Expert Perspectives—Technology for Empathy and Plurality
Compassionate Computer Agents

Fig. 1: Timothy W. Bickmore, Northeastern University
Figure 10.1
Timothy W. Bickmore, Northeastern University

Health represents one of the most primal human needs, and technologies designed to positively impact health thus compose an important aspect of positive computing. Over the past decade, many researchers have investigated the development of automated caregivers that promote physical and mental health in people who interact with these systems. Many of these researchers believe that the most promising design methodology is to emulate the best role models we have—expert, compassionate, human caregivers—with as much fidelity as possible. The resulting systems are conversational, communicating with people in some form of natural-language dialogue, often accompanied by a virtual or physical representation of a human caregiver’s persona.

Media used in these systems span voice only (interactive voice response), animated agents (also known as “embodied conversational agents”), and social robots. The anthropomorphic qualities of these interfaces not only serve to make them approachable by a wide range of users but are essential for conveying many of the subtle cues used by human caregivers, such as empathy and compassion. These systems are necessarily limited in scope, but they promote some narrow dimension of health and wellbeing among their users, typically over the course of many conversations that can last weeks or months.

Most of the automated caregivers developed to date address physiological and mental health needs, including interventions for medication adherence, exercise and diet promotion, cancer prevention, prenatal care, and depression counseling. Most of these systems have demonstrated significant improvements in health relative to nonintervention or standard-of-care control groups, and many have been shown to be equal to interventions performed by human caregivers.

An increasing number of automated caregivers are being developed whose primary goal is to provide companionship, typically for older adult users who live alone. Although these agents may provide health or wellness coaching or cognitive stimulation activities, they also engage in a wide variety of social behaviors whose goal is to provide a feeling of social support among their users, which has been shown to be an important factor in decreasing mortality among isolated elderly. One study demonstrated that a companion agent provided to isolated older adults significantly decreased loneliness after a week and that the more users talked to the agent, the less lonely they were.

In addition to general anthropomorphic qualities, these agents use specific human behaviors to improve their effectiveness and their users’ sense of wellbeing. Social bonding between a caregiver and client (referred to as “therapeutic alliance”) is crucial in maximizing outcomes for all kinds of health interventions. Agents that use behavior designed to improve social bonding and therapeutic alliance have been referred to as “relational agents,” and several studies have shown relational behaviors can lead to significant increases in engagement with these agents. One of the most important agent behaviors for social bonding is the display of empathy, comforting, and compassion for the user when they are in a distressed state. Even though these behaviors are simulated and thus do not exhibit “true empathy” (by many definitions), a wide range of agent behaviors—including facial displays of concern, empathic language, and even physical touch—have been shown to significantly reduce user frustration, increase positive affect, and decrease loneliness over time—impacts that are arguably among the most important objectives of positive computing at the individual level.

Expert Perspectives - 공감과 다수성(Plurality)을 위한 테크노롤지
Compassionate Computer Agents

Fig. 1: Timothy W. Bickmore, Northeastern University
Figure 10.1
Timothy W. Bickmore, Northeastern University

Health represents one of the most primal human needs, and technologies designed to positively impact health thus compose an important aspect of positive computing. Over the past decade, many researchers have investigated the development of automated caregivers that promote physical and mental health in people who interact with these systems. Many of these researchers believe that the most promising design methodology is to emulate the best role models we have—expert, compassionate, human caregivers—with as much fidelity as possible. The resulting systems are conversational, communicating with people in some form of natural-language dialogue, often accompanied by a virtual or physical representation of a human caregiver’s persona.

Media used in these systems span voice only (interactive voice response), animated agents (also known as “embodied conversational agents”), and social robots. The anthropomorphic qualities of these interfaces not only serve to make them approachable by a wide range of users but are essential for conveying many of the subtle cues used by human caregivers, such as empathy and compassion. These systems are necessarily limited in scope, but they promote some narrow dimension of health and wellbeing among their users, typically over the course of many conversations that can last weeks or months.

Most of the automated caregivers developed to date address physiological and mental health needs, including interventions for medication adherence, exercise and diet promotion, cancer prevention, prenatal care, and depression counseling. Most of these systems have demonstrated significant improvements in health relative to nonintervention or standard-of-care control groups, and many have been shown to be equal to interventions performed by human caregivers.

An increasing number of automated caregivers are being developed whose primary goal is to provide companionship, typically for older adult users who live alone. Although these agents may provide health or wellness coaching or cognitive stimulation activities, they also engage in a wide variety of social behaviors whose goal is to provide a feeling of social support among their users, which has been shown to be an important factor in decreasing mortality among isolated elderly. One study demonstrated that a companion agent provided to isolated older adults significantly decreased loneliness after a week and that the more users talked to the agent, the less lonely they were.

In addition to general anthropomorphic qualities, these agents use specific human behaviors to improve their effectiveness and their users’ sense of wellbeing. Social bonding between a caregiver and client (referred to as “therapeutic alliance”) is crucial in maximizing outcomes for all kinds of health interventions. Agents that use behavior designed to improve social bonding and therapeutic alliance have been referred to as “relational agents,” and several studies have shown relational behaviors can lead to significant increases in engagement with these agents. One of the most important agent behaviors for social bonding is the display of empathy, comforting, and compassion for the user when they are in a distressed state. Even though these behaviors are simulated and thus do not exhibit “true empathy” (by many definitions), a wide range of agent behaviors—including facial displays of concern, empathic language, and even physical touch—have been shown to significantly reduce user frustration, increase positive affect, and decrease loneliness over time—impacts that are arguably among the most important objectives of positive computing at the individual level.

