Matarić is a professor of computer science, neuroscience, and pediatrics at the University of Southern California, where she oversees the development of socially assistive robots. These intelligent machines are being designed to help patients with convalescence, rehabilitation, and training through social interactions rather than physical ones.
Socially assistive robotics—a term Matarić and a graduate researcher coined 15 years ago—is being studied at her pioneering Interaction Lab for use in stroke rehabilitation, autism behavior therapy, dementia, and traumatic brain injury, among other health applications. The robots monitor and chat with patients, coaching and motivating their human companions to do tasks they might be disinclined to do, like exercise a stroke-affected limb.
Matarić and her collaborators in the growing field are harnessing the power of social influence to change behavior. She says that interacting with socially intelligent machines can give patients the “psychological oomph” they need to do the work their physicians have prescribed.
Matarić hopes to release the first commercially available socially assistive robots through her Pasadena-based start-up Embodied in 2019. She recently spoke with JAMA about her work and the future of socially assistive robots. The following is an edited version of the interview.
JAMA:How long have you been working on socially assistive robots?
Dr Matarić:I actually founded the field 15 years ago. As the technology started to be developed, this idea of having robots interact with people through social means in order to aid in their recovery was starting to float about. Now it’s an actual field and there’s an National Institutes of Health call for proposals for socially assistive robotics. It has arrived.
JAMA:What do these robots do?
Dr Matarić:The term “socially assistive robots” or “socially assistive robotics” refers to a field of assistive robotics, which is all about assisting people in convalescence and rehabilitation. But how the robot assists the person is the interesting part. When I started looking into the field, I spoke to a lot of stroke patients, occupational therapists, and behavioral therapists. And consistently what I was told was that the patients didn’t so much need to be told what to do or how to do it, but they needed a great deal of help with motivation, and coaching, and support, and what you might call companionship during the process of recovery. One can, in theory, recover lifelong. However, how to keep doing this when it is demoralizing, depressive, isolating, lonely, and stigmatizing—that is the challenge.For socially assistive robotics, what we’re really leveraging are 2 things. One, that we know that the brain reacts to physically copresent creatures, such as robots, differently [than screen-based agents] and is influenced more strongly by robots than screens. And 2, that we are very fundamentally influenced by social factors. We are social creatures. So we’re creating robots that are physically embodied [and that] interact and influence us socially.
JAMA:Why can’t people just do these things that the robots are doing?
Dr Matarić:Of course—people have for millennia been helping people. But unfortunately now we have so many people who need help. We have 1 in 45 to 68 children being diagnosed with autism. We have 800 000 new strokes per year, and that number will double soon. We have millions of people with Alzheimer’s. So unless our demographics of care change entirely and literally, we have half of the population helping the other half of the population. We are not going to have enough resources to provide affordable and trained care.
JAMA:Isn’t there something sad about robots filling social roles in our lives in this way?
Dr Matarić:I would challenge you to ask that question of anyone who has ever interacted with a socially assistive robot. I think that we need to separate our ideas that we might have collected from literature or movies from the actual science. If we’re going to be scientific about it, then we must run experiments and we must ask the actual patient.When we do exactly that, which we and others have done for the last 15 years, we have never met a patient who felt lonely [with a robot]. In fact, we find that patients feel cared for. They report that they prefer to interact with the robot than to just do this on their own or they prefer to interact with a robot than to do it with a screen, or a phone, or an app. They smile more. They talk more. These effects have been observed repeatedly over a variety of populations across the age spectrum. But also, almost more importantly, people do more exercises and they recover more willingly than otherwise.
JAMA:Let’s talk about some examples. How can a socially assistive robot help a stroke patient?
Dr Matarić:Working with socially assistive robots for stroke patients was actually one of the very first things we did. There’s a need for doing activities of daily life using the stroke-affected limbs so the brain [can] exercise its plasticity and very gradually begin to regain function. Every time that the person who, let’s say, is right-handed and has had a stroke that affects the right arm is reaching for a coffee cup, or to open the microwave, or to get the clothes out of the drawer, they need to use that stroke-affected limb to do it. The reason that people don’t is because it is incredibly slow, inefficient, and depressing to do this. You can much more easily do this with your left arm and not take 4 hours to get dressed. On the other hand, if you use your left arm, then you’re not recovering and you will forever remain disabled.So that notion of, how can I get people to do this very emotionally and physically challenging task for themselves? That is what the robot can do. The robot can say, “Oh, come on, Mary. Let’s try and get that shirt on. You’re doing great.” [It’s] someone who’s providing coaching, and motivation, and support at just the right time. There is a combination of motivating them to do it and then encouraging them to do it properly, all day long.
JAMA:What about children with autism? How do socially assistive robots help them?
