Cyclic Perception

Perception is the activity through which living beings (which, for convenience and generalization, we will call agents) acquire information from the external environment; from a relatively superficial analysis, one might, therefore, think that a mechanism feed-forward governs the whole process, that is, that there is a single stream of data coming from the context in which the agent is located and directed toward it.

In this sense, what we call “activity” loses all inherent characteristics to give rise to a completely passive mechanism; the agent receives information without the will being able to intervene, and there is no possibility of selecting sources. This type of approach accords well with the theory of communication formulated by Claude Shannon and predicts the existence of a sender, a channel, and a receiver; of course, the latter knows only the structural characteristics of the message passing through the channel but is unable to predict precisely how it will perform. Applying these principles to perception, it is tempting to attribute the sender’s role to the context and that of the receiver to the agent who, through the organs responsible for capturing information, has only the task of decoding and processing incoming flows to meet specific objectives.
For example, if we consider a sound message, we might say that the ear is the organ functionally suited to transform pressure changes into electrical signals that, once conveyed from the auditory nerve to the brain, become the subject of a series of processes that culminate in full awareness of what is being heard.

In reality, the issue is quite different, and one can easily realize this first by considering the different types of sound messages: background noise, peculiar sounds, music, spoken language, gibberish, etc.

If perception were based solely on serial analysis, each of the above alternatives would have to follow the same procedures, and possibly only in the final stage of processing could it, without any guarantee that it would do so, be channeled to different pathways which would have as their destination particular brain areas suitable for a precise action on the percept.

In the case of natural language, one might expect the ear-brain pathway to terminate in Wernicke’s area (responsible for semantic comprehension), while conscious listening to a piece of music that arouses particular emotions would refer to areas located predominantly in the right hemisphere.

But what criteria are used for such differentiation to take place?

A linear pathway of information flows has as its fundamental characteristic the isomorphic processing of data: in other words, the electrical signals traveling along the nerves, according to this view, are received by the brain always disregarding content and form; the percept, therefore, must arise as an autonomous entity, only at the end of the processing chain.

Such an approach, valid for communication systems, has, however, been shown to be fallacious and inadequate to the conscious perception characteristic of humans; in particular, Ulric Neisser has pointed out a variety of factual data emerging through laboratory experiments that show undoubted “activity” of the subject (understood as the logical opposite of passivity) toward the same incoming flows.

To better understand this point, let us take the example of a conversation between two people and assume that both speak the same native language. Still, neither is initially aware of it: when subject A begins to speak, a stream of information in the form of acoustic signals reaches subject B’s auditory apparatus, which begins decoding with the ultimate goal of understanding the meaning of the words (of course, the first stages of the process serve to identify the type of message received, only after the acquisition of awareness does the actual decoding take place).
At this point, it seems clear to me that, should the point-to-point feedforward view be accepted, it would be irrelevant whether the two interlocutors speak the same language since what would matter is only the flow of data.

One should be most concerned about its intelligibility and any background noise that might alter its informational content, and so, for example, for an Italian to listen to a compatriot or an Arab would be perceptually the same thing. Of course, anyone who would disagree with this view would bring as an example just the learning of foreign languages, for it is not enough to know grammar and a set of words to be able to understand English or German perfectly: it is also necessary to master inflections, to be able to separate terms that seem to be united (think of the French liaison), etc. In other words, the receiver must be able to “format” the incoming data and make perceptual anticipations about them.
At this point, one should ask how this kind of discrimination is possible within an entirely passive scheme: the answer is clearly no, for it is logically unthinkable for any action to occur before processing. For the data to be shaped, it is necessary for the collection of the data to take place according to a particular criterion (or, to use Neisser’s words, a pattern) that must guide the exploration of the data and that is, at the same time, modified by the data. In other words, the variations in the schemata are equivalent to the continuous change of the modus percipiendi in that they represent not so much the type of form taken by the information but rather the process of acquisition itself; we will clarify this particular point of view later.

Returning to our example, a question arises: when does interlocutor B realize that A speaks the same language as him? Hearing is perceptually governed by a variety of schemas that serve to deal positively with all the different situations we have mentioned above. Without a doubt, we can say that learning a language has as its ultimate (and optimal) result in the creation of a main schema suitable for the gathering of information encoded in that particular idiom; to put it better, our ability to understand dialogue is developed by starting with a very crude initial exploration and proceeding through refinements that must culminate in the choice of the most appropriate perceptual schema.

