by Matteo Berlucchi, CEO, Your.MD
@matteoberlucchi
London: 16 June 2016
My heart sank when Google’s artificial intelligence (AI) driven AlphaGo computer beat Lee Sedol, the world’s best GO player, four games to one.
A strange reaction, perhaps, for a digital entrepreneur who has set out to build the world’s first AI-driven Personal Health Assistant.
While geeks celebrated the victory I fixated on the loss. Why did AlphaGo lose the fourth of the five game series? A clean sweep would have made me happy. I could have slept soundly in the knowledge that artificial intelligence was able to deal with human unpredictability.
However, in that fourth game Lee Sedol made an unexpected move that took AlphaGo by surprise. Worse, it took the programme driven by Google’s DeepMind technology eight moves to realise the move’s implications.
The move literally wasn’t in AlphaGo’s lexicon. Indeed, it was so counterintuitive it sent the programme into a tailspin. AlphaGo literally couldn’t compute the implications. The progamme’s developer, Demis Hassabis said the defeat was a valuable learning exercise, but for the developer of an AI Health Assistant it produces sleepless nights.
Go is fiendishly more complicated than chess but it has rules and parameters — large but definable limits to the board’s permutations and strategies.
Medicine has rules and parameters as well but a lot of them are fuzzy. Medicine also has to interact with people and they are unpredictable, just like Lee Sedol. A mistaken move in Go loses a game; mistaken advice by a Personal Health Assistant can lead to over reaction or underreaction that may have consequences.
My company, Your.MD is trying to unravel a truly immense Gordian knot.
Think of it in these terms. Take 10 people, a mixture of genders and ages. The group has the same condition but each person chooses to describe it differently: emphasizing different aspects, using different words to describe symptoms, introducing diagnostic red-hearings, omitting information and sometimes only partially revealing information that might be embarrassing.
The condition may be affecting them differently because of their age and they may also be at a different stage in the conditions development. Layer on top of that the annoying tendency for many conditions to have overlapping symptoms and the potential to contract more than one condition at a time.
If you are a fan of Hugh Laurie in House you’ll have come across the concept of differential diagnosis, where his team try and separate two or more conditions that have the same, similar, crosscutting or masking symptoms. Put all that together and you have a challenge that makes a Gordian knot look like a half hitch.
Your.MD’s AI Personal Health Assistant launched last November in its current beta form. We are keen to stress that it is still in development since it needs to learn from people’s interactions to gain in intelligence, in the same way a child learns from their life experiences and observations. Your.MD uses a combination Machine Learning and crowdsourcing from NHS certified doctors to grow in medical accuracy.
Everyday my incredibly bright team pit themselves against our 800,000 users, feeding back on how right or wrong Your.MD got their assessments and advice. Everyday we learn something new about how people describe their ailments.
Doctors have distinct advantages. They are likely to have your personal history. For instance: your coughing, it’s winter but they know you’re a heavy smoker. You say you have back pain. The doctor knows it snowed heavily at the weekend, did you perhaps do some shoveling? You are 15 feeling tired, fever and swollen throat — could be tonsillitis, but what’s that hint of a rash the parent hasn’t seen indicating glandular fever.
Doctors have all four senses to draw on, environmental knowledge, historical perspective, psychological awareness and seven years of medical training. Pulling in patient records will give Your.MD some historical perspective. We can become much smarter about environmental factors and we can replicate the medical training but the rest is tricky.
However, everyday we get close to our goal and what surprises me is how little attention the medical community is paying to advances in artificial intelligence.
Artificial intelligence, combined with the ability to self-test, presents the potential for massive disruption of primary care and the democratisation of healthcare. Even in its beta form Your.MD has already been the number 1 health app in 70 countries.
Bruce Wilcox has won the annual Loebner Prize for the most human-like chatbot three times. His view on the rise of the AI doctor is simple: “This must happen. What I don’t understand is why doctors aren’t fighting it like they do every other attempt to disrupt their profession. If they don’t recognise AI is a threat then they are impervious to the future.”
“It is inevitable that these apps are going to get better. Once you start adding sensor technology and giving people the freedom to order tests themselves it is going to be difficult to see what the doctor is adding. What we are going to create is a digital Dr. House.”
The future is not here yet but it is on its way and the NHS and the medical professions should be debating the implications of a patient with a doctor in their pocket.
Your.MD is free and available on iOS and Android apps, popular messenger platforms (such as Facebook Messenger, Kik, Skype, Slack and Telegram), and via the web.