Computers and the Condition of Human Plurality

Fig. 1: Mihaly Csikszentmihalyi, Claremont Graduate University
Figure 10.2
Mihaly Csikszentmihalyi, Claremont Graduate University

It’s been half a century since Marshall MacLuhan wrote his treatise about the media, which was often wrong-headed but contained some truths as well. As is usually the case, the really important contributions of his thinking seem to have been forgotten. Or at least they have not been applied very much to the new media that have arisen since his writing.

If one just takes the most memorable line from MacLuhan’s work, “the medium is the message,” and applies it to computers and to the Internet, some interesting issues are foregrounded.

Let us just think about how computer-mediated communication differs from interpersonal face-to-face communication. The latter is the expression of what the social philosopher Hannah Arendt has called the fundamental fact of the human condition: plurality. What makes us human is that we can communicate our individual uniqueness to each other, and in so doing we can grow beyond the boundaries set by our genes and early environment. The only reason we can say, with Walt Whitman, that we contain multitudes, is that we are open to information coming from people who are so much like we are, yet also so different.

This information, which makes us who we are, is not contained just in the meaning of words. For us to take the information seriously, we must feel that the sender is to be trusted. And trust depends on many things: we now know it depends in part on how the person in front of us smells as well as on looks, expressions, tone of voice, and so on. After all, we are not going to take seriously messages sent by an unknown source or by someone who is completely different from us.

What is essential to the condition of plurality is this fine balance between similarity and dissimilarity among the partners of the exchange. If there is too much similarity, learning something new is unlikely; if there is too much dissimilarity, the information is unlikely to be relevant. For growth to occur, the information that contradicts our beliefs and prejudices must come from trusted sources.

What is missing from the so-called social media is the condition of plurality. Social media are social, but rarely plural. In other words, in order to maintain contact on the web, the parties usually try to emphasize their similarity. But because the communication relies only on the written word and not on the looks, smells, and other dimensions of physicality that facilitate openness and trust, each party essentially remains himself or herself, without having the growth-producing experience that an encounter among humans has, at least when the conditions for it are right.

Of course, we can and do learn a great deal from computer-mediated information—especially when the source has been established as trustworthy and when the information we seek is factual and objective. But some of the most important knowledge we seek is existential, or subjective and dialogical—that is, knowledge that helps us become fully grown persons. The computer is still not a very good medium for gaining that kind of information. Will it ever be? I am not a betting man, so I will stop right here.

Computers and the Condition of Human Plurality

Fig. 1: Mihaly Csikszentmihalyi, Claremont Graduate University
Figure 10.2
Mihaly Csikszentmihalyi, Claremont Graduate University

It’s been half a century since Marshall MacLuhan wrote his treatise about the media, which was often wrong-headed but contained some truths as well. As is usually the case, the really important contributions of his thinking seem to have been forgotten. Or at least they have not been applied very much to the new media that have arisen since his writing.

If one just takes the most memorable line from MacLuhan’s work, “the medium is the message,” and applies it to computers and to the Internet, some interesting issues are foregrounded.

Let us just think about how computer-mediated communication differs from interpersonal face-to-face communication. The latter is the expression of what the social philosopher Hannah Arendt has called the fundamental fact of the human condition: plurality. What makes us human is that we can communicate our individual uniqueness to each other, and in so doing we can grow beyond the boundaries set by our genes and early environment. The only reason we can say, with Walt Whitman, that we contain multitudes, is that we are open to information coming from people who are so much like we are, yet also so different.

This information, which makes us who we are, is not contained just in the meaning of words. For us to take the information seriously, we must feel that the sender is to be trusted. And trust depends on many things: we now know it depends in part on how the person in front of us smells as well as on looks, expressions, tone of voice, and so on. After all, we are not going to take seriously messages sent by an unknown source or by someone who is completely different from us.

What is essential to the condition of plurality is this fine balance between similarity and dissimilarity among the partners of the exchange. If there is too much similarity, learning something new is unlikely; if there is too much dissimilarity, the information is unlikely to be relevant. For growth to occur, the information that contradicts our beliefs and prejudices must come from trusted sources.

What is missing from the so-called social media is the condition of plurality. Social media are social, but rarely plural. In other words, in order to maintain contact on the web, the parties usually try to emphasize their similarity. But because the communication relies only on the written word and not on the looks, smells, and other dimensions of physicality that facilitate openness and trust, each party essentially remains himself or herself, without having the growth-producing experience that an encounter among humans has, at least when the conditions for it are right.

Of course, we can and do learn a great deal from computer-mediated information—especially when the source has been established as trustworthy and when the information we seek is factual and objective. But some of the most important knowledge we seek is existential, or subjective and dialogical—that is, knowledge that helps us become fully grown persons. The computer is still not a very good medium for gaining that kind of information. Will it ever be? I am not a betting man, so I will stop right here.