Why Does Machine Learning Work?

By Boris Dzhingarov

Machine learning is the study of powerful computer algorithms that improve automatically by the application of real-time data and through experience over time. It is often viewed as a sub-field of artificial intelligence, with many of the methods used in this growing niche being similar to natural language processing. However, there are significant differences between these two areas of application. Natural language processing is typically a supervised method, whereas machine learning tends to be unsupervised. There are also differences between the two in terms of the resources that are available to the programmers.

One of the most compelling arguments put forth against the effectiveness of machine learning and artificial intelligence is that humans can achieve much better results simply by staying the way they are. Humans can solve problems much quicker than computers. They have higher IQs, better problem solving skills and are better able to learn from experience. The main argument of machine learning and artificial intelligence is that computers can achieve some of the same things that people can, but due to the complexity of today’s computer systems, they will invariably fail. Deep learning, deep artificial intelligence and other forms of artificially intelligent systems cannot compete with human cognitive skills and can often outperform them.

This argument has some merit. If humans could learn how to solve complex problems such as complex calculus, then they would do just that. In fact, researchers have made some progress toward machine learning and artificial intelligent software which may be capable of solving some of the more challenging problems in scientific research. However, the majority of current machine learning methods are unable to compete with and surpass human experts at tasks requiring advanced reasoning and analytical skills, as well as those requiring specialized knowledge that only humans can acquire. For instance, researchers have made some progress in developing software that can analyze past data in an effort to predict future data that may prove relevant.

This type of system has some advantages. An algorithm can function by itself without the need for humans to oversee its development or apply any form of human judgment. Although an algorithm can learn without any supervision, this type of machine learning requires supervised learning. Supervised learning involves a supervised data scientist who provides labeled training data sets so that the algorithm can learn what it is supposed to learn.

In contrast, supervised learning requires two conditions: a human expert to supervise it and a computer system that can understand the intent of the human expert. Computer systems are good at supervising themselves. When the computer system is faced with a problem, it can ask a series of questions and determine its own solution. As long as the computer is run by a person who is capable of explaining what he or she is doing, the computer can typically solve any problem that is presented to it.

Nevertheless, in the case of deep learning, a human expert must oversee the development of the machine learning application in order to ensure that it is not accidentally creating a mistake. Experts in this area are called artificial intelligence specialists. Their job is to catch and correct mistakes and to continue to improve the machine learning system even when other experts have abandoned it.

Some researchers believe that machine learning works best when the algorithm is not supervised by a human expert. In other words, the algorithm is given enough new data and enough time to achieve some sort of accuracy. Unfortunately, not every algorithm can survive this test. This is especially true in self-driving cars that must rely on data collected by the sensors and lasers of other vehicles around it.

Nevertheless, there are many machine learning models that can work well even if there are no human experts supervising their development. They can be given explicit programming which will allow them to adapt to changing situations and circumstances and can also function even if the current computer system does not have an artificial intelligence system in place yet. These models, called supervised machine learning systems, can typically learn from past data and can continue to do so even when other machine learning technologies are actively trying to make the same mistakes.