The latest NAB Health Insights Special Report takes an in-depth look at health consumer attitudes and satisfaction levels. It’s a revealing investigation available for download now.
Chatbots are ready to answer questions 24 hours a day – a big plus when you’re talking health or finances. Dana Bradford, senior research scientist at CSIRO, and NAB’s head of innovation Jonathan Davey discuss how conversations with machines could make life easier.
We’re getting used to the idea of chatting with a machine. Virtual assistants such as Apple’s Siri, Amazon’s Alexa and Google Home are paving the way for chatbots to play a more significant role across all business sectors. And they’re already having an impact in healthcare and banking – two areas where this application of artificial intelligence could have wide-ranging benefits.
Unplanned hospital readmissions place a serious burden on healthcare systems around the world – in the US it’s been estimated they cost more than $US17 billion a year. There’s evidence, however, that follow-up care can help. For example, a recent study published in the Australian Health Review suggests that telephone support could reduce readmissions by almost a third.
“The problem is that we currently don’t have the workforce to provide that level of service,” says Dana Bradford, a Senior Research Scientist in the CSIRO’s Australian eHealth Research Centre.
“This is one of the areas where chatbots could help fill a gap. In the future, chatbots could prompt patients to take their medication, check how they’re feeling and ask questions [to establish] whether their pain is under control. If something isn’t quite right, they could recommend anything from seeing a GP to calling an ambulance.”
Similar chatbots could help people manage a chronic illness more effectively.
“One has been developed in the United States to provide support for children with diabetes,” Bradford reveals. “This is a great example of a tool that works in concert with a GP to improve patient health, provide more independence and reduce the number of visits to the doctor’s surgery.”
Bradford also predicts chatbots will play a crucial role in meeting the needs of an ageing population.
“We know that healthcare requirements increase disproportionately as you get older, but we have fewer and fewer people to provide those services,” she says. “There are currently about five people of working age for every person aged over 65 and, by 2050, that number will have halved. I believe this is where chatbots will come to the fore.”
So, what exactly is a chatbot? Bradford defines it as “a natural language processing agent that receives input in the form of text or speech and provides a response that leads to the next input”. In other words, it’s a machine that can hold a conversation.
They currently work in one of two ways.
“Some go through a scripted decision tree and formulate the best response from those they have in their database,” Bradford explains. “Others are programmed with machine learning so they can adapt to the conversation and learn as it progresses.”
A decision tree is less flexible as the database may not hold an appropriate response.
“In this case, the chatbot will simply say it doesn’t know and pass the enquiry on to a human,” Bradford says.
Creating the scripts for decision trees can also be challenging.
“When my colleague David Ireland and the team were programming their chatbot HARLIE (Human and Robot Language Interaction Experiment), they wanted to be able to include family members in the conversations,” she says. “But they had to be very careful about the context. You don’t want the chatbot to ask: ‘How’s your brother Richard today?’ if his name has been mentioned because he died.”
One advantage of decision trees is they provide more control over the responses.
“Machine learning adapts as it goes so it can pick up inappropriate language,” Bradford says. “Microsoft had to take down an entertainment chatbot called Tay very quickly when it started speaking in a racist and offensive way.”
Not everyone is comfortable with the idea of conversing with a chatbot.
“When we rolled HARLIE out to community groups, the conversation stopped as soon as it showed its fallibility and they realised they weren’t talking to a person,” Bradford says. “I think it’s inevitable that the more human-like chatbots become, the more users will engage. Already, the inventor of the Mitsuku ‘virtual friend’ has found that people often can’t tell whether they’re talking to him or his chatbot.”
Bradford also points to research at the University of Southern California, which found people spoke more freely about their mental health issues when they knew they weren’t talking to a human.
“Chatbots may prove valuable in this area because they remove some of the barriers to mental health outreach such as judgement and stigma.
“It’s vital that healthcare providers start thinking about how chatbots could be incorporated into their organisation,” she adds. “Preparing for these developments now will bring benefits in the future.”
NAB is currently developing chatbots to communicate in two ways.
“The first helps customers find out more about a product on our website,” says Jonathan Davey, Executive General Manager of Digital & Innovation at NAB. “You can simply type in a question and a chatbot will supply the answer. At the moment, our chatbots can respond to about 200 different kinds of questions and they’re learning all the time. If a question is too complex, it’s directed to a human.”
NAB is also building banking services into the Google Assistant platform and recently announced it would integrate services with Amazon Alexa, arriving in Australia in February.
“This is a voice-activated service which allows you to talk to Google rather than keying in questions on a keypad,” Davey says. “NAB customers can ask questions such as ‘Where’s my nearest ATM?’, ‘What’s the interest rate on this product?’ and ‘What are my credit card options?’. Shortly, we’ll be launching the service into our authenticated environment via Amazon Alexa devices, which will allow our customers to ask questions relating to their accounts, such as ‘Have I been paid yet?’ or ‘What were the last three transactions on my credit card?’ or ‘How much do I owe on my home loan?’.”
The aim is to make accessing information as fast and simple as possible, giving customers more flexibility and ease.
“We’re using technology to provide a greater level of convenience to our customers,” Davey says. “The less time they spend looking for banking information the more time they have to spend on their business. Judging from the interest we’ve seen so far, I believe the use of chatbots will become more and more prevalent in the future.”
© National Australia Bank Limited. ABN 12 004 044 937 AFSL and Australian Credit Licence 230686.