Today, the scientist Fernando Reinach, a former professor of biochemistry at the University of São Paulo, wrote in his weekly column, at the newspaper O Estado de São Paulo, an article entitled “The brain isn’t a blank paper”. In his article, he does some considerations about the innate capabilities of the brain and argues why teachers and educators don’t take advantage of them, learning more about the findings of modern neuroscience .
He starts his article reporting a cute behavior displayed by his little toddler : during this morning breakfast, he observed his two-year-old son handling a spoon when he noticed, apparently at the first time, his face reflected on the convex side of the spoon. After some moment contemplating the image of his own face reflected, the boy turned the spoon to the other side and looked at its concave face. Fernando noticed a remarkable reaction of surprise in his son’s face, as his eyes brows revealed, when the boy found his reflected image now inverted. Then the kid started playing with it, turning to the convex side and back again to the concave, over and over, figuring out that one of the sides was constantly inverting his reflection. Finally, he concentrated on the concave side and did the unexpected: he rotated the spoon in such a way that the handle, which was first pointing to the ground, should then be pointing to the ceiling, as he would be trying to “uninvert” the image. Fernando then noticed that the boy looked somehow frustrated because he couldn’t “correct” his reflex to the “right position” by this expedient, since he then had thrown away the spoon, seemingly bored with the loose of time.
Fernando used this fact to illustrate the principle that the brain of a child isn’t, as many believe, like a blank paper, where parents and educators “write” the teachings and explanations. There are some, or rather, many innate abilities and presumptions about how things would behave, and these can be exploited with advantage during education. The blank paper metaphor is quite ancient, dating to the epoch of Aristotle. It was after called tabula rasa, a term in Latin denoting the idea that at birth a child hasn’t any knowledge about the world. However, studying the behavior of babies and toddlers, one can find signals that there is an innate knowledge concerning primary assumptions about the rules that govern the physical world. What is not known is exactly what could be these primitives. However, it is possible to observe that the babies quickly develop expectations, like “one object cannot occupy the place of another at the same time” or, “the light comes from above” , and “objects fall downwards”, etc. These expectations presuppose the existence of some prior knowledge, which we just referred as the primitives. One can say that the baby starts with some primitive innate knowledge that we could call the initial state of knowledge. From this beginning, the knowledge is constantly expanded and rebuilt, as more experience is acquired. And this process of continuous re-elaboration follows ahead during the childhood, adolescence and so on, to the adulthood.
Piaget stated that there are four sequential stages during the development of the child, each one characterized by a series of constructive operations, that assemble, step by step, the cognitive abilities. There are some disagreements with this theory, related to the evidences of modularity and parallelism in the perceptual and cognitive processes in the neuro-sensorial system. It is observed that the order of development of some operational and functional abilities can be somehow different from a child to another, inside the stages and between the stages.
The mirror stage and the mirror test
During the toddlerhood, there is a phase called the mirror stage, that is related to the moment when the self-recognition arises. A very simple experiment with babies and toddlers, that can be done by parents, is the mirror test. First, you have to paint a red spot on the forehead of the child, with a size that he can see it easily. Them, put him in front of a mirror. A baby that hadn’t yet developed self-recognition will try to touch the spot in the “other” baby (i.e., the image reflected in the mirror), as he would be trying to “clean” or check the spot on the virtual baby. On the other hand, toddlers are generally able to recognize themselves on mirror images, and they show a different reaction: in this case, the child touches his own forehead, to check or to clean the spot.
The mirror test can also be performed with animals, to verify their cognitive capacity of self-awareness. However, the results with animals are often controversial. There are other factors that can lead the animal to fail in the test, besides an eventual lack of self-awareness. Among these factors, a possible color blindness or a limited stereoscopic vision can affect the animal capacity in identifying itself in the mirror image. These problems can eventually show up in the human case. Besides these visual impairments , behavioral factors can also influence the test result: in many species, the eye contact with other individual is a cue for threatening, and it can be a disturbance to the self-recognition in the image.
There are two aspects to consider when analyzing the results of the test :
- to fail in the mirror test means that the individual is not self-aware ?
- to pass in the mirror test means that the individual IS self-aware ?
