When I first started doing this blog several months ago, one of my first topics was "nanny-bots"--- hypothetical robots built to do child care. An article about this by Noel Sharkey had appeared in the journal Science, and a lot of discussion had ensued. Now Noel Sharkey and Amanda Sharkey are publishing another article, "The crying shame of robot nannies", in the journal Interaction Studies, and I've been asked to submit a commentary on the article. So, I think I'll use this space to "noodle" a bit and go over some relevant issues.
To build a child-care robot that does a good job, and whose care results in normal development for the child in question, we need to know two broad categories of things. One is how to build a very sophisticated robot that can respond to sounds and sights (and actually a sense of smell would be quite helpful too). The other is to understand a great deal about early development, and especially to know how caregivers' actions influence infants. I'm afraid that at this point we know a good deal more about the robots than about the babies. Or, to be more accurate, whereas we know some things about groups of babies, we don't have very clear ideas about how things work in the lives of individual babies.
Although nanny-bots could easily "watch" or "mind" children (for example, sounding an alarm or creating a barrier if a child approaches a dangerous situation), and although they could entertain children by providing music, videos, or even interactive games, much more than that is provided by a human caregiver. For example, as emotional attachment of child to adult develops, the adult "woos" or "courts" the baby, making an effort to start social interactions and please the baby by playful, interesting games and talk. When babies are very young, adults may have to work quite hard to get the baby to respond, and may have to pick up small and subtle cues that indicate they are doing the right things. They also have to stop if the baby shows that the stimulation is too much or that it's time to feed or sleep instead. What's more, different babies give different cues, and adults have to get to know a given baby's signals--- so it's hard to imagine that a nanny-bot could be programmed for babies in general and be competent with a specific baby, unless it was capable of "learning".
Discussions of child-care robots usually seem to focus on one robot and one child together, perhaps for lengthy periods. In real family life, though, or even in group child care situations, babies do not spend all their time in one place with one person. One important experience for a baby is involvement in a triad or group of three people, in which there are shifts between interactions of different pairs of people. For example, if Mom and Dad are talking, the baby may make efforts to get each of them to pay attention, by smiling or vocalizing. A parent may then turn to the baby, or they may continue to talk to each other while glancing and smiling back at the baby. The baby learns that social interactions may take longer to occur when there are three people than just two, but that if you wait you'll get your turn-- an important piece of information for later social development. In the same way, the family baby goes out with parents to places where adults interact with each other; depending on the culture, this may involve a trip to the grocery store, or going out to work in the fields.
A final point has to do with the amount of variability a baby experiences when with human beings. For example, when the baby learns language, he or she has to begin by identifying the sounds people make that are sounds belonging to the family's language. We make a lot of noises with our mouths that are not part of the language (cough, hiccup, whistle), and all languages do not include all our possible speech sounds. Babies have to hear a lot of speech and other sounds before they can be sure which sounds make a difference to meaning (that is, are phonemes in their language). In addition, different people create different sound patterns when they intend to speak a particular phoneme, and even a single person makes a sound differently on different occasions. The baby needs experiences with all these different sounds in order to learn which of a whole array of sounds are intended to fit into a particular phoneme category. Similarly, human beings show a lot of variability in their other interactions with babies. A mother may be quick to show joint attention with her baby when the two are sitting together and watching a kitten run around, but ignore the baby's social approaches completely when smoke is billowing out of the oven, or when an older child has a skinned knee. Babies benefit from these experiences of variability because they learn that a lack of social response does not necessarily mean that a person isn't interested in you any more.
It's hard to see how a nanny-bot could do all these (and many more) complicated things, without being as complex as a human---- and we do already have humans, many of whom like to care for children and do it well.