Dr Matarić:Many, many children with autism love to engage with robots. It kind of gets them over the inhibitions that they have around people. So many children with autism are more engaged and more readily interact with socially assistive robots than they do with strangers. When the child is interacting with a robot and another child is brought in as a social partner, the robot [becomes] a bridge. The robot becomes a catalyst to create more opportunities for social play, first just with the child, then with the child and, let’s say, a sibling or parent, and then with other children. [It] thereby expands the child’s ability to actually practice play and learn social skills.
JAMA:What are some of the challenges you’ve had in designing robots to interact socially with people? What have you learned that people want in their robots and don’t want in their robots?
Dr Matarić:That is really the big question and the topic of long-term research. One example is personality. When you create extroverted robots and introverted robots and you randomly assign them to study participants who are introverted or extroverted, you find that individuals who are a personality match actually do more rehabilitation exercises than those that are not personality matched with their coach robot. [Also] the one thing that’s incredibly important is to not be boring and nagging. We know that nagging doesn’t work as a behavior change strategy. But what does work? Humor is a very powerful tool for behavior change, but how much humor and when and for whom? That turns out to be the kind of question that actually is best answered with a machine learning personalization algorithm than it is with running a large number of social science studies. The exciting thing about socially assistive robotics is that it can really be a tool to understand human behavior better in a much more quantitative way in order to inform therapy.
JAMA:How intelligent are the robots you work with? Do they learn and adapt to people?
Dr Matarić:Yes, indeed. It’s incredibly important that the user feel that the robot cares. It [also] cannot do things that appear not very smart. So for example, repeating itself—that’s a deadly thing. A robot cannot use the same phrases repeatedly. And indeed, our robots do learn. If, for example, the robot makes a joke or laughs and the person does not respond positively, then the robot will decrease its probability of doing that in the near future. It will try it again, but exactly how much to try, when to try—those are all features of the machine learning.
JAMA:So what is the robot looking for in terms of the patient’s response? Is the robot looking at their face, at the words that they’re using, at their body language?
Dr Matarić:The robot is looking at everything, just the way that a human therapist would. Is the person doing the activity that they’re supposed to be doing, whether it’s using their stroke-affected limb, or making eye contact in the case of a child with autism, or moving about for the elderly person—whatever it may be. And we’re looking at all of the embodied social signals. So things like, is the person smiling? Are they leaning in when they’re interacting with the robot? Are they gesticulating? What affect are they projecting? If they’re interacting verbally, what are they saying? Fully understanding what someone is saying is difficult [for the robot], although those systems are starting to come online so that people can also feel like they’re chatting about something.
JAMA:One of the types of robots you’re working on is called the squash-and-stretch. What’s unique about it?
Dr Matarić:In order for people to feel like they’re interacting with something lifelike, the robot needs to behave in a lifelike manner. [But] almost no robots today have really flexible spines. So there is an interesting intermediary solution, which are these squash-and-stretch robots. They really are exactly what it sounds like. They squash and they stretch, so they go up and down. They have this effect of having a flexible spine. It’s a very simply way to give lifelike appearance to a robot.
JAMA:And the squash-and-stretch can be “skinned” to look different for different users?
Dr Matarić:Yes. In our labs, we have this underlying squash-and-stretch body that we developed and published. It’s open source, so anyone can use it to develop their robots to facilitate this research. We worked with a talented puppeteer in Los Angeles [to] develop [our] skins. When we worked with first graders [on nutrition], we had this dragon-like “skin” to engage them. Again, the goal here is to create a character that the patient will want to interact with for an extended time. We’re also studying habit formation in the elderly. We did focus groups to talk about what [the robot] should look like. We came up with a skin that looks a little bit like an owl. With socially assistive robots, you have to bring multiple fields together, and design is one of them.
JAMA:Robots sound expensive. Can you make them affordable for use in health care?
Dr Matarić:Yes. What’s really nice to see is now, [compared with] 15 years [ago], there are quite a few start-ups and very successful pathways to create more affordable robots for the home. I think within the next 5 or so years—certainly within the decade—we will be looking at affordable robots that are going into not just hospitals, not just managed care, but into homes. And then it really becomes a question of how much can they do, how useful they are, and how useful do they stay after 2 or 3 months?
JAMA:What do you see happening in the future, maybe more than 10 years out? Will everybody have a socially assistive robot in their home that helps them with their wellness and maybe is also their friend?
Dr Matarić:There are going to be multiple technologies coming online. Some are wearables. Some are robots that will be in people’s homes. Those things will be used as platforms for health and wellness. I’m very excited about that because I think if we look at beyond 10, 15 years, we’re really looking at interacting with machines constantly through everything around us. How will that affect behavior change? I think for that we will need companions. I want to believe that there will be a robot per person to help you with your current challenges, whether you’re procrastinating on something you need to do, or you really need to lose weight, or you have this ongoing challenge that’s lifelong, [like] autism.