The U.S. National Institute of Health (NIH) and the National Library of Medicine have been trying to devise a system that can intelligently answer medical questions. What they wanted was a program that could be asked common questions, in plain English, and give an appropriate response. What they needed was artificial intelligence.
“What our hypothesis is and what our research in this area has been in for a couple of years now is; if we have a computer analyzing a requests and behaving like an intelligent person, can it pick out a document that better helps the consumer?” said Milton Corn, deputy director for the National Institute of Health. The tricky bit, however, isn’t just amassing the information. In a previous project headed by Corn, MedLine, “consumers” (or patients if you’re not at the NIH) query a database of medical information. But, since it’s basically a search engine with access to medical documents, consumers need to mostly know what they’re looking for in formal medical terms and then decipher the answer.
This is where Winnipeg, Man.-based AskTheDoctor comes in. “The reason we became interested in AskTheDoctor is … the computer has to be trained,” said Corn. It involves a technology called Natural Language Processing (NLP). For the program to be useful, Corn said it needs to be able to understand common language and answer in kind. “The computer has to be trained in the same way that a child has to be trained to learn something new,” he said. “To train the computer we needed actual questions, asked by patients, with actual answers that could be considered authoritative.”
AskTheDoctor.com has been taking questions from the public for two years now, since its founder, Prakash Chand, secured the dot-com rights in 2010. “There are eight physicians on the site and we create the content from people asking the questions,” he said. “Up to date, we’ve answered about 11,000 to 12,000 questions.”
As a database of medical questions and answers, he said that Chand says Ask TheDoctor may be the most diverse. “Compared to all the other universities we have a much larger sample size than other projects like this,” he said. He attributes this to the kinds of users of university programs, which is mostly students. “We are getting age range demographics and where they are in the country (as well).”
Corn recognized this, but also the frank way in which AskTheDoctor questions were answered. “I saw that the questions that they, Mr. Chand, had on his site, were actually in the same English that the patients had used,” he said. “There was no attempt to clean it up.” He said that this is particularly important because the system only becomes useful if it can understand the English that people really use, not just medical jargon.
As for the security of patient info, divulged during the exchange between doctor and patient, both Chand and Corn said that the information was stripped of all identity. “It’s totally de-identified, we have no idea who really asked these questions,” said Corn.
The addition of AskTheDoctor’s questions, among other databases, has the NIH feeling confident. Not confident enough, however, to let the public use it yet. “We’re not going to send these answers out to people yet, there’s too much risk,” he said. “But we’re going to see how the computers do with questions they’ve never seen before. That’s about where we are.”
Which brings it back to NLP. In an ideal world, with each database addition, the computers in the project become more adept at parsing questions for tone, context and meaning. Ideally, when a question is asked, only one answer is given, and it’s the right one. “NLP has been the Everest of artificial intelligence,” Corn said. “No one is claiming that the computer is going to be conscious or intelligent like a human, but it has to behave that way.”