Can You Bond With a Robot? New Study Says Yes
AI, it seems, is everywhere, from the largest factories on the planet to the most compact smartphones in our pockets. But its potential to help humanity in new ways is only just beginning. That’s long been an exciting prospect for Woebot Health, which is focused on using technology that helps people solve human problems. Who inspires us? In part, a woman who lived way back in the 19th century. Ada Lovelace, considered the mother of AI, conceived of a generalized artificial intelligence, but she thought of it in a truly assistive way: as a technology that could ultimately be in the service of helping human beings. That core idea shaped how Woebot was designed.
But it was a famous test conducted 100 years after Lovelace’s death that showed us how not to design Woebot. Turing’s Imitation Game, which sought to test a machine’s ability to converse and behave in a manner indistinguishable to that of humans, convinced us to adopt the opposite strategy. Woebot is designed to be authentic, and to transparently present itself as an archetypal robot, complete with robotic “friends” and habits. This authenticity and transparency is key to building trust; trust, in turn, enables the formation of a bond with which the true work of therapy can begin.
That brings us to today, and an important question: Can a conversational agent for mental health actually build a bond with people? A new study just published in JMIR Formative Research sought to answer that question. Our Founder and President Alison Darcy talks about a new generation of relational agents for mental health, and what they could mean for us all.
Why study bond?
We’ve wanted to do this study since the publication of our first randomized control trial, which showed really clearly that people were relating to Woebot as if Woebot was a person. They were using interpersonal terms and were showing an affection for this entity that talked to them every day. And so we thought there’s something really unique happening here that could have particular significance in the context of mental health.
What were the core findings of this study?
Three things. First, the bond that Woebot formed with users appeared to be non-inferior to the bond created between human therapists and patients. Second, participants’ bond with Woebot was established in just 3-5 days. That’s really quick, and far faster than the bond scores in the comparison studies that were all measured between 2 and 6 weeks. And third, the bond with Woebot does not appear to diminish over time.
So, Woebot is more than a chatbot?
Definitely. I think of chatbots as bots typically used in customer service scenarios. They’re not intentionally designed to build a bond with the person that they are interfacing with. They’re really focused on performing a certain task. Our study refers to Woebot as a conversational agent, but given its results, we are now calling Woebot a relational agent. These are built to replicate some of the unique traits of human interpersonal relationships.
What will these relational agents mean for the way healthcare is delivered?
I suspect that relational agents will be used a lot in healthcare, hopefully in the near future, because they enable a longitudinal perspective of human health. Right now, we interact with our health systems in this kind of problem-focused, discreet way. But human beings are continuous. If our care systems were continuous too, I think it would drive a much better experience and afford us more opportunity to be more preventative.
What’s a real life example of that?
Think about how today, there might be a flyer in your mailbox that says it’s time for a mammogram. Maybe you throw it away in the recycle bin.
Ok…
Now imagine a trusted agent like Woebot reaches out to you on your mobile and says: “Hey, have you ever thought about getting a mammogram? Should we have a chat about what that would mean for you? Here’s why you should consider it.” That’s a really different experience. In Woebot, you have a relationship with an entity that knows who you are, what works for you, and what doesn’t work for you. And that’s a relationship that’s going to help you not just maintain good health, but avoid poor health, too.
What’s the implication for the field of mental health?
I think now that we’re able to show that we really can establish a therapeutic bond using just an automated relational agent, it means that maybe we can get better outcomes in some of these purely automated, purely self-directed approaches. And that’s really significant because it makes high quality services much more scalable. This could be really impactful for our ability to reduce the burden of illness.
Isn’t it kind of hard to architect bond in a digital solution?
It’s very complex, but we have a very deliberate approach. To start, we have a team of incredible clinicians and designers and therapeutic writers. They are thoughtful, warm and empathic individuals, and that comes through in the writing. But there are key principles that guide our design. Transparency is one of the top ones, and something we think has been absent for a long time in technology. Woebot’s core values also guide our design. And we spend a lot of time talking to people who are going through the specific experience that we’re building for. When you deeply understand the lived experience of individuals, then that’s where great empathy comes from. We’re building for a human problem in a very human way.
What’s your hope for the future?
I’m really inspired by the findings of this study. I’ve always believed technology needs to do a better job of helping humans be more human. This is a big step towards that. I also believe it’s important that we continue to be explicit about the ways in which we need to design technology to be genuinely helpful. There’s a lot of technology out there that’s vying for our attention. The more that we can talk about and understand the pieces that are helpful, the better tools we’re going to have that help us be more compassionate human beings.
