Artificial neural networks were designed to imitate biological neural networks, as they are meant to “learn” based on inputs given to it. In a way, artificial networks memorize what it has “seen” in the past and uses these memories influences its future decisions. For example, an Artificial Intelligent being (AI) that has been designed to have a neural network may be shown millions of videos that can be separated into two categories. The first category of videos has humans in it, whereas the other set of videos have no humans in them at all. The AI will be told which videos have humans in them and which don’t, that way it can determine what makes a video with a human in it different from a video without a human. It may see that typically, in videos with humans, there is an ovular shape that has two circular objects within it and a crescent moon-shaped opening‒this is a human head. Then, by creating specific criteria for what constitutes a video with a human in it, an AI with a neural network may then be able to determine whether a human appears in other videos that it has never seen before. This is all done by the AI “learning” what a human looks like and what a video looks like when there is a human shown. There is absolutely no prior knowledge given to the AI.
Now, it is important to understand that artificial neural networks are actually much different from biological neural networks. Biological neural networks found in humans are much more complex than current artificial neural networks, showing that AI currently cannot learn as well as a human possibly could. This may be refreshing news for some of you, as you may fear that a Terminator-like apocalypse could happen if AI ever eclipses human intelligence.
Similar to how a biological neural networks work with biological neurons, artificial neural networks have artificial neurons. Basically, these artificial neurons actually take some sort of input that is fed to it, and then based on that input, it relays an output. An extremely basic example would go as follows: Pretend we are still observing the AI that has been “taught” how to recognize whether a human appears in a video or doesn’t. One artificial neuron may be responsible for determining whether human eye’s appear in the video. So every single frame is given as an input to this neuron. Then, if a human eye does appear in the video, as determined by this neuron, it outputs a specific piece of data. For simplicity, let’s say it outputs a “0” when there is no human eye seen, but a “1” when there is. A 1 or 0 will be output for every single frame of the video, and thus there will be a series of 1’s and 0’s that have been output by this single artificial neuron. This binary series will then be fed into a non-linear which also accounts for the outputs of every other artificial neuron, and then a final output will be given by this function. That final output will show whether there is actually a human that appears in the video or not. This means that while the specific artificial neuron responsible for finding out whether the video has a human eye in it or not is a factor in determining if there is a human that appears in a video, it is not the deciding factor. There may be a few frames in the video where this artificial neuron thinks that there is a human eye, but if the neuron responsible for determining if there is a head or the one that determines if there is an ear does not see these body parts, then it is likely that the human eye neuron is just mistaken. Therefore, the non-linear function may only determine that a human-being is present when multiple artificial neurons recognize different body parts of a human.
As you can see, the concept of machine learning and neural networks is quite fascinating and I recommend that you explore these ideas more on your own. It is amazing that in just a century, we went from basic calculators that could only do simple arithmetic to creating artificial intelligence that can mimic the human brain. Imagine where we may be fifty years from now if the exponential advancement of technology does not slow. Maybe by that point, we will actually have real hoverboards. A man can only wish.