Humans have had a love-hate relationship with AI. We love our car autopilot. But then we get scared when the media focuses on the one car accident involving an autopilot, even though there are 1,500 car accidents per day (according to the National Highway Traffic Administration) that have nothing to do with autonomous driving. We love chatting to Amazon’s Alexa, but we get upset when it only knows the answer to 1 out of every 10 questions about your health.
We would take Alexa’s honesty and lack of medical knowledge all day long versus Siri’s “know it all” attitude and usage of an Internet search engine for medical answers. Ironically, the reason users with questions about their health are more satisfied with Siri than with Alexa is because Siri – you guessed it – knows it all. Unfortunately, this erroneous claim results in 14 (!) times more potential deaths for Siri vs Alexa. 1 in every 5 (or 20.9%) of medical inquiries in Siri and its consequent diagnostics and treatment advice could cause potential death, according to a 2018 study in the Journal of Medical Internet Research. In other words, Alexa is not designed for medical knowledge, and because it didn’t know the answers, it could not cause much harm. Siri, on the other hand, was designed to answer every question. However, as the study points out, for most of the medical questions Siri just sends the user to a web page. (Full disclosure: this study only analyzed 394 medical inquiries from 54 subjects, which is a small sample by most data science standards, but it is still a decent sample to infer some interesting conclusions.)
Many thought that the COVID-19 pandemic would accelerate the acceptance of AI by medical professionals. In fact, the opposite happened. According to a recent industry study by Eagle Hill Consulting, 63% of healthcare workers said that digital health solutions added more stress, while 83% said their jobs were harder to do when new digital tools were introduced.
It’s the same “Siri effect” mentioned earlier. When the front-line health workers are burning out and getting desperate, giving them the most popular AI tool is not helping, and in fact creating an aversion to AI. Healthcare needs AI tools that would make the healthcare workers and their patients’ lives easier, not harder. Healthcare is in great need of the right digital tools and innovative wellness solutions.
The reason 1 in 5 health diagnostic and treatment answers by Siri could cause potential death is simple. Using the Internet or a search engine to offer medical advice, based on one question, or one or two symptoms, is outright dangerous.
Think about your experience with a human doctor. It’s interactive. The doctor would ask you a question. Based on your answer, the doctor would ask a follow-up question. And so on. The doctor would consult your chart. A good doctor would also recommend a second opinion.
Good healthcare is interactive healthcare. Good healthcare is based on science and depends on a variety of sources of medical knowledge, including digital knowledge.
Interestingly, a simple disclaimer that a digital assistant should only be used exactly for what it is –assistance – and a doctor’s advice is still highly recommended makes a huge difference. A brand new study in the Journal of Medical Internet Research concluded that patients do respond to such disclaimers, and that statically significantly improves their health outcomes.
What about AI and doctors working together? Critics would suggest that’s not a good idea. And they may have a point. According to a brand new study by data scientists at Google’s Deepmind, human medical experts are still much more accurate than the machine learning algorithms, including Gopher, a 280 billion parameter transformer language model.
So human doctors are better than machines. Case closed, right?
No one is suggesting that human doctors are not good at what they do. Their experience is irreplaceable, and no machine would be to replicate their experience, their empathy and their human touch for many decades and centuries to come.
However, machines are, by definition, better in speed, calculation and aggregation. 200 to 1,000 peer-reviewed medical studies are being added to WellAI’s medical dataset daily. There is no way any human can read and learn this much information. Do you know who can read these many articles? Machines can.
According to Bayes’ Theorem for Digital Health, combining data from an interactive digital health assistant (e.g. WellAI) and human doctor’s expertise creates the best outcome for the patient, versus relying on just doctor’s expertise or just digital health assistant’s advice. AI and humans work together to create better health outcomes. Wow! What a concept.
So please don’t dismiss AI in helping you feel better. Please give digital health a chance.
A recent report from the World Economic Forum found that around 66% of healthcare providers spread across 14 global markets are investing significant amounts of resources and time into digital health solutions.
The future of healthcare is doctors using AI as a tool to create better outcomes for their patients, the same way they are using a stethoscope, MRI or CT scan.
Stay healthy! Stay knowledgeable about your health.