I would absolutely love to learn how machines learn. I tried once, at that time I had only heard the term on a few different occasions, each when I was in the company of some of my engineering friends. Now of course I can’t turn on the television or radio or visit a website without seeing or hearing or reading about some machine that has learned to sort cucumbers or put the number 8 into rows or some other totally amazing and not at all trivial task. Then there are the endless array of for profit “universities” offering “degrees” in “machine learning”. You can get a twofer if you drop another 3 grand and get your fake degree in another fake topic AI at the same time.
That’s the present day but back then it did not make much of an impression on me. However, something about the combination of those two words (machine) and (learning) struck me as odd. I’m sort of stickler for vocabulary and grammar so I thought it would be fun to look up each individual word that together make the term. First I looked up the definition of machine and then the definition of learning, then I put the two together, and uh-oh it seems something strange had happened, a logical contradiction had arisen. When a term is composed of words that logically contradict one another it is nonsense. How can I learn how machines learn if machine learning is nonsense? See below for details.
Can a Machine Learn?
Machine learning is one of the terms claimed by tech culture that continues to resonate, even among the learned. It is…
I do not think it is possible to gain knowledge of nonsense, unless nonsense itself is considered a form of knowledge. I realize of course that when two words are combined to form a new term they sometimes shed or morph their original meanings to become something entirely new. I would be all for that in this case, unfortunately, it is abundantly clear that the advocates of this “exciting new area on the cutting edge of technology knowledge” mean it exactly as it sounds. Even though machine learning is nothing more than computing with sophisticated algorithms developed with and employing advanced statistical and mathematical methods and models that are tasked with processing huge amounts of input data and outputting information in the form of more data, or predictions, or whatever as determined by their programming. In other words, computing as it has been done for the past 5–15 years depending on whom you ask.
Then we have my old friend the neural network. On the plus side the term itself is not a logical contradiction, on the minus side neural network is in the top four most overhyped, overappreciated, overinvested in, overinflated, terms in all of technology today, and that is saying something given the basal level of hype inherent in the sector. I place neural networks at #3 just after machine learning (current #1) and artificial intelligence (former number #1 now #2), but slightly ahead of our fourth place and falling fast contender (nanofillintheblank). I am just too tired to take the time here to expound on this but please can we stop already. Artificial neural networks cannot learn anymore than a machine can. In fact I would go even further and say that actual, biological neural networks cannot learn. This is an instance of the mereological fallacy. Ascribing to a part of a thing a capability, or trait, or characteristic, that can only logically be said of the whole thing. A neural network cannot learn anything because only a whole human being (or many/most animals), with a (mostly) functioning neural network (i.e. brain, nervous system) can logically be said to learn anything.
Finally, just because a given technology makes use of difficult, or unusual or highly specific mathematical methods like Gradient Descent, Backpropagation, or others does not in any way grant special status to that technology. It certainly does not imply that the technology has or is capable of human like abilities such as intelligence and/or the ability to learn. Last I checked the human brain does not use any of those methods to learn and is clearly intelligent. In fact it was the human brain and human intelligence that developed or invented or discovered (depending on your leanings re: philosophy of math) those mathematical methods in the first place. It is the nature of mathematical methods to be specific to the application for which they were devised. The obscurity of the math required has no bearing on the rarity, power, or importance of the thing it was developed for and is now applied to.