Whatever Happened to All Those Care Robots?

The first thing Pepper told me was that he was running out of battery. “He’s got about 15 minutes before he dies,” Emanuel Nunez Sardinha, a Ph.D. candidate in robotics at Bristol Robotics Laboratory, told me. That turned out to be plenty. Sardinha greeted Pepper; then I did. I asked Pepper how he was doing, to which he replied, “How are you doing?” Then Sardinha resumed telling me about the sorts of things Pepper, a friendly, wide-eyed robot designed to assist humans through social interaction, can do, such as talking through an exercise routine while demonstrating upper-body movements (he doesn’t have legs). But Pepper can get “nervous” in crowds—that is, his voice recognition short-circuits in an environment with multiple people talking—which is what seemed to happen at the lab that day. He kept piping up unprompted as we chatted, flustering Sardinha, who, with a gentle apology to Pepper, put him to sleep.

For such an underwhelming little robot, Pepper has managed to inspire remarkable faith in his potential over the years. He wasn’t designed for any particular purpose; he was introduced by SoftBank Mobile and Aldebaran Robotics in 2014 as “the world’s first personal robot that can read emotions.” But roboticists in private companies and academic institutions quickly set about retooling his software for elder care. Ads showed Pepper monitoring the corridors of a care home for wandering residents, and guiding elderly visitors to the appropriate room of a hospital. In the media, researchers voiced lofty aims for him: He might function as a helpful companion for elderly folks living on their own, reminding them to take their medication while engaging them in sorely needed conversation. In a care home, Pepper might help keep an eye on residents, entertain them with games and jokes, or simply offer some of the friendly interaction that overstretched staff cannot. In 2018, Pepper himself appeared before the U.K. Parliament, citing his potential to “reduce pressure on health-care services” and “boost independence, reduce loneliness, and improve the quality of life among elderly people.” The following year, the U.K. government cited Pepper when it announced that it would invest 34 million pounds in developing care robots that “could revolutionise [the] UK care system.”

Read: The robot revolution in caregiving

But Pepper has yet to make it very far out of the lab. He and other social robots have been tested out in care settings in multiple countries over the past decade, but very few nursing homes actually own one. Hard data are hard to come by, but Chris Papadopoulos, an expert in health technology at the University of Bedfordshire, guesses that fewer than one in 1,000 U.K. care homes uses a humanoid robot on an ongoing basis. A city council in England made headlines in 2017 when it recruited Pepper to work in its adult-social-care team, but when I called their office to ask how he was getting along, the woman who answered the phone had no idea what I was talking about. Likewise, media portrayals of seemingly widespread use of robotics in Japanese care homes have little basis in reality, James Wright, a visiting lecturer at Queen Mary University of London and the author of Robots Won’t Save Japan, told me. About 10 percent of care homes in Japan use any sort of robot—including monitoring systems or mobility aids—let alone a humanoid. Production of Pepper was paused in 2020 due to lack of demand. (Aldebaran was eventually acquired by United Robotics Group, which still advertises Pepper as “an ally in Healthcare” that can “interact, entertain and provide companionship,” “enhance the efficiency of the administrative process, improve quality and consistency of patient experience” and “support caregivers.”)

There are likely many reasons that the long-predicted robot takeover of elder care has yet to take off. Robots are expensive, and cash-strapped care homes don’t have money lying around to purchase a robot, let alone to pay for the training needed to actually use one effectively. And at least so far, social robots just aren’t worth the investment, Wright told me. Pepper can’t do a lot of the things people claimed he could—and he relies heavily on humans to help him do what he can. Despite some research suggesting they can boost well-being among the elderly, robots have shown little evidence that they make life easier for human caregivers. In fact, they require quite a bit of care themselves. Perhaps robots of the future will revolutionize caregiving as hoped. But the care robots we have now don’t even come close, and might even exacerbate the problems they’re meant to solve.

Some researchers have not given up on Pepper. “There are so many benefits to continual contact and interaction that we are unable to provide to our elderly because of health-care-worker shortages,” Arshia Khan, a roboticist at the University of Minnesota at Duluth, told me. Her lab deployed a fleet of Peppers into eight nursing homes in Minnesota in 2022. She admits that the robots have limitations—they can’t perform physical care yet—but Khan believes that lives would have been saved during the coronavirus pandemic if more elderly people had had robots to interact with when they couldn’t be with others. “Loneliness doesn’t just make a person feel depressed. It actually kills,” Khan said.

There is, to be clear, no evidence that care robots can save lives. And although some research suggests that social robots reduce loneliness or otherwise improve well-being, the conclusion comes with a few asterisks. Many studies involve robotic pets—usually Paro, a soft robotic seal designed to soothe and stimulate people with dementia—not humanoids. Many of the studies are bad: Multiple meta-analyses have lamented that studies on social robots have methodological issues that make it difficult to know what to make of the results.

Even higher-quality studies on humanoids such as Pepper have some limitations to consider. As part of a large collaboration between the European Union and Japan, Papadopoulos conducted a study that tested a “culturally competent” version of Pepper. In practice, that meant loading Pepper with knowledge about the local culture—at an English nursing home, Pepper might talk about rugby, for example—and then installing him in residents’ rooms for up to 18 hours over the course of two weeks. Compared with those who weren’t around a robot, residents who got to hang out with Pepper—particularly the “culturally competent” version—reported a boost in emotional well-being. Of course, that doesn’t mean Pepper was actually satisfying residents’ need for human connection: Residents’ self-reported loneliness didn’t significantly improve. In fact, Papadopoulos told me that many residents were initially wary that Pepper might replace human caregivers but came around to him as it became obvious that was “absolutely impossible.” Instead, they saw him “as more of a fun, assistive therapeutic bit of kit … like a television or an iPad or something,” he said.

