Neural network structure and function result in information processing and learning.
Introduction
How come you can complete the last word in this ________? How do you know it’s your friend that just entered the elevator, even though you only saw him for a split second, and from behind? And how can you tell a cat from a dog, every time, even meeting ones for the first time? If you were asked to explain which features allow you to tell a cat from a dog, for example, you might have difficulty knowing what your brain focuses on to make this determination. And yet, your brain draws these types of conclusions all of the time, usually accurately, even though we generally remain unconscious of how our brains exactly achieve these things. This routine ability of our brains to process information and learn from experience over time in order to draw accurate conclusions is essential to the daily success of humans and to human society overall.
The Strategy
In human (and other) brains, information gets analyzed through sequences of groups of neurons, known as a neural network. The number of neurons in these interconnected networks gives brains a variety of pathway options through which electrical signals can travel. For a given amount of information, multiple signals can travel in parallel through a network, establishing connections between different neurons and neural groups in numerous possible combinations. This structural complexity and its resulting versatility is vital, because it allows for the successful accommodation and processing of a large amount of distinct and detailed stimuli coming from the outside world.
Another critical feature of these network pathways is that they modify themselves based on the outcome of their processing (i.e., they are self-optimizing). When a young child misidentifies a cat by calling it “dog”, and is corrected by an adult, the child’s brain automatically changes the course and relative strengths of the pathways of electrical signals traveling through the neural networks involved in distinguishing cats and dogs. Over their development, the child repeats this process of neural pathway modification until they reliably and correctly identify cats from dogs, recognize even fleeting glimpses of their parents, learn to drive a car, and do thousands of other things that we call “learning”. The self-optimization of neural pathways continues for all information of interest over the course of a person’s life.
The Potential
These two features of human neural networks––a high variability of pathways through the sequential structure of neurons, and the process of self-optimization in these pathways based on the relative success of the pathways’ outcomes––are the two seminal features used by artificial neural networks (ANNs) in computational technologies. By emulating these fundamental aspects of human neural networks, ANNs (which use numerical values to represent network pathway and strength/weighting of path sections) can often process information as accurately as human brains can, and can sometimes meet or surpass human brains in their analytical ability, speed of processing, and/or capacity to generate content. ANNs form the underlying structure and function of most artificial intelligence (AI) today, from AI that can identify and categorize, as well as generative AI (such as ChatGPT). Whether it’s helping people decide what to make for dinner based on what’s in their refrigerators, or determine if a skin mole is cancerous by analyzing pictures taken by a cell phone, artificial neural networks owe their expanding capabilities to the human neural networks they .
