I am always fascinated by the sweet smile of my younger brother. I watched him born and grown up. His smile can always me feel happy, because I think that is an expression of recognition and love. Is that true? What do infants mean when they smile? Thus I tried to look for researches that can explain that.

Dr. Daniel Messinger is an associate professor of psychology and pediatrics at the University of Miami. In Just the fact baby he divided infants smile into 4 stages.

For infants from 0 to 1 months, they smile instinctively. The action is described as “occur while the baby is drowsy or during REM stages of sleep. And for 1 to 2 months infants, they smile to respond to circumstantial stimulation, auditory stimulation first, and then visual ones. Baby learn to have face to face interaction when they are 4 to 6 months. “They are learning to regulate emotions and the joy may be too intense,” says Dr. Messinger. And when they are 6 to 12 months, babies can make different kinds of smile, and are able to express the extent of their joy, and babies can sophisticatedly use smiling as a regular communication method.

“Some research shows that they smile most often as they’re falling asleep and waking up,” says Alan Fogel, Ph.D., professor of psychology at the University of Utah in Salt Lake City and author of Infancy: Infant, Family and Society. “Those times are probably inherently pleasurable for babies, and their good feelings seem to create a smile.” (Cited from Parent)

From the two studies mentioned above, It seems like that smiling of babies is rather a spontaneous action or a method to express joy in communication. However, the latest study shows that actually babies smile for some purposes. They always expect you to smile back to them.

According to the study published on Plos, Infants Time Their Smiles to Make Their Moms Smile, and they try to minimize the time they smile and maximize the time their mother do.

“Thirteen infant-mother dyads were observed in weekly face-to-face interactions between the ages of 4 and 17 weeks [8]. Mothers provided written informed consent, and the Purdue University Institutional Review Board approved all procedures. For each of the 13 dyads we computed the probability of 4 hypothetical goals separately for each agent (mother and infant): (1) maximize the time of mother smiling / infant smiling (simultaneous smiling), (2) maximize the time of mother smiling / infant not smiling, (3) maximize the time of mother not smiling / infant smiling, and (4) maximize the time of mother not smiling / infant not smiling. The basic logic of how we computed these probabilities follows the four steps described in the introduction. For instance, in the case of determining the probability of each goal for the infant, we perform the following steps: (1) fit a predictive model of mother’s smiling behavior, (2) hypothesize a particular goal for the infant, (3) compute the optimal smile timing for the infant to realize that goal as well as possible, and (4) determine how well the optimal smile timing fits the empirically observed smile timing. “

And the results are indicated by the figure below. Babies want their mother to smile, but not themselves, when their mothers expect mutual smiling.

However, the most interesting part of the study is that to further validate the results, the researchers designed a sophisticated child-like robot, named Diego-San, simulating face to face interaction with adults. And the robots smile for a period of 3 minutes following the control strategies below:

“1- Infant: The robot smiled using the control policy synthesized from the inverse optimal control analysis of infant smile behavior described in the previous section.

2- Replay: The robot smiles were locked to the timing of the smiles generated during the session with the previous participant using the Infant controller. Notably, there was no contingency between the participant’s smiling and Diego-San’s smiling.

3- Mirror: In this condition Diego-San always matched his smile to that of the participant, as if he were a “smile mirror”.

4 Infant Plus: The smile timing of this controller was identical to Infant with the modification that Diego-San was more likely to modulate his expression to be the same as the participant (elevated probability of matching of 50% per second). This controller was designed to test the effect on the participants of increasing the contingency of the robot’s smiling to the participants’ smiling (as compared to the Infant controller).”

The results come out that “the robot maximized the amount of adult-only smiling just as infants had maximized mother-only smiling”, which confirms the original hypothesis.

The study is convincing. Control theory methods are applied, and the mathematical analysis is rigorous. But the most interesting part of the study, the robot simulation part, has many uncertain things. I would question the effectiveness of the robot. All human simulation designs are based on what we have already know. If we don’t know when and why an infant smile, how could we design a program to simulate his or her reactions. Even if we assume the hypothesis is right, and perform the robot simulation for a probability test, I don’t think there is a computer program that can fully replace a human brain. However, I think the robot-human interaction program is a good try, and if one day there is a perfect copy of human brain appeared in computer, which means we digitalize human brains, it is not far from Artificial Intelligence.