Meme Menagerie

An article in this week’s New Yorker describes medical robots under development  which adapt their communication style, when they interact with human patients.  As a result of some very simple strategies of “style flexing,”  these robots are more effective coaches for recovering stroke victims and patients with Alzheimer’s.

Here are some choice excerpts from the article:

(Maja) Matarić’s work on social robots, however, must address a higher level of complexity. “The challenge is to have cognitive models built into the robots, so the robot understands how to motivate people,” (Allison) Okamura says.

(…)

(The patient) said, “When I’m at home, my husband is useless. He just says, ‘Do it.’ I much prefer the robot to my husband.”

(…)

..robots that were programmed to behave as introverts or extroverts. A robot’s degree of sociability was defined by how far it positioned itself from the patient, the speed of its movements, and its type of communication. For people who were more extroverted, Matarić programmed the robot to move close. “We are not talking sociopathically close, because we always maintain three to four feet of safety distance between the user and the robot,” she explained. “But, with the extroverted robots, they move into your area, and talk with a slightly higher pitch, more words per unit time, and they say things that are more forceful, like ‘Come on, you can do three more. I know you can do better than that.’ ” The more introverted robots were programmed to stay farther away from the user, to gesticulate less, and to speak with a slightly lower pitch and at a slower tempo. “You don’t want to make the introversion glaring,” Matarić said. The introverted robots also said more soothing things and offered more praise.

(…)

..Matarić and her co-workers developed an algorithm for learned behavior, with the robot adapting to match the participant’s preferences in terms of therapy style, interaction distance, and speed of movement. The robot was able to gauge the subject’s time and success in performing the assigned task and then to modify its behavior accordingly. “We actually had the robot slightly shift its personality, gradually, while interacting with the user,” Matarić said. This capacity to adapt is called “machine learning.” The programming has to be carefully done, she explained, because “you don’t want the robot to schizophrenically suddenly change. You don’t want it to become a dictator all of a sudden, because that breaks the whole engagement.” She went on, “But over time, with slow changes, you end up somewhere that’s quite different from where you started. The notion of social engagement is to keep people doing something even if they really don’t want to do it. It may be painful, it may be boring, it may remind them of their disability, which is frustrating. But we know you need to not be in your comfort zone, because if you are fully comfortable, then you are not pushing yourself enough.”

Human coaches, Matarić explained, might read a person’s facial expression, but even an intelligent robot has difficulty interpreting nuances in lighting and appearance. Also, patients may mask their feelings. To overcome these problems, Matarić’s team placed galvanic sensors on a band on the subject’s upper arm, whose readings Matarić believes will provide the robot with sufficient information about whether a patient is being challenged or becoming frustrated.

(…)

So, the next time the person you’re working with is “too introverted” or “too extroverted” for you, just  remember:  There’s a robot out there just waiting to take your job.

The article ends with some cautionary notes: 

Sherry Turkle, a professor at M.I.T. who has expertise in psychology and sociology, is concerned about both the stated need for robots and, she says, the risks they pose to “the most vulnerable populations—children and elders.”  

(…)

“..there is no upside to being socialized by a robot.” Based on her observation of groups of different ages, Turkle has found that “children and the elderly start to relate to the object as a person. They begin to love it, and nurture it, and feel they have to attend to the robot’s inner state.” With this attachment and projection of their emotions, Turkle says, people begin to seek reciprocity, wanting the robot to care for them. “We were wired through evolution to feel that when something looks us in the eye, then someone is at home in it.”

Robots, Turkle argues, risk distorting the meaning of relationships, the bonds of love, and the types of emotional accommodation required to form authentic human attachments. She questions whether robots are necessary in the settings that Matarić and others are exploring. “What human purposes are served by fostering these attachments?” Turkle asks. “The benefits have to be extraordinary, and, as far as I’m concerned, the jury is still out. You are dealing in deception about what is fundamentally human—the nature of conversation, attachment, nurturing.” She is not convinced that the elderly in nursing homes need robots. “Why not people?” And she is not convinced that robots serve as a bridge for autistic children to learn how to connect with family or friends. Only a small number of children have participated in studies, Turkle notes, and there are no data on long-term effects. And while Turkle does not doubt the good intentions of roboticists like Matarić, she points out that the direction, if not the purpose, of their research is to produce a robot that can function independently of a human therapist. “Is it something a robot can really do that a person cannot?” she asks. “Why is a machine touching something in us that is so appealing?”

In Turkle’s interviews with people who interact with robots, she has been struck by how many state that they “can’t trust people,” and that the robot offers a safe and secure relationship. “We need to really think through now where we are headed with social robots, whether we really don’t have people for these jobs.” The idea that robots will teach people to relate to others, she says, is as fallacious as the argument that e-mail facilitates telephone conversation and then direct discussions. “People lock into the place where they can hide and feel safe,” she said. “And while we know this with computers, we seem ready to move ahead with robots that are designed to perform in a way so that a person believes there is somebody at home. If the patient actually learns something about himself, then I could imagine that these objects would be valuable. But that is not proven. Right now, it’s a giant social experiment with real risks.”

Matarić is aware that intelligent social robots raise worries about emotional impact. She and her team, in the course of their research, have asked, What happens if a robot breaks down, or is taken away, after the person invests the robot with the qualities of a grandchild or a companion? What if a user begins to treat the robot like a slave, and then extends this destructive behavior to a family member or a friend? And, even if the machines are unaware of morality, robots must be prepared to act ethically. Her team is trying to envisage future ethical dilemmas. For example, if a patient being assisted suddenly needs emergency attention, what is the robot’s responsibility? Matarić is trying to create independent robots that are able to perform the tasks of human caregivers and are capable of displaying empathy toward patients. “But robotic interaction should not replace human interaction,” she said. “It should only improve it.” 

I think this technology provides some good hope/evidence for all the extremely shy geeks out there:  Sociability can be learned.  You just need to improve the accuracy of your sensors and hard-wire your behavior with some good, adaptive algorithms. Run the program from a few years and before you know it, AI  (or in this case, EI), may actually emerge on its own…