This summer, Riva-Melissa Tez was searching online for research that might help her father. He’d gone into a coma after suffering a stroke, and she wondered what the latest recommendations said—whether playing music to him in his native language could keep him connected to this world, or if giving him Prozac could boost his chances of recovery as it had done for mice in a study last year. Doctors are so busy saving lives, she thought, that they couldn’t possibly keep up with all the papers published every day.
Her concern is shared by doctors, who wonder what they could be missing in the 2.5 million scientific papers published every year. Popular sites like MedCalc and UptoDate are useful tools for doctors to consult diagnostic criteria and double check on treatment guidelines. But there’s plenty of room for improvement, and some believe artificial intelligence could be a solution to science overload: machine learning assistants to read incoming papers, distill their information, and highlight relevant findings.
Last month, a company named Iris launched a first version of that type of assistant. The machine can currently read the abstract of a paper, map out its key concepts, and find papers relevant to those concepts. It provides a quick way to get a sense of the scientific landscape for a given topic, something especially useful when you don’t know the exact keywords for the type of research you are looking for. The Allen Institute for Artificial Intelligence also recently launched a search engine, Semantic Scholar, that takes search beyond keywords.
“One of the problems is getting research out of the dusty digital drawers and into the hands of people who can implement it,” says Anita Schjøll Brede, the CEO of Iris. Her tool should make it easier to navigate the literature, especially for people doing interdisciplinary research, she says. In three years, the company plans to make a proactive version that remembers which papers you read last week and gives you new ones based on your project description. And in 10 years, she hopes the AI will be powerful enough to discover new concepts—based on its reading and understanding of the literature—all on its own.
Iris’s machine is discipline-agnostic. It doesn’t care if you ask to find research about cancer or composite material. But other groups are homing in on the problem in medicine. IBM is using its AI technology to take on the high-stakes field of cancer treatment with Watson for Oncology, an application trained by expert oncologists at Memorial Sloan Kettering Cancer Center. It draws from papers, patients data, and clinical trials to help generalized cancer doctors keep up with developments in the field.
IBM’s application doesn’t expand to other medical fields, and Iris’ machine currently just improves how literature is organized and accessed. For a typical doctor with a typical schedule, just finding the proper research is not enough: Someone has to read and understand that research. “This is a huge problem,” says Setareh Alipour, a medical resident in New York. “Scientific data is becoming so vast that even specialized doctors can’t know everything that is being discovered about their field. And I’m talking about larger studies, not small and unreliable data.”
The idea that scientific literature should have a place in clinical practice—so-called evidence-based medicine—is a fairly recent departure from medicine’s tradition of practicing what you learn in medical school. Physicians’ knowledge doesn’t always age well: Only about half of patients in the US receive the recommended course of treatment. Any efforts to bridge the knowledge gap or just make it a bit easier to track new science would be welcomed by doctors. “I would love it if a machine could act as my reliable and smart memory,” says Alipour.
Tez’s father came out of the coma and is now recovering. But some of the papers she found online actually piqued the interest of on-call physicians, who printed them and put them on the notice board in the neurology ward. Following this experience, Tez, who is the co-founder of an AI-focused venture capital fund Permutation, has thought a lot about the future of medically-minded machines.
An AI physician’s assistant could, she imagines, plug in to a universally accessible electronic health record that keeps all your information—cross referencing your symptoms and medical history with the most up-to-date recommendations to guide treatment choice. It could also alert your doctor about new research that could be of interest “The problem with hospital research and applying AI is that people who work in AI don’t understand hospitals,” says Tez. Just as shown by IBM’s partnering with those in the medical field, if there’s an AI solution to improve healthcare, it will likely come out of a collaboration between hospitals and technologists.