According to a publication from PR Newswire, CloudMedx, a technology company from Palo Alto, is touting impressive results following a simulated exam pitting human doctors against their “deep learning” Clinical AI assistant.
The news wire, originating from CloudMedx, made some lofty claims. Let’s break some of them down.
The Rig
CloudMedx’s Clinical AI is a program designed to take inputs in the form of a patient complaint, and output a diagnosis.
Yes, it’s a doctor-bot. Supposedly.
In their news wire, CloudMedx claims that their program uses “natural language understanding” and “deep learning” to help diagnose patients. Creating an AI capable of either is in itself a remarkable achievement.
Natural language understanding is what it sounds like – the program can supposedly understand human speech as it’s spoken conversationally. This is a problem that has been facing computer scientists for some time, and the closest anyone’s gotten to true natural language understanding is IBM’s Watson supercomputer. Achieving natural language understanding of any meaningful degree would be a truly incredible accomplishment for a company operating out of a simple three-story low-rise in Palo Alto.
Deep learning refers to a type of machine learning (another buzz-word) based on the input and analysis of data. Theoretically, it can be used to parse data into meaningful patterns that let the program draw complex conclusions.
If the AI program is exactly as CloudMedx claims (this writer is a personal skeptic) – it’s certainly a pretty sweet piece of software.
The Simulated Study
In a simulated version of a modified USMLE (United States Medical Licensing Examination – the exams you take to become a doctor), CloudMedx’s AI outscored a “small-scale” group of doctors by an average of 10% (85% versus the doctors’ 75%). The technology company also reported a 91% exam score when the doctors and AI cooperated together on a test. The findings are startling, though I naturally have a lot of questions.
The company emphasizes that it believes the true potential of AI lies not in outperforming physicians, but by helping them make decisions and assisting them in parsing relevant data.
I’ve written here before about the application of AI and big data in medicine, and to make a long story short, other tech experts believe there are serious problems that can arise when trying to apply machine learning principles to diagnostic medicine. Mainly the problem seems to lie not in the algorithms, but in the data we provide them – it’s often too scant, too random, and too stilted to draw any conclusions beyond those that would already be immediately obvious to a competent physician.
CloudMedx has partners across the country hoping that their AI programs will provide a meaningful difference in the efficiency of their organizations. Though I’m skeptical to some of the claims they’ve made in their news wire, I do share their optimistic sentiment for the limited use of AI programs as a supplementary tool for professionals.
The lines between medical technology and computer technology are blurred more every day. Do you think one day you’ll have a computer for a doctor? Share your thoughts with Patient Worthy!
