Why AI medical chatbots still can't tackle coronavirus concerns
Artificial intelligence chatbots would help minimise exposure to coronavirus patients but they're a long way off from being useful
It is the technology that the coronavirus pandemic is crying out for: remote, swift, smart diagnosis that doesn’t involve human-to-human contact. But now that the crisis is here medical chatbots, the great hopes of the healthtech industry on which billions are being wagered, “aren’t ready”.
Chatbots are dialogue boxes which pop up on many websites or apps offering to help solve customer questions. Some run on simple “multiple choice” decision trees; others are more sophisticated, powered by artificial intelligence algorithms.
The NHS, for example, uses a chatbot licensed from Babylon Health, a major name in UK healthtech valued last summer at £1.65bn.
But instead of updating that chatbot to sift through patient symptoms and make a coronavirus diagnosis itself, Babylon is diverting concerned users with a cough or fever to the NHS 111 hotline, actively telling patients: “do not use the symptom checker”.
Keith Grimes, Babylon’s clinical innovation director, told The Wall Street Journal that “in a situation like this, the tools aren’t ready.”
Another British health app, Your.MD, which is based in London and last summer received significant funding from Reckitt Benckiser, the health and hygiene multinational, has updated its app. But critically, only the simple “if-this-then-that” decision tree has been tweaked, rather than the AI algorithms which may be capable of deeper, more nuanced diagnoses.
The problem is that such algorithms are themselves powered by data - vast troves and spreadsheets of information and numbers which they can crunch in order to spot patterns that might elude even experienced human clinicians.
But because the coronavirus is so new, the verified, peer-reviewed data simply doesn’t exist. What distinguishes Covid-19 from the wider family of coronaviruses that can cause anything from the common cold to the Middle East Respiratory Syndrome is anyone’s guess at this very early stage.
That means for a patient inputting symptoms such as coughs, high temperatures and shortness of breath, an artificial intelligence chatbot would find it difficult to accurately determine if the individual has been infected by Covid-19 or something more benign.
“Chatbots are perfect for triaging “ifs” and “maybes” but definitely not for diagnosing,” says Pete Trainor, chief executive of Vala Health, which specialises in video-consultations and who used to build chatbots for financial services companies.
Triage by basic chatbots, automated in decision trees known as robotic process automation, still has a purpose, he says, if it prevents potentially infected people going to GPs surgeries or schools.
But while hospitals are being asked to conduct more consultations remotely to limit the spread of the virus, those will be video-consultations - person-to-person, not person-to-bot.
Juliet Bauer, UK managing director of doctor app Livi, and former chief digital officer of NHS England, reckons video consultations will play the critical role in maintaining patient access to GP services as the number of coronavirus cases climbs to 85 in the UK.
“With six million – one in five UK employees – likely to be affected during the worst of the outbreak, video consultations give worried patients immediate access to GP advice and support, safe from spreading or catching infection,” she said.
“Video consultations enable GPs to continue treating patients even if their surgeries are forced to close.”
Livi, the UK division of Swedish start-up Kry, is currently offering its video consultations to 2.7m patients in the country who can access doctors 24/7 video consultations through its app. Babylon’s GP at Hand app offers similar uses.
But even video has its limits. Seeing a doctor in person is often the most reassuring solution to getting symptoms checked out. All the tools a doctor would typically have in their arsenal to assess a patient simply aren’t there in a video chat.
Healthtech companies, meanwhile, are potentially wary of fallout as an error in their AI or other tech solutions could have fatal consequences. “First, it might look cynical, like they’re trying to cash in,” Trainor says. “Then there’s obvious caution. If a bot diagnoses a child incorrectly and the child dies, well, there’s a lot of apprehension about that.”
That, he suggests, is a shame, because chatbots would be useful ways of tracking symptoms that may be related to the outbreak. “You can track queries by postcode, looking at the volume of queries by area.”
At his own company, he says, ad hoc video consultations are up by 50pc in the last two months, with teachers and pregnant women among those most likely to be seeking reassurance. “We remind them that this is peak cold and flu season, and try to use automation so they keep calm and carry on and don’t clog up the system.”