Artificial intelligence has come a long way since its humble beginnings in the 1940s, with recent discoveries catapulting it into the mainstream.¹ The goal of scientists and engineers behind it all has always been to build an intelligent machine equal to a human. That has been the driving force behind all the advancements in AI. Everything popping up in the field, from chat GPT, self-driving cars, facial recognition, intelligent assistants, etc, stems from artificial networks inspired by the human brain. 

These technologies utilize networks made of software or hardware nodes that replicate neurons in the brain.²  They mimic nature by combining artificial neurons with many connections to form a network. As humans, we are only in the early stages of developing functional networks and are just beginning to realize their potential. Current artificial networks are primitive versions of their biological counterparts and are a great starting point for understanding how a network leads to anything significant.

The current wave in AI, sparked by works from Geoffry Hinton and Ruslan Salakhutdinov in 2006, is powered by deep learning.³ These networks place artificial nodes of varying numbers into single rows called layers. The nodes from one layer heavily connect with others in the next layer, forming rows of interconnected signaling units. Each layer processes information and passes the result to the next one in line, refining the solution. The last layer ties it all together to provide a result. Deep refers to the total number of rows the network has or how many layers deep it is.

Here is how it would work if we were to look at a hypothetical fiver-layer deep photo recognition network that can identify images of a cat or dog. The first row or layer of nodes will analyze pixel brightness. The next layer identifies edges for the image, and the one after that evaluates textures and shapes.  Each row examines a particular feature of the image, and the final layer brings it together for an answer.⁴ 

Generally, the first layer is the input, sending signals through the subsequent layers. These are called hidden layers, each solving a piece of the puzzle. They pass information down the line toward the final output layer that collects it. Information cascades from layer to layer, generating a collective response by the time it flows through the network. With our hypothetical network, an image of a cat or dog enters the input, producing a specific pattern that determines the correct answer by the time it reaches the output. 

Modern AI, like Chat-GPT, self-driving vehicles, image generators, etc, are networks of different shapes and sizes made of millions of nodes, hundreds of layers deep. They can do one thing well, usually better than any human, but that is all they can do. Our brain is equipped with general intelligence, meaning we can learn many things. Everyone can learn to ride a bike, drive a car, and play an instrument. Both humans and AI use networks, but ours is far more complex and advanced. 

To be fair, nature had a 3.8 billion year head start on developing a biological network with general intelligence.⁵  Humans have been at it for less than one hundred years and already have artificial networks with the same number of connections as a cat brain.⁶ We are seeing impressive results, but these are early representations of what a learning network can do. Deep learning is only the beginning; as time passes, more discoveries and breakthroughs may lead to artificial general intelligence. These networks have a long way to go in size, connectivity, and structure before they even come close to being in the same league as the human brain.