The failure case, as we considered before, doesn’t mean that it was a consequence of lacking of self-awareness, but it could be due to other causes. The tough question is the second, the pass case. In the human case it seems natural to admit the intervening of the self-awareness to produce the result of the test. However, when considering other animals, one could be reluctant in admitting the presence of self-awareness in many animal species. We expect that some species eventually could have it, but others no.
To create a test for the detection of self-awareness, we need some definition of self-awareness. It is very difficult to define it, however we can establish certain requirements that seem to be fundamental if we are going to find self-awareness on an individual. For a creature to be self-aware, it seems reasonable to require the fulfillment of the following requisites:
- It must be able to perceive the patterns of its own behavior, separating them from other events. This is equivalent to say: “These observations refers to a separate class of events”. This separate class of events are its own actions.
- It must be able to behave consistently with the hypothesis that it can correlate the perceived patterns of its own behavior with its own actions. This is equivalent to say: “These observations occur at the same moment as these actions”. These observations are the consequences of its own actions.
- It must have a sense of causality. Or, it could say: “These observations were caused by these actions”. In order to do this, the creature must be able to make experiments. To make an experiment, it must be able to control variables, that is, select a specific value of it and check its effect on an observation.
- It must be able to categorize the causes, labeling them to a same origin. In other words, the observations and the actions that caused them can be categorized as an agent. This is equivalent to say: “The source of these facts is unique”. That is, it must “conceive the agent”.
- It must control the agent. The self-awareness is something emerging from this control. It exhibits the intention to make the agent perform specific actions.
Now, let us check if the passing case of the mirror test provide issues that fulfill these requirements. If an individual or animal is said to have passed in the test, it is because: (i) the subject displayed a behavior consistent with the identification of the red spot (or something similar) on its forehead and (ii) it gave signs of disapproval of the presence of the spot on its forehead (e.g., tried to clean it). In order to draw the conclusion that the red spot is on its own forehead, the subject has to see its head moving and bringing the spot with it. So, it separated the movements of its forehead from other events that it could perceive at the same time (1). It moved its head, following the movement of the spot (2). If it performed specific movements with its head, expecting to bring the spot to a specific position, following its head, then (3) is also satisfied. This should be considered if we agree that the animal passed the test. However, (4) and (5) are not evident. This experiment is not telling us much about the conception and control of an agent.
Robots taking the mirror test
In the field of machine intelligence, there is a great desire to improve the cognitive properties and abilities of machines, such as robots. One could consider that it should be very interesting if a machine could exhibit self-awareness. So, why not try the mirror test as an experiment to check the claim that some robot could be self-aware ? This was exactly what Takeno, Inaba and Suzuki had done at Meiji University.
This work was reviewed at Conscious Robots by Raúl Moreno. In his words: “The robot is able to recognize its own image reflected in a mirror without confusing it with the image of another robot with the same physical aspect. The mirror image cognition system is based on an artificial neural network. The aim of this system is to recognize and differentiate robot’s own behavior from other robot’s behavior. Takeno also suggests that imitation is a proof of consciousness as it requires the recognition of other subject’s behavior and then the application of that behavior to oneself.”
Let us check if the requirements for self-awareness can be fulfilled by the test in this case. If the robot can separate the region in the image in the mirror that corresponds to its own reflection in a video sequence, and maintain it correlated with a consistent set actions, it can certainly fulfill (1) to (3). If it can also group the actions-observations causal pairs in classes consistently with (4), then the robot would be able to conceive agents as categories of behaviors+explanations. But how can one say that the robot can also control the agent that corresponds to itself ? It could control activities that are consistent with the agent as uniquely distinct from the others. However, is the robot selecting its own actions because they are consistent with the agent actions (i.e., consistent with the idea of an agent) ? Or is it simply performing the control action in order to observe specific results in the estimation of a perceived image ? For instance, it could select the actions that bring the centroid of the pattern to a specific trajectory , without association that it is controlling the agent that it perceives. In order to claim that the robot is controlling the agent, we must be able to say that the control must be consistent with the idea of an agent.
Well, in my point of view, more specific experiments should be devised to answer these questions.