Whatever care robots’ impact on well-being, multiple studies have found that, far from easing the demands on human caregivers, they can create additional work for them. The most obvious reason is that introducing a social robot into a care home means bringing a fragile machine into a setting full of fragile people. Leaving Pepper or other such devices lying around is simply not an option, because they and residents risk harming each other. As part of the research for his book, Wright spent six weeks in a Japanese care home that was testing out Pepper; the robot was stored away when not in use and closely monitored when he was.

Potential safety issues aside, Pepper didn’t seem to work terribly well without help, Wright told me. Initially, the plan was for Pepper to run exercise classes with residents. “The staff members found out very quickly that if they just let Pepper stand at the front of the room and do its thing, basically, the residents would kind of ignore it,” Wright said. A caregiver had to stand next to Pepper, repeating its words and mimicking its movements to get the residents involved.

Naonori Kodate, an associate professor in social policy and social robotics at University College Dublin, observed something similar while producing a documentary about care robots in Japan; the social robots did seem to get the residents talking and boost morale in the home—but only with some elbow grease from staff. “It’s not like you can just leave the robots and then all the people speak to them and have fun together,” Kodate told me. In fact, to be of much use at all, the robots often needed the help of a human who really knew the residents well.

Read: The friends who are caring for each other in old age

The nursing home where Wright conducted his field work also tested Paro, the cuddly seal robot, which was designed for regular handling. The hope was that such a hands-off robot might help soothe the home’s more agitated residents and thus cut down on some of the attention they required from staff members. (Paro’s manufacturer does not advertise it as a labor-saving device; it emphasizes Paro’s capacity to reduce patient and caregiver stress, and to enhance socialization of patients with one another and with caregivers.) Paro didn’t seem to interest the home’s neediest residents much, but others became so enamored with it that staff became concerned. One woman in particular seemed to develop a fixation with the robot, taking any opportunity to wheel Paro back to her room, where she’d put it to bed like a baby and often cry while talking to it. She refused to take meals or go to bed without Paro. So the staff started keeping tabs on who was using Paro when and for how long. “In the end, it just got put on a shelf, because it was easier to do that than to constantly monitor everybody,” Wright told me.

These types of challenges are likely underreported. Care workers are largely overlooked in research on care technology, Cian O’Donovan, a researcher at University College London who is leading a project aimed at developing robotics that empower care workers, told me. One review of research studies on robots in assisted-living facilities noted that the majority of studies do not collect data on the experiences of caregiving staff with the robots, instead focusing on residents’ experiences.

Papadopoulos and his team did consult staff as part of their project, and the concern that Pepper might create additional work for caregivers didn’t come up. But that might be because, as a result of various safety and ethical concerns, the researchers were doing the monitoring themselves, and staff were instructed to carry on as though the robots weren’t there. Such oversight is a common feature of this sort of research. One widely cited study reportedly found that Paro reduced loneliness in the elderly even more effectively than their usual activities did, such as going on a bus trip or playing bingo—but the team tested him in one-hour group sessions guided by a researcher or member of staff. Multiple studies investigating robots’ effect on well-being and loneliness employed a “Wizard of Oz” approach in which all of the robot’s questions and answers were keyed in by a human at a laptop out of sight. The tightly regulated nature of these studies adds an important caveat to their findings: Social robots seem to improve well-being under the careful watch of humans.

Some of the researchers I spoke with are certain that whatever shortcomings Pepper has will be overcome with better technology. Both Khan and Papadopoulos see a future in which robots can do anything a human caregiver can. Recent developments in AI are already allowing social robots to engage in more sophisticated conversation. Even the physical limitations of modern robots are on the precipice of being solved. Papadopoulos pointed me to Google’s newly released Mobile Aloha, a comparatively low-cost robot that researchers have trained to cook shrimp and wash laundry.

Read: The new casualties of automation

Other researchers are far more skeptical. Caring for someone isn’t as simple as jumping to do their bidding. Even a robot that can have a satisfying interaction with an elderly person may nevertheless fail to care for them. Paro successfully captivated the woman in the home where Wright did his field work, but only a human caregiver recognized that her reliance on it had curdled into something self-destructive. It’s not just a human touch that Pepper lacks, but a human perspective and the capacity to act on it.

Caregiving is not the fulfillment of a set of discrete tasks; it’s the management of someone’s quality of life. The sort of knowledge required to do it well is person- and community-specific. Kodate told me that he was fascinated by the subtlety of information caregivers relied on to ascertain desires, frustrations, and needs that individuals themselves might not know or cannot express. That’s why care is done best in the context of strong relationships.

The robots we have now may offer a glimpse of both the promise and peril of what care robots could come to be. Most researchers I spoke with saw potential for robotic technology to assist and even bolster a strong caring relationship, but they were doubtful it could ever supplant one. If they are correct, then even future, more capable robots could lead us down a very strange path. Pepper and Paro did not alleviate the demands of caregiving, but they did change them. Carers spent less time interacting with residents and more time monitoring resident interactions with robots. Instead of coming up with their own exercise routines, they mimicked Pepper’s. In other words, care itself became more “robotic,” Wright noted in his book. That’s an attractive prospect from a business standpoint; minimizing the intimacy of care could make the humans who do it more interchangeable. But such a robotic revolution in caregiving would succeed only by further imperiling the relationships that overstrapped and underpaid carers already struggle so much to build.