This week, Arthritis Research UK released the results of its survey of over 2,000 British adults about their attitudes to taking exercise, from running and cycling to simply walking and stretching, for at least 20 minutes at a time.
Predictably enough, there was clear evidence of an ‘intention gap’ - three-quarters of respondents said that they wish they exercised more than once per week, but only around half of them (53 percent) do.
Reasons (or excuses) given included lack of time (33%), feeling too tired (32%) and cold weather (28%). But a significant number of respondents - 49% - said that they suffered from joint pain, and of these, 51% said that it had put them off taking exercise, even though research shows that regular exercise can actually reduce pain and stiffness in joints.
There can be many underlying reasons for joint pain - but arthritis is a major cause, which is why the charity has produced a guide to everyday exercise to alleviate symptoms. And for those already living with arthritis, it’s going further still, using a virtual assistant powered by artificial intelligence (AI) to offer information and support via smartphones and computers, in a form that feels like a natural conversation.
The assistant, which has been given the name of ‘Arthy’ at least for the prototype stage of its development, is powered by IBM Watson and is already being tested by 300 people with arthritis before it is launched to the wider public on the charity’s website later this year.
It’s part of a two-pronged approach by Arthritis Research UK to helping people with arthritis answer the questions they have about the impact the condition has on their day-to-day life, symptoms and treatment options, says the charity’s chief digital officer Zoe Camper.
In April, she says, the charity will also launch a telephone helpline and this will work in tandem with Arthy, so that questions that demand a more in-depth response or a more personal touch can be passed to a human, phone-based advisor. It’s all about making sure that people with arthritis have access to information and support, whenever and wherever they need it, she says.
From a financial and technical perspective, there’s a lot to consider when embarking on such a sophisticated technology deployment - especially for a charity that must ensure that the bulk of its funding is spent directly on its mission. Camper can’t go into details as to how Arthritis Research UK is paying for its work with IBM Watson and its Watson Conversation API, beyond saying that there’s a ‘partnership’ element at play here. On the technical side, it’s clear that a great deal of thought and debate was invested upfront into finding a technology to help the charity get information into the hands of people who need it. Says Camper:
It all started with our organizational strategy to support people with arthritis more directly and we spent about 12 weeks to understand what the problems were that we were trying to fix and then producing something that people with arthritis would want to use.
We maybe spoke to around 26 different individuals or organizations in the business of delivering relevant information digitally and, in the process, we came across IBM Watson. They didn’t approach us, the initial contact came from our side.
We also looked at Google and at open-source companies producing their own cognitive algorithms. I’d like to think we did a very good sweep of the market, in terms of how information could be provided in a way that puts people at the centre of everything.
Cognitive editors at work
Once the decision had been made to go with IBM Watson, a further effort was spent on preparing content so that answers could be delivered via the platform in direct response to users’ questions. Here, Arthritis Research UK has assembled a dedicated team, as Camper explains:
We now have a group of people who we refer to as ‘cognitive editors’. They’re a very new breed of content producers and I like to think we’re very much leading the way in this field. I’m not sure there are many other organizations like us that would have cognitive editors and it’s one of the things we’re most proud of developing. They have sped up the production cycle, so that all of the questions and answers go through a review cycle, are edited and then get sucked back into the platform.
The charity also spread its net widely in recruiting external help and expertise where needed. This was vital, says Camper, as she had never worked with cognitive computing before. Arthy was developed over five months using the charity’s 80 years of research-based knowledge and expertise, as well as advice from healthcare professionals, people with arthritis and IBM Watson cognitive computing experts. Academics and consultants from external organizations were also pulled into the effort:
This is sophisticated technology and that’s brought a lot of worry with it - it isn’t something we can just mess about with. But we contacted a lot of people out there at the real ‘bleeding edge’ of AI and cognitive computing, we asked them questions, we told them we needed their advice. I never went out and said I knew everything, because I absolutely don’t. The goal has always been to assemble a team of advisors with experience, fire lots of questions at them, get answers to my worries in advance and get evidence to present to our management team on the best roads to go down with this project.
In time, Arthy’s knowledge base will grow to enable it to answer more questions - including those around diet and treatment options. As IBM Watson learns from each interaction, it will automatically refine the information that is surfaced, along with developing Arthy’s conversational style. Here, however, some patience is required, says Camper:
The true learning part of all this isn’t happening yet. It can only occur when we increase the scale of the project. We’ve built the Watson conversation service and used automated testing to help the learning along, so that the cognitive element can start to get to work. Over time, we will see a move away from the very specific answers that we know we wrote ourselves, to answers that have a more nuanced feel in response to the specific question that was asked, a gradual change in the way that the natural language processing forms an answer. But I feel we’re on the cusp of seeing that, which is very encouraging.