Science Magazine—Pure Reasoning in Young Infants

May 27, 2011

Budapest, 27 May 2011—An international team of researchers (E. Teglas, Central European University; E. Vul, University of California, San Diego; V. Girotto, University IUAV of Venice, Venice, Italy; M. Gonzalez, CNRS, Marseille, France; J.B. Tenenbaum, Massachusetts Institute of Technology, Cambridge, MA; L.L. Bonatti at Universitat Pompeu Fabra, Barcelona, Spain) published a new study in the 27 May issue of Science Magazine (http://www.sciencemag.org/content/332/6033/1054.abstract), in which they found that babies only a few months old have a solid grasp on basic rules of the physical world. One of the co-lead authors of the study is Erno Teglas, from the Cognitive Development Center (CDC), CEU.   

Infants show surprisingly sophisticated abilities in reasoning about the physical world. However, the mental operations that help them integrating different sources of information in order to be able to deal with a large variety of new situations are still controversial.  When facing uncertainty, as it often happens in the early stages of development, infants would be helpless and unable to form expectations about new, unfamiliar situations they have never encountered before in the absence of these abilities. Therefore, if no relevant past experience is available, how do we build predictions about the possible ways an event can unfold? In fact, humans excel at making "one shot" predictions, integrating multiple sources of information to form rational expectations about most of the new situations never directly experienced. The study recently published in Science Magazine probes the roots of this ability in early infancy, which is called "pure reasoning." The researchers showed that the probabilistic inferences involved are powerful and coherent already in preverbal infants.

Examples involving this kind of reasoning abilities are frequent in real life situations. Imagine a table with blocks of different colors arranged in different configurations. If the table is shaken, one block will fall off. If the blocks are close to the edge, the probability of seeing a yellow or a blue block falling can be estimated depending on the number of blocks of each type. If there are 3 blue and 1 yellow blocks, one might guess that the blue block will fall. However, this intuition will change as soon as there is a possible interaction between the number and the spatial arrangement of the blocks. Even in absence of any previous experience with this exact situation adults would find these judgments easy. The list of possible arrangements is basically infinite, thus, statistical relations previously acquired through experience with similar scenes would not be useful in coping with all the possible ways this scene can unfold. While adults might encounter such problems often, such problems are even more prominent in infancy. The researchers’ experiments intended to test whether pure reasoning is available already in young infants, permitting them to deal with a large variety of situations with an increased flexibility in the early stages of development.

In order to test this in infants, the researchers designed situations that tackle on the most important aspects of these problems. Consider a scene in which three yellow objects and one blue object bounce inside a container with an opening on its lower side, as in a lottery machine. For a variable period, an occluder hides the bouncing objects from sight. Following this, one has to estimate which of the objects will exit the container first: will it be a blue or a yellow object?

A scene like this contains different kinds of information about the properties of the objects, such as their color and shape; about the number of objects in the scene; and about the dynamics of their trajectories, collisions and changes in speed. All that piece of information is potentially relevant for predicting which object will exit first. However, the challenge is to select the relevant pieces of information, integrate them in an optimal way and update them if the dynamic of the scene makes some of them irrelevant.

If the container is occluded for a very short time before an object exits, the object most likely to exit first will be the one which is closer to the exit before the occlusion. In this case, the most relevant information for an optimal prediction is the distance of the objects from the exit prior to occlusion. If the occlusion period is long, in a dynamic scene involving objects that continuously change position, information about distance looses relevance. Instead, what counts most will be the number of yellow and blue objects: in the absence of any other information, the most rational bet will be that an object of the most numerous classes will exit the container first (in this example, a yellow object). In the named experiments, 12-month-old infants saw scenes of this kind, which contained different sources of information, and it was tested under what conditions they could use and integrate such different sources to predict inexperienced outcomes.

The role of expectations over behavior is well documented in psychology. In infancy, violations of these expectations are reflected in specific looking time patterns: thus, longer looking time might signal a greater violation of infants’ prior expectations.

In the studies the researchers find that infants’ looking times varied as a function of their degree of belief in an outcome that is in accordance with a probabilistic inference model that optimally combines the information sources described above. For instance, if the occlusion was short, they looked longer if the exiting object was previously seen as far from the exit. If the occlusion was long, they looked longer if the single different object exited. These findings show that optimal predictions require to flexibly integrating different sources of information; to weight their relative importance, and to adapt one's reasoning accordingly. Thus, when facing uncertainty, as often happens in new situations or in the early stages of development, a rational agent would benefit from a mechanism that can take into account all the relevant possible continuations of encountered scenes and make rational predictions.

Source: Erno Teglas, Edward Vul, Vittorio Girotto, Michel Gonzalez, Joshua B. Tenenbaum, Luca L. Bonatti, (2011), “Pure reasoning in 12-months-olds as probabilistic inference”. Science, 27/5/2011.

Some international media coverage includes:

http://www.ibtimes.com/articles/153970/20110529/mit-tenenbaum-barcelona-babies.htm

http://www.medicalnewstoday.com/releases/226470.php

http://www.sciencecodex.com/inside_the_infant_mind

http://oneclick.indiatimes.com/article/03v3d4q5Qs5E1?q=California

 

For further information, please contact Ildiko Rull, Hungarian Media Relations Manager at CEU on 327-3800 or rulli@ceu.hu