What is the difference between AI, Deep Learning, Machine Learning and Natural Language Processing?

In the world of business, science and technology, there is a lot of talk about these concepts, but many times we don’t have a clear idea of their meaning, their relationships and their limits.

In this post, we will define the concept of Machine Learning and its differences with respect to the concepts of Artificial Intelligence, Deep Learning and Natural Language Processing.

What is AI?

The concept of Artificial Intelligence is not new. It has been around the world at least since the 1950s when Alan Turing created his famous test to determine if a computer has real intelligence. Shortly afterward, in 1956, the first Artificial Intelligence Conference took place in Dartmouth, which was the official starting point for this new field of science.

In this conference, the conjecture was raised that both learning and human intelligence, in all its aspects, could be described in sufficient detail to be reproduced by a computer.

Because the fundamental idea on which Artificial Intelligence is based is to get a computer to solve a complex problem as a human would do. Sometimes those “complex” problems for a person are not so complex. For example, for a person, it is very simple as:

* Identify a cat in a photo* Decipher a blurry text or one in which some letter is missing.* Identify a sound* Prioritize tasks* Driving a car* Play a game and win* Or do something creative like write a poem or summarize an idea with a drawing

How does a person solve these tasks?

In order to emulate the process, a computer must add to its capabilities in terms of computing power, processing speed and data storage capacity, new ones that allow imitating the reasoning of the human mind in a functional way. Thus, it needs power:

Capturing information from the environment: Perception

Humans, we communicate with our environment through our senses. Today there are many sensors of all kinds that can perform this function, collecting information from the environment and sending it for processing on the computer. They can even surpass the “human senses” as they are not subject to the limits of our biology.

Understanding natural language (Natural Language Processing): Interpreting spoken and written language

Understand the meaning of a sentence, understand different accents. This is a difficult task since the meaning of a sentence can vary greatly according to its context.

Representing knowledge

This AI capable of perceiving people, objects, concepts, words, mathematical symbols etc., needs to be able to represent that knowledge in its “artificial brain”.

They also need the ability to reason. Be able to connect all that knowledge, data and concepts, in order to solve problems using logic. For example, a chess machine detects the movements of the pieces on the board and applying the rules of chess to the data it has collected, decides the best move. – Be able to plan and move around. To really look like a human, it’s not enough to think like a human. Our AI must be able to move in a three-dimensional world, choosing the optimal route. This is what autonomous vehicles already do, but they must do it well because in this case, mistakes cost lives.

What factors have helped drive the development of AI?

Over the last few decades, what initially seemed like a utopia has become a reality. Several factors have helped drive the development of AI.

One of the factors that has contributed most to the advancement of AI, in addition to the investment of major technologies in R&D, has been Moore’s law. Moore’s law is not a law in the scientific sense, but rather an observation. In 1965 Gordon Moore predicted the continued increase in the complexity of integrated circuits (measured by the number of transistors contained in a computer chip), while reducing their cost.

This allowed the then-nascent semiconductor industry to create the microprocessor (in 1971) and other integrated circuits that were originally applied to computers, but today we can find them in any device (mobiles, televisions, vehicles) or even in living beings (such as identification chips implanted in animals). Thanks to this, Artificial Intelligence applications are now part of our daily lives.

Another factor that has largely driven the development of Artificial Intelligence (AI) has been Big Data technologies. In 2012, Google struck the bell when it proved capable of identifying the image of a cat in a photo with 75% accuracy. To achieve this, it used neural networks that it trained with a corpus of 10 million YouTube videos. Obviously, this would not have been possible without Big Data.

What is Machine Learning (ML)?

The first IA-based programs, such as Deep Blue, were rule-based and programmed by one person. Machine Learning is a branch of Artificial Intelligence that began to gain importance in the 1980s. It is a form of AI that no longer depends on rules and a programmer, but the computer can establish its own rules and learn for itself.

Google’s DeepMind, which managed to beat the world champion Go, did so by applying automatic learning techniques and training with a large database that collected plays from experts in the game. Therefore, it is a good example of ML application.

ML systems work on large volumes of data, identify patterns of behavior and, based on them, are able to predict future behavior. That way they are able to identify a person by their face, understand a speech, distinguish an object in an image, do translations and many other things. It is the most powerful tool in the IA business kit.

For this reason, large technology companies such as Amazon, Baidu, Google, IBM, Microsoft, and others, offer their own “ML for business” platforms.

And how do machines learn?

Automatic learning takes place by means of algorithms. An algorithm is nothing more than a series of ordered steps that are taken to perform a task. The objective of ML is to create a model that allows us to solve a given task. Then the model is trained using a large amount of data. The model learns from this data and is able to make predictions. Depending on the task you want to perform, it will be more appropriate to work with one algorithm or another. The models we obtain depend on the type of algorithm chosen. Thus, we can work with geometric models, probabilistic models or logical models. For example, one of the best-known logic models is the one based on the decision tree algorithm, which we will also see in detail later.

What is Deep Learning?

One of the ML algorithms that arouses more expectation is the neural networks, a technique that is inspired by the functioning of the neurons of our brain. They are based on a simple idea: given some parameters, there is a way to combine them to predict a certain result. For example, by knowing the pixels of an image, it is possible to predict a certain result.

The input data is sequentially passed through different “layers” in which a series of learning rules modulated by a weight function are applied. After passing through the last layer, the results are compared with the “correct” result, and the parameters (given by the “weight” functions) are adjusted.

Although the algorithms and in general the learning process are complex, once the