Over the past 3 billion years, life has evolved from single-celled organisms to humans, the most intelligent living beings on the planet. Over the past 80 years, information technology has also evolved and sought inspiration from human intelligence, more specifically from neural networks, to develop more advanced software.

Artificial intelligence emerged almost 100 years ago from scientists' hypothesis that machines would be capable of complex reasoning. It would be natural to think that, for this to happen, machines would have to function like the brain. Therefore, we looked to biology—more specifically, anatomy—for the sources to build artificial neural networks.

Follow along with us in this article to understand how we "discovered" neural networks and how they inspire some of today's most widely used applications!

From biology to technology

Although it may sound obvious, it is worth clarifying: strictly speaking, a neural network is a network of neurons, the cells that make up a large part of our nervous system. The point worth noting here is that, unlike some other tissues in the body, nerves form a network. Understanding them in this way changes our understanding of how human thought works.

In the second half of the 19th century, research in the field of health identified the morphology of neural tissues. In 1851, German physician and professor Henrich Müller identified the existence of vertical fibers of nervous tissue crossing successive layers of the retina. In 1872, German physician Theodor Meynert published his "Treatise on the Brain of Mammals." Studying cases of dementia, Meynert described a group of hyperchromic magnocellular neurons and suggested how this structure is associated with memory and cognition. And in 1875, English histopathologist Herbert Major described the six-layer structure of the primate cortex.

Understanding that cognition occurs, in part, thanks to the morphology of the brain was fundamental to understanding how reasoning is processed. If human intelligence exists because of the connections between layers of neurons, for artificial intelligence to exist, it would be necessary to create "neurons" connecting to each other in networks that cross layers.

Based on this idea, Walter Pitts and Warren McCulloch, in a 1943 article entitled "A Logical Calculus of the Ideas Immanent in Nervous Activity, " proposed a simplified mathematical model to try to illustrate how the human brain supposedly works. Pitts and McCulloch's work suggested a way to build a "thought structure," an artificial neural network.

Six years after Pitts and McCulloch's classic work, Canadian psychologist Donald Hebb, in his book "The Organization of Behavior," proposed a theory of learning. However, before exploring Hebb's findings, it is important to discuss the most important feature of a neural network: its ability to learn.

Neural networks: understanding “learning”

Observing how the mind works has been a concern for different societies around the world: Egypt, China, Greece... The oldest historical record of this concern is possibly a collection of maxims, or advice, produced by Ptahhotep, vizier to Pharaoh Djedkare Isesi, about 4,500 years ago. In one of these maxims, he states: "if he who listens hears fully, he then becomes he who understands."

The senses—in this case, hearing—are an input for understanding. Two thousand years after Ptahhotep, Theaetetus, in a dialogue about knowledge with Socrates and Theodorus, suggests that knowing is perceiving through the senses. In response, Socrates and Theodorus argue that the ease of learning requires quick understanding and a good memory.

Thousands of years ago, Ptahhotep, Socrates, and other philosophers associated understanding and learning with perception and memory. These lines of thought advanced through different sciences to the point where we now understand that mental processes derive from our biological makeup.

As we said at the beginning of this text, by studying the morphology of the brain, we discovered neural networks. From this advance, we understand that reasoning occurs because of the connections between layers of neurons. However, synapses explain only part of how intelligence works. How is learning generated from them?

It was psychology that provided these answers, more specifically the theory of Donald Hebb, the Canadian psychologist and professor at McGill University in Montreal. In his theory, Hebb introduces the concept of "cell assemblies": groups of neurons that work together as a processing unit. According to Hebb, it is in "cell assemblies" that perception is transformed into "information." In addition, he proposes that "assemblies" that are repeatedly activated generate structural or metabolic changes in that set of cells, allowing memory to be stored in a stable manner.

With the memory stored, then it is possible to learn. According to Hebb's theory, persistently repeated activities enable the growth of the volume of connections and the formation of other, broader "cell assemblies." Thus, the moment we learn is when we associate information that was initially memorized in more restricted "assemblies."

If we are living in the age of artificial intelligence, then we are also living in the age of neural networks. They currently make up the architecture of applications that perform various activities: large-scale testing, automation flows, image generation, data protection... Practically everything that applications with other architectures do, neural networks also do.

If you want to experience the power of neural networks in your company, contact us to learn about our AI solutions. They bring rapid positive impact by connecting to your company's actual operations. Click here and talk to us!