All* Of Your Artificial Neural Network Questions Answered

Author's preface: Reposting the classic FAQs about the future series from way back in 2017-2018. Part III still holds up pretty well.

1. What is an artificial neural network?

An artificial neural network is a computational model based on (what we think we understand about) the structure and functions of biological neural networks. Information moving through the network affects its structure because it changes in response to this information. Because it changes it can be said to learn — in a sense, based on input and output. The sense it which it can be said to learn is of the non variety. Rocks also change based on (information) input from the environment e.g. rain and wind. Do we say that a rock can learn? I do but only on Saturdays when I hold my igneous rock extension course. At a more basic level an artificial network is a series of wires attached to an electron source with some other electronic components and computer chips controlling the distribution of electrons throughout the network. Code is written that instructs the electrons how to flow through the various wires and other components. The electron source is often a power strip in the computer lab on West Green. Sometimes the networks’s “learning” is interrupted when drunk coeds stumble into the lab looking for the bathroom.

2. How can I get rich off of artificial neural networks?

Once again my young friend of friends you ask the most important question early. Hopefully you have read parts I and part II in the series because if you have you will already know the answer and if you have not I ask what in tarnation is wrong with you dickweed, go back and read them then move to part III, this is a sum-bitchin series, it needs to be read in order if you hope to understand it. Damn, where did you go to school, that fake ass Udemy or something. If you don’t get that reference its your own damn fault for jumping in out of order. Anyway the answer once again is hype.

3. I heard artificial neural networks are able to remember stuff and memorize things and even make associations between concepts, I thought only natural neural networks could do those things?

Who told you that little buddy because you should punch them in the face, they are flat out lying to you. Artificial neural networks do do not remember or memorize anything. They are not capable of the act of memorization or remembrance nor can they make “associations between concepts.” Only a human being with a (mostly) functional biological neural network (brain/brain stem and nervous system) and some or perhaps most non human animals are capable of remembering and memorization. Associations between concepts is a trickier topic and it is not clear that non human animals are capable of this. However, it should be beyond dispute that artificial neural networks are not. The situation is similar for the biological neural network, brain, nervous system, brain stem or any combination of those is/are also incapable of memorization or or remembering. Only of a whole, intact person, and possibly some or most non human animals (maybe, though I doubt some insects) can we say that they have memorized something or have remembered something. To say otherwise is to commit the mereological fallacy, attributing characteristics/traits to a part of a thing that logically can only be attributed to the thing as a whole.To say an artificial neural network remembers or memorizes or associates between concepts is false on two levels: being non alive it is incapable of those things and being only a part of a potential whole it is a mereological fallacy. Patently false and logically impossible. The dreaded double whammy of absurdity.

4. So an artificial neural network is really nothing more than a different kind of algorithm?

You hit the nail on the head buckaroo. It is essentially a model based on sophisticated (actually not all that sophisticated) statistical and mathematical rules that accepts and processes inputs and outputs. Unlike a standard algorithm however, it changes in response to these and a given input at t=0 may equate to output A, where the same input at t=20s may result in output X. Another term for it might be something like adaptive algorithm. In fact at one time this was what artificial neural networks were. However, once the very first engineer/programmer/mathematician working in the area realized he could get that interview on NPR if he said he was working on artificial neural networks, and was stuck on the local news if he referred to his work as with adaptive algorithms, hype world protocols were enacted and the rest is history.

5. You are so damn smart, how come nobody listens to you?

I gotta say you are on fire today kid-o because I have been asking myself that same question for years. Turns out nobody listens to me because I am an idiot who likes to talk out of his asshole. I have no formal training in computer science, engineering, advanced mathematics (beyond differential equations), advanced statistics or any other field that would qualify me to comment intelligently on these matters. That said, formal training, knowledge, and experience are overrated. Who is to say what is right or what is wrong in this topsy turvy world we live in today? It’s a complicated world what with AIs on every street corner, in every home, and machines in every university classroom learning stuff. I can’t walk two feet without some damn AI hitting me up for 50 bucks and then wanting to try and school me on the philosophy of Wittgenstein. Me, the mother fucking Wittgenstein expert. And you know what? they do it, every fucking time. You know why? because they are so mother fucking smart. These God damned AIs know everything. How can I, a lowly human ever hope to keep up? Sorry we were talking about artificial neural networks. I’m done. Data “science” is up to bat next.