 percent, which is not very good. In order to get an answer in which I will be sure more, please give me information about molecular pathology, which is associated with the results of these specific tests. '”

During Watson's triumphant performance in the Jeopardy! David Ferucci, IBM’s lead researcher for the project, talked about the motives that led the company to invest so much in Watson development. “The desire to solve this problem is simply irresistible,” says Ferrucci, “because when we strive to solve the problem of understanding the natural language, then, in fact, we want to penetrate the very essence of what we consider to be human intelligence.”

Natural language processing can be the starting point for a wide range of different applications. IBM is already considering other areas of application for Watson, such as its use in financial analysis. However, the specialization required by the "doctor" Watson says that IBM aims to make Watson not a general practitioner, or even an omniscient oncologist, but rather an expert in certain types of cancer. It seems that every area of ​​activity that Watson deals with brings its own specialized language and raises new questions.

No one knows this better than Caroline, who does all the intricacies of teaching Watson medicine. “This is completely unlike plug and play,” he says. “There is no such off-the-shelf natural language processing tool that could just be plugged in, and it would immediately interpret anything.”

But, despite its specialization, Watson is still a big step towards the creation of universal artificial intelligence, comparable to the invasion of IBM earlier in the development of unsurpassed artificial intelligence gaming systems. Deep blue systemdeveloped by IBM, which defeated the then world chess champion Garry Kasparov in 1997, except for playing chess, could not do anything, even could not play checkers.

We have yet to find out if Watson can count on repeating its success, adding to its knowledge in new practical areas. The project demonstrated the attractive potential of machines that can communicate with us in the same language, but at the same time it serves as a reminder that it’s too early for us to admit defeat from our computers.

This article was originally published in the print version of the IEEE Spectrum magazine, entitled Watson Goes to Med School.

PS In this translation, Watson is feminine, firstly because it is referred to as a program (even though it is a supercomputer too), and secondly because, as you know, Watson was a woman ;-)

Also popular now: