Brain System Theory

Primordial Learning

Small children's cognitive abilities are virtually nonexistent, yet they acquire a tremendous amount of knowledge at a very early age. Besides learning how to use their bodies and exploring the world around them, children also learn the most complex and abstract of human skills: language. What's exciting is that children learn without any formal explanation. Rather, they learn through observation and trial and error.

As we get older, we become more and more dependent on our developing cognitive abilities. We consciously search for rules and create logical connections between facts. We describe the knowledge we acquire using formulas, algorithms, and rules. The formalized representation of this approach to knowledge is science.

There is, however, one serious issue with the cognitive approach to learning: the number of facts we encounter. As the number of facts grows, describing them by formulas becomes quite a difficult task. After a certain threshold of facts, it is virtually impossible to describe them using an algorithm and rules. Science solves this issue by reducing the number of facts either by splitting the problem into smaller chunks or selecting and discarding facts of marginal impact. However, as we collect more and more data, reducing or filtering the number of facts can be a similarly complex task. To solve the situation, scientists who were inspired by the human brain created the artificial neural networks.

What is the fundamental difference between the algorithmization of a problem and solving a problem using artificial neural networks?

Algorithms and artificial neural networks both represent conserved knowledge about the world. In both contexts, this conserved knowledge lets us ask questions and obtain responses. The fundamental differences between the two are:

  • how the knowledge is internally represented, and
  • how the knowledge was built.

Algorithms descriptively represent the knowledge through formulas or series of logical steps. We can say we know exactly how the algorithm sees the problem. Creating an algorithm requires a conscious process called thinking and depends on the various cognitive abilities of the algorithm's inventor.

Artificial neural networks, on the other hand, represent knowledge using the internal state of the artificial neurons and the connections between them. With the exception of elementary neural networks, we are not able to understand relations between the inputs, outputs, and various intermediate states. The internal representation of the knowledge of artificial neural networks is not descriptive. Artificial neural networks create knowledge by learning, which is a process of constant internal feedback followed by immediate reorganization. Each time an example is learned, the neural networks get closer and closer to the "correct understanding" of the problem. The beauty of the solution is that the process of learning is a purely mechanical one and does not involve any cognitive activity.

While the algorithm inventor has to be, to some extent, an expert in the relevant area, we can train the same artificial neural networks to work in different domains. Like children, neural networks learn by trial and error. They shine where human cognitive abilities have failed to come up with a reasonable algorithm.

Chick sexing is an interesting example of adults' non-algorithmic learning. Chick sexing is the method of distinguishing the sex of day-old chickens. The process depends on subtle visual clues, which are not easy to explain. In one approach to teaching it, professional sexers stand behind the students, and their only work is to validate the student's choices with simple yes or no answers. Clues are not described to the students, but after a few weeks of training, the students are able to separate male/female chickens correctly. All knowledge is acquired without cognitive skills.

The above example confronts us with an important question: algorithmization is the manifestation of cognitive abilities, but what human qualities do artificial neural networks represent?

Let us look at human unconsciousness. When we talk about unconsciousness, we think of a place within our psyche, the source of thoughts and actions that come virtually out of nowhere and without any reason. It is a dark place of our mind with a constant flow of often chaotic emotions and feelings. Unconsciousness can confront us with desires we are afraid of. There is the feeling of something within us taking control of our thoughts and lives. It is very natural to be frightened of, or at least have respect for, something that is beyond our conscious reach. Unconsciousness is the mysterious land we both want to understand and are afraid of—it is our forbidden chamber.

Or is it not?

Perhaps the unconsciousness is not the dark and scary place we tend to think of. The reason for unconsciousness to exists is the same as the reason for artificial neural networks to exist. It is the number of facts that has to be processed.

The brain has to guide us through our time-constrained lives, taking into consideration social norms, personal goals, a vast number of everyday tasks, our own needs for life preservation and continuation. It's our primordial learning system.

The primordial learning system is the counterpart of the artificial neural network. After all, artificial neural networks were inspired by the human brain. They represent a system of learning that does not involve cognitive tools. We have the primordial learning system from the beginning of the time. It is the same basic learning system all living beings have.

However, just as cognitive learning has an issue with large numbers of facts, the primordial learning system has a problem with structure. As mentioned above, the representation of knowledge in neural networks is by neurons and synapses state. The same goes for our unconsciousness. The knowledge of unconsciousness is unstructured and not descriptive. It is just there.

In summary, there are two systems that help us learn about the world: the primordial learning system and the conscious cognitive learning system. The cognitive system has an issue with large numbers of facts, while the primordial learning system overcomes this limitation at the expense of structure. In the next essay, I will describe how knowledge flows from one to the other, forming a knowledge loop.