Scientists have labelled the 1970s and 1990s as two distinct “AI winters,” when optimistic predictions about AI turned into gloomy pessimism as projects failed to live up to the hype. IBM sold its AI-powered Watson Health It sold what analysts called residual value to a private equity firm earlier this year. Does this deal herald a third AI winter?
Artificial intelligence has been with us longer than most people realize, and the 1960s TV show “The Jetsons” saw the robot Rosey attract a huge audience. This application of artificial intelligence—the omniscient maid who keeps the family running—is a sci-fi version. In a healthcare setting, AI is limited.
The concept is designed to operate in a task-specific manner, similar to real-world scenarios, such as a computerized machine defeating a human chess champion. Chess is structured data with predefined rules for where to move, how to move, and when to win. The electronic medical records on which AI is based do not fit into the neat confines of a chessboard.
Collecting and reporting accurate patient data is the problem. MedStar Health Seeing sloppy electronic health record practices hurt doctors, nurses and patients. The hospital system took initial steps in 2010 to focus public attention on the issue, an effort that continues to this day. MedStar’s campaign usurped the acronym “EHR,” turning it into “mistakes often happen” to clarify the mission.
MedStar analyzed software from leading EHR vendors and found that input data was often unintuitive, and displays made it difficult for clinicians to interpret the information. Patient records software is often irrelevant to how doctors and nurses actually work, leading to more errors.
Examples of medical data errors appear in medical journals, the media and court cases, and they range from erroneous codes that delete key information to mysteriously changing the gender of a patient. Since there is no formal reporting system, there is no clear data-driven number of medical errors. There is a high chance of bad data being dumped into an AI application, undermining its potential.
Developing artificial intelligence starts with training an algorithm to detect patterns. Input data, and when a sufficiently large sample is achieved, test the algorithm to see if it correctly identifies certain patient attributes. Although the term “machine learning” implies an evolving process, the technology is tested and deployed just like traditional software development. If the underlying data is correct, properly trained algorithms will automate functions, increasing physician efficiency.
For example, diagnosing medical conditions based on eye images. One patient’s eye was healthy; the other eye showed signs of diabetic retinopathy. Images of healthy and “sick” eyes were captured. When enough patient data is fed into the AI system, the algorithm will learn to identify patients with the disease.
Andrew LiangA Harvard professor with private-sector experience in machine learning presents a troubling scenario of what could go wrong without anyone knowing. Using the eye example above, suppose more patients are seen, more eye images are fed into the system, which is now integrated into the clinical workflow as an automated process. So far, so good. However, suppose the images included treated diabetic retinopathy patients. Those who were treated had a small scar from the laser incision. Now the algorithm is tricked into finding small scars.
Adding to the confusion, there is disagreement among physicians about what the thousands of patient data points actually mean. Human intervention is required to tell the algorithm what data to look for, and it’s hardcoded into machine-readable labels. Other issues include potentially erroneous EHR software updates. Hospitals may switch software vendors when information is moved elsewhere, leading to so-called data transfers.
That’s what happened at MD Anderson Cancer Center, yes Technical reasons Why IBM’s first partnership ended. IBM’s then-CEO Ginni Rometty described the arrangement, announced in 2013, as “the company’s healthcare”moon landing. MD Anderson said in a statement Press release, which will use Watson Health in its mission to eradicate cancer. Two years later, the collaboration failed. To move forward, both parties must retrain the system to understand the data from the new software. This is the beginning of the end for IBM Watson Health.
AI in healthcare is as good as data. Precise management of patient data is not science fiction or a “moon shot,” but it is critical to the success of AI. Another option is for a promising healthcare technology to be frozen in time.
Photo: MF3d, Getty Images