Subject B hears sounds, and his consciousness prompts him to analyze them. After a very few moments, he will have modified the basic schema because, from the first information, he realized that the message is linguistic; if A can understand it (i.e. if he possesses the specific decoding structures), the process will guide the interlocutor toward the acquisition of the final schema (at least as far as word comprehension is concerned) that should enable him to give an appropriate form to the formless stream reaching his brain. We can say that comprehension is acceptable when B is in a position to anticipate the information coming from A; that is, the schema will allow correct exploration and awareness of the semantic content of the discourse.
The perceptual process is thus no longer governed by a linear trend but rather by what is technically called a feedback loop or feedback: information is collected through a pattern (forward action), but this pattern is continually modified by the information (backward, backward action) and conditions subsequent acquisitions. The role played by anticipations lies precisely in the return phase when it is necessary to adapt the schema to a possible variation in the type of data that is about to be captured.

For example, if we are in a car and observe, in the vicinity of a traffic light, a slowdown of the cars in front of us, we tend to think that the first motorists started braking because the light had turned orange (and later red) and so we brake as well.

Although we have not yet been able to ascertain whether our prediction is correct, we can “predict” what the senses might sense based on a future perceptual pattern. Anticipatory adaptation is fundamental, and without a cyclicality in the decoding of information flows, it could never take place consciously: in fact, to predict a sequence of values requires continuous monitoring of the error between the predicted and actual data; only in this way can the estimate be kept within acceptable limits of validity.

In the case of natural language, semantic comprehension of a text or speech is strongly constrained by the speed with which one can anticipate and modify one’s patterns based on actual perceptions: a good interpreter must succeed in this task much more quickly than a person who knows only a few words of a given foreign language, and similarly, an automobile driver is forced to anticipate perceptions regarding the spatial placement of other cars much more celery than a driver driving at very low speeds.

The cyclicality of flows is not only related to anticipations, but more importantly, it is responsible for the modification of patterns to achieve a certain perceptual result; to be honest, this concept is not immediately understood and, in my opinion, requires some clarifications that would allow us to identify the nature of the process in question correctly. In my opinion, it is correct to say that the cycle is based on a continuous extraction of information from the input streams to further specialize decoding: in the dialogue example, the listener, after hearing a few words, can understand that the acoustic signals are related to a natural language that he knows and therefore is aware of the possibility of understanding its meaning. Pattern modification accomplishes precisely this state of consciousness (of possibility) since it is by no means specific that the person will use it. Still, at least one is confident that he or she will be able to do so.

The higher the specialization of the schema becomes (such as the identification of a piece of poetry), the greater the contribution of the will to extract information: elementary perceptions are often characterized by “raw” content that can be perceived without much difficulty; in contrast, more refined processes require a relatively high intellectual effort that, in a sense, can be correlated with the computational load of the anticipation ability. After all, anticipation arises precisely from the action of the scheme, so it is not misleading to think that should this prove particularly challenging, the reason should be sought in the level of abstraction at which the scheme itself operates; for example, while it is trivial to understand a sentence spoken in the mother tongue, the same cannot be said in the case of choosing a language known but only superficially acquired. In the former case, the subject can extract the information even if the background noise (i.e., the set of internal or external disturbances that undermine intelligibility) is very high. In contrast, in the latter, considerable attention is required to minimize “losses” due precisely to interference.

Perceptual anticipation can be simple or complicated, and the reason for this is that specific patterns are used frequently and, therefore, mobilize a greater number of resources. In contrast, others play a very minor role in our lives, making the perceptual process much more tiring at the time of use.

In summary, we can say that conscious perception requires the active participation of the subject to make a selection of information, and this process requires a continuous adaptation of the schema used; referring back to Neisser’s statement, the schema can be thought of as the set of an adaptive algorithm and a data structure: the former has the task of appropriately directing the exploration of the perceptual space. At the same time, the latter is the basis for formatting the data.

What has been stated is of fundamental importance for developing cognitive science and artificial intelligence. Indeed, it is the latter discipline that provides the best territory for testing any theory, and it is through it that it is possible to test whether robotic agents can behave in a human-like manner through the implementation of structures functionally analogous to perceptual schemas.

Bibliographical references

    1. AA. VV. Ferretti F. (a cura di), Dizionario di scienze cognitive. Neuroscienze, psicologia, intelligenza artificiale, linguistica, filosofia, Editori Riuniti
    2. Neisser U., Conoscenza e Realtà, Il Mulino
    3. Spinicci P., Sensazione, Percezione, Concetto, Il Mulino


Share this post on: