A practical guide to using AI

What are neural networks?

Neural networks are designed to mimic human intelligence using neurons and connections which are the building blocks for human-brains and for neural networks. The neural networks "learn" by changing the strength of the connections between neurons and changing the configuration of those connections and neurons creating structures somewhat analogous to a brain that has specialized parts to deal with some functions -- for example, humans have a very specialized visual cortex that enables it to process visual information.

The way a neural network behaves i.e. functions comes from the relative strength / weakness of its connections and the way that "signals" move through the network getting amplified or dulled at various points.

Roughly speaking, humans have eyes and ears which are each an "input" to the neural network in our heads (our brains) neural networks ALSO have inputs that they receive from an "environment" (basically the universe of information that they have available to them) whether they get this information passively via a camera or, more generally, a data stream or it is fed to them (like speaking to a person directly).

Neural networks take this information, process it through their structure of neurons and connections and return some output "make a decision" / "have a thought", which is also what happens in people -- when you hear an alarm this cascades through the neural network in your head and prompts some output (elevated heart rate, eyes darting for the nearest exit, etc. etc).

The effect of this concept is that a neural network can't learn about something that it can't "sense" (i.e. has or is given data for).

This makes a lot of sense when thought about from the perspective of a person, if a human doesn't have access to an input related to solving a problem it would be nearly impossible for them to get to the right answer -- for example asking a person that is completely blind to figure out what color you're holding in front of them by sight is obviously problematic.

How does the structure of neural networks get built?

Today, the structure of a neural network is often decided by a person creating a neural network somewhat arbitrarily. There are various algorithms to change the structure of neural networks algorithmically, one of them is the NEAT algorithm