While COVID-19 remains an occupying force in our world, and a viable vaccine is still out of reach, dealing with the pandemic comes down to mitigating health risks while coping with stress. If you’re a smoker, that’s going to be tough, on both accounts. This isn’t the Mad Men era; everyone knows the risks of tobacco use, but cessation of addictive substances isn’t easy.
Could a virtual human help?
In a May advisory on COVID-19 and tobacco use, the World Health Organization said “COVID-19 is an infectious disease that primarily attacks the lungs [and] smoking impairs lung function making it harder for the body to fight off coronaviruses and other respiratory diseases. Available research suggests that smokers are at higher risk of developing severe COVID-19 outcomes and death.“
We took Florence's artificial smarts for a spin, and then spoke with Soul Machines and the WHO for the backstory. Here’s what happened.
Your AI Will See You Now
I’m sitting in my apartment in Los Angeles looking at who-en.digitalhero.cloud on my laptop. The start screen says the AI known as Florence needs to hear my voice and see my face. I reluctantly peel off the piece of tape that usually covers my microphones and camera.
Florence appears on screen. As someone who wants to avoid bias, or cliché, with regard to emerging technology, I am going to refer to this AI in gender-neutral terms. However, as you can see from the pictures, the AI presents as female.
The WHO digital health worker isn’t in scrubs, but an orange T-shirt. I would have gone for a more medical/professional look in blue, but that’s just me. Florence is pleasantly friendly, moving around the screen in a natural(ish) manner, blinking in the right places, looking approachable and smiling from time to time.
It’s definitely not an Uncanny Valley situation. But that’s because Florence is based on a real person. Soul Machines did motion-capture sessions with Shushila Takao, a New Zealand-born actor of Maori, Indian, and French descent, known for her role on the TV series Tatau. In fact, her likeness has been used before for clients such as Autodesk. Audio matching for Florence is done in multiple languages, so we don’t hear Takao’s Kiwi accent, but I digress.
The Session Begins
Before we proceed, a confession: I stopped smoking many years ago. But I had a two-pack-a-day habit for a decade, and it was hell to quit, so I’m familiar with all the techniques—I read all the books, mainlined gum, and so on. Hypnosis is what finally did it for me, but the concept of a virtual human to engage in supportive dialogue would have been helpful.
The session kicks off with two options: we can talk about smoking cessation or dispel COVID-19 health-related myths. I want to get straight to the point, so say "I want to quit smoking." You can use the text/chat function, but I went for voice. Florence mostly understood me. I had to repeat some things twice, but that was fine.
Florence asks me how often I smoke—with a note of concern—and provides options to repeat back. The NLP (Natural Language Processing) requires trigger words as opposed to a more free-form conversation, which is disappointing, but understandable. This stuff is still pretty new.
“Regularly," I lie.
"Thank you for your honesty," says Florence. Of course I feel instantly validated by this classic reinforcement-based response, but that’s why this stuff works.
"Are you ready to quit tobacco and start living a healthier life?" says Florence. I nod and mumble yes. "Great. I can help you build your own personal quit plan."
Florence then suggested I tell someone in my life I was ready to quit. Hmm. Seriously? Friends and families of smokers IRL have heard all the excuses. And if they’re still smoking, they’re not going to be thrilled you’re leaving the pack (couldn’t resist, sorry).
In contrast—which is why this virtual human stuff should be effective—embodied AIs have many attributes that bio-based humans don’t possess, such as machine learning-based communication tools, 24/7 availability, plus zero ego or competitive zeal.
Which is why, as I found out when I interacted with WOEBOT a few years ago, virtual beings are great, because there’s no loss of face. That's vitally important with certain cultures averse to showing vulnerability and/or seeking professional help.
Anyway, here’s a rough outline of the plan Florence came up with:
Remove all tobacco products from your environment.
Tell someone you’re committed to quit.
Expect withdrawal symptoms but know that a smoke-free future increases long-term health benefits.
And know that if one uses tobacco again “I did not fail. I successfully tried. And I can successfully try again.”
That last statement was annoying—split infinitives alone. Then it got worse. “Would you like a toll-free tobacco quit line?” said Florence.
No. I would not. Because the mid-1980s just called and asked for their 1-800 telephony-based strategy back. I logged off.
Visually, aurally, and tonally, Florence is fine.
But the AI (or “digital brain,” as Soul Machines dubs it) needs to go to med school or hang out with ex-smokers to formulate better quit plans. Florence doesn’t have enough clinical cognitive-behavioral therapeutic experience right now to be truly effective.
The script is also on a closed loop (I know because I tried it several times during different sessions) and retains no personalized information.
You might want this level of anonymity but, having tried out MABU, the pint-sized healthcare robot and mobile-based AI therapist WYSA, I know the future of AI-delivered, machine learning-based health care has got to be rooted in personalization. It also requires proven research, while deploying sophisticated socio-emotive H2M (Human to Machine) behavioral science.
A truly effective smoking cessation digital health worker can’t stop at a simple scripted conversation and delivering a plan (or a toll-free helpline). Florence needs to be an AI that you can check in with when you’re struggling and want to smoke. Talk therapy works because it’s good to talk when the going gets tough. But these are early days for virtual humans. Florence is a worthwhile attempt and, I’m sure, will improve over time.
After the session, I contacted Greg Cross, Co-Founder and Chief Brand Officer of Soul Machines in New Zealand and Jaimie Guerra, Communication Officer for Non Communicable Diseases, at the World Health Organization in Geneva.
How did this partnership between Soul Machines and the WHO come about?
JAIMIE GUERRA/WHO: Early on in the pandemic WHO research showed that smokers are at higher risk of developing severe disease and death from COVID-19. As this evidence was reported, millions of more people wanted to quit. We were acutely aware that health workers were overwhelmed, and cessation services would not hold up during the pandemic. So we looked for a solution that would engage people, pique their interest, and would be available in their home 24/7.
GREG CROSS/SOUL MACHINES: When the COVID-19 pandemic hit, Soul Machines worked quickly to create COVID Sam, a first line responder for the pandemic, followed by Bella, a Digital Kiwi guide and friend, born during quarantine who can share information on a range of topics including mental health and wellbeing. WHO saw a demo that Soul Machines worked on for another public health agency.Bella (Image: Soul Machines)
Is it a managed services software model?
GC: Speed was important, so we delivered Florence fast as a managed service with a couple of modifications. Firstly, all data belongs to WHO and all public health advisories that Florence delivers comes from WHO. The conversation was developed and is an ongoing body of work between WHO health and tobacco cessation experts and our conversational engineers. To help us with this effort, Florence was designed and delivered in partnership with AWS and Google Cloud.
Why did you call this AI Florence?
JG: We have an incredible colleague named Florence at WHO and we wanted to pay her homage for all of her work. It is also the name of Florence Nightingale, the founder of modern nursing, so it works! We chose Florence to be female because the majority of healthcare workers are women, more than 70% globally. We also wanted someone who looked like she could be from many places and would work on a global scale. Florence today exemplifies diversity and empathy.
What sorts of data sets did you train Florence on?
JG: Florence is trained on decades of research into tobacco cessation. We also worked with AI conversation specialists at Soul Machines to ensure that the conversation is as fluid as possible.
How much of Florence’s platform is proprietary?
GC: Florence leverages sentient analysis from Soul Machines’ technology. Soul Machines is the only company that is creating Autonomous Animation with a patented Digital Brain. Autonomous Animation is what allows companies to create and bring to life digital people (for various use cases), with an incredibly lifelike, dynamically interactive experience. The Digital Brain technology provides its Digital People with the ability to sense, learn, adapt, make decisions, and communicate interactively and in real time with the user in a way that feels alive and engaging.
Are the results in? Has Florence stopped anyone smoking? I saw in my demo that there’s no data gathering or follow-up (unless that’s via the helpline option).
JG: It’s very early days in this initiative, but we expect she will launch people on their path to quitting. WHO measures cessation at six months of a person being tobacco-free. This is because it often takes people several quit attempts before they finally quit. What we do know is around 70 percent of tobacco users want to stop but less than 30 percent have access to the resources they need. This is where Florence comes in, to empower ‘quitters’ with the resources and information to quit and to help them build the confidence to tackle their addiction.
Finally, what's next for Florence?
JG: Right now WHO and partners are focused on scaling the experience to reach more of the 1.3 billion tobacco users by adding all six UN languages. This will mean Florence will provide WHO’s public health advice in English, French, Arabic, Spanish, Russian and Chinese.
- Amazon Pharmacy Lets You Order Prescription Meds Using an App
- Is Contact Tracing Doomed in the US?
- FDA Approves Apple Watch App Designed to Stop Nightmares
- LA's COVID-19 Response Runs on Data Science
- More in Health & Fitness