The Difference between Artificial Intelligence, Deep Learning and Machine Learning

Nowadays, when it comes to new data processing technologies, there are many different terms. One person says they are using machine learning, another calls it artificial intelligence, and some may claim to be doing in-depth learning. What does all this mean?

Although many of these terms are relevant and may overlap in some respects, some key differences may be important, which may help people understand the definitions between them correctly.

Artificial intelligence means letting computers imitate human behavior in some way.

Machine learning is a subset of AI. It includes technologies that enable computers to find problems from data and deliver AI applications.

At the same time, deep learning is a subset of machine learning, which enables computers to solve more complex problems.


A kind of

The following will briefly illustrate the difference between AI, machine learning and in-depth learning through some easy-to-understand examples.

What is artificial intelligence?

Artificial intelligence as a discipline was founded in 1956 in the United States. The goal at that time was, as it is now, to allow computers to perform tasks that are considered unique to humans: tasks that require intelligence. Initially, researchers studied chess and logic problems.

If you look at the output of a checkers program, you will find that there is some form of "artificial intelligence" behind these actions, especially when the computer beats you. Early successes made the first group of researchers show almost unlimited enthusiasm for the possibility of AI, matched only by the extent to which they misjudged how difficult certain problems were.

Therefore, AI refers to the output of a computer. The computer is doing some intelligent things, so it shows artificial intelligence.

The term AI does not explain how these problems are solved. There are many different technologies including regular and professional systems. One of these technologies has been more widely used since the 1980s: machine learning.

What is machine learning?

The reason these early researchers found some of the problems more difficult was that they were simply not suitable for early AI technologies. Hard coding algorithms or fixed rule-based systems are not satisfactory in image recognition or content extraction from text.

It turns out that the solution is not just to imitate human behavior (AI), but to imitate human learning.

Think about how you learned to read. Before picking up your first book, if you don't sit down to learn spelling and grammar, you can only read simple books. Over time, you will read more complex books. In fact, you learned the rules of spelling and grammar from reading. In other words, you process a lot of data and learn from it.

This is the idea of machine learning. Put a lot of data into the algorithm (not your brain) to make things clear. It contains many types of programs that will be encountered in large data analysis and data mining. Ultimately, the "brain" that drives most prediction programs, including spam filters, product recommendations, and fraud detectors, is actually machine learning algorithms. Input a lot of data about financial transactions into the algorithm, tell it which is fraudulent, let it find out which is fraudulent, so as to predict future fraudulent behavior. Or give it information about your customer base and let it find the best way to segment it.

Data scientists can use a range of technologies and languages to write machine learning algorithms, including Java, Python, Scala and so on. They can also use pre-built machine learning frameworks to speed up the process.

With the development of these algorithms, they can solve many problems. But some humans still find it difficult for machines to recognize simple things, such as voice or handwriting. However, if machine learning imitates human learning, why not directly imitate human brain? This is the idea behind the neural network.

The idea of using artificial neurons (synaptic neurons are the main elements of the brain) has been around for some time. Software-simulated neural networks are beginning to be used to solve some problems. They show great potential to solve complex problems that other algorithms cannot solve.

But machine learning is still trapped in problems that many elementary school children can easily solve, such as: How many dogs and wolves are there in this picture? How to distinguish raw bananas from ripe bananas? What makes this character cry so much?

It turns out that this problem has nothing to do with the concept of machine learning, or even the idea of imitating the human brain. Simple neural networks, with 100 or even 1000 neurons linked together in a relatively simple way, are unable to replicate the functions of the human brain. If you think about it carefully, you shouldn't be surprised: the human brain has about 86 billion neurons that are very complex interconnected.

What is in-depth learning?

Deep learning is a form of machine learning, which can use either supervised algorithm or unsupervised algorithm, or both. But it uses neural networks with more neurons, layers and interconnections. We still have a long way to go to simulate the complexity of the human brain, but we are moving in this direction.

When you read the progress of computer technology, from self driving car to go super computer to speech recognition, you will find that this is actually one.

Please read the Chinese version for details.