The health AI revolution
TV dramas are full of genius doctors and detectives like Dr Gregory House, Sherlock Holmes and Columbo who effortlessly spot the links and hidden meanings between seemingly inscrutable clues while their diligent colleagues stumble along in their wake. Barring some inspired guesswork, we too have to wait until the grand finale to make sense of what we’ve seen and heard.
Sherlock Holmes is a fictional intelligence of course, but artificial intelligence (AI) is real and becoming ever more mainstream. One important area where AI is developing rapidly is healthcare – where spotting a causal link and reading between the lines is quite literally a matter of life and death.
AI is a very short acronym for a very broad and complex subject. At its most basic level, AI is simply a software application that is complex enough to simulate human intelligence. Many developers differentiate between Weak AI and Strong AI.
Weak AI is software that has been designed to perform a particular task perfectly. A game that can beat you at chess or robotic machines used in manufacturing are examples of Weak AI – the software processes the input that it’s given and proceeds flawlessly, forever to perform its task according to the designer’s intentions. Apple’s Siri is another example which uses the whole internet as its database but nevertheless can only perform a limited set of tasks. Most would agree that this isn’t artificial intelligence at all, just incredibly clever software that processes responses given to it by human beings or other software programmes. The ‘intelligence’ is simply a hard-wired response to a given input.
Strong AI is the next step whereby a software application can intuitively ‘learn’ from its experience and reshape its answers to suit an unrecognised parameter, re-writing and improving its own code as it learns. This, of course, is the stuff of science fiction, of computers running out of control and becoming more powerful than their designers. There are no examples of Strong AI currently in existence. It exists only in theory and the Turing test remains impassable.*
While we wait for true AI to lead us into a brave new world, a process called deep learning is taking AI to astonishing levels using pattern recognition. Software that can recognise, and act upon, intricate levels of detail and can compare and contrast it with a database as broad as the surface of the earth.
That’s a key skill – that’s what doctors do.
Every major pharmaceutical company in the world is now heavily invested in using AI to develop devices and applications that can improve critical decision-making in healthcare environments. Skilfully implemented AI can perform the most delicate surgery, diagnose ailments before complications set in and suggest treatment alternatives that are not immediately obvious to the physician. By recording 750 coughs from existing patients and 1,500 coughs from healthy volunteers, a start-up company in Toronto has developed software that can tell the difference between tuberculosis, pneumonia, upper respiratory infection or bronchitis by listening to the patient cough.
At home, people with minor ailments are likely to self-diagnose by checking symptoms on the internet and then ‘chat’ with a virtual doctor or nurse. Home health monitoring devices are a huge growth area enabling users to get reassuring feedback without leaving the house. The applications of this technology are seemingly endless – from low tech apps that measure blood pressure or remind patients to take their medication, right the way through to highly sophisticated mixes of hardware and AI that can scan a wound or rash and offer immediate advice.
In the last 20 years, the relationship between the public and the medical profession has changed profoundly, and in coming years AI is set to be at the forefront of that change. An AI system that can entirely replicate the countless and subtle human skills of an expensively trained physician is still a work in progress, and a long way off. Human traits such as empathy, compassion and sensitivity cannot be coded. Although AI in healthcare is undoubtedly of enormous benefit to us all, humans must remain in the driver’s seat.
*The Turing test, developed by Alan Turing in 1950, is a test of a machine’s ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human.