Machine Learning: A Truthy Lie?

Franck V.

For all these years, we all have been misguided by the term machine learning. We have been told that machines learning makes a machine capable of how to think, how to act like a human. Machine learning is the most misused term. It does not really mean what it sounds like. It is a lie, a truthy lie.

What is meant by a truthy lie?

Each year Merriam-Webster releases a top 10 list of most searched words. In 2003, the top word in the list was democracy. In 2004, the word blog made it to the top. The winning word for the year 2006 was trustiness, "Truth coming from the gut, not books; preferring to believe what you wish to believe, rather than what is known to be true". A word which could be a lie is used so often that it eventually feels like truth. "Bet on the jockey, not the horse" is a truthy lie.

Similarly, "machine learning" has been used over time for any kind of activity to train a machine or a computer so it could think or act like a human. The word is used so often that now it feels like machines are really capable of learning.

If machine learning is not actually machine learning then what is it?

To answer this question, let's first understand what is learning. Learning by definition "the acquisition of knowledge or skills by observation, experience or being taught". Anything which has a capability to learn learns from its environment or experience. Any living thing lives the way it lives because it is grown in such an environment where others live that way. A chick of flying birds would never fly if it is grown with chickens and a cub may never eat meat if it is grown with cows and goats. We speak the language our society speaks and learn a new language from a teacher. Someone from Spanish parents would never speak Spanish if grown in the tribes of Africa. We also learn from examples. We do not touch a kettle if we see someone burning their fingers by touching it. An investor would never invest in a company which has pushed other investors into a loss.

Machine learning is a process in which machines learn from experience and that experience comes in the form of examples. In order to teach a machine how to do a task, it is provided with a lot of examples. It has been told what is right and what is wrong either explicitly or implicitly. This is how machine learning defined by machine learning scientists to the rest of the world.

But behind the scene, machine learning is nothing more than a mathematical function estimation. It is just the math that the human mind is incapable to solve not due to its complexity but because of its amount. A huge amount of simple calculations is required to make a machine able to do even a simple task. A simple neural network may have several thousand if not millions of multiplication operation just to detect a human face in a picture.

Unlike the good-old-fashioned-AI, machine learning makes a computer system able to learn from experience and change its behavior without being implicitly programmed. This statement might sound fascinating to you that how could a computer do a task it has never been programmed that way? Well, that is a truthy lie but a lie anyhow. Each step in the learning process is written by a human. From simple addition and multiplication to the process of learning, everything is programmed in a dumb computer.

Computers today have the so-called intelligence because of their speed and it could be a threat to humanity because of this quality. Artificial intelligence is a war between speed and real intelligence. If something is intelligent enough to make a counter-attack strategy to destroy its enemy but it lacks the speed to implement the strategy on time, its enemy will crush it for sure. That is why I say AI will take this world to its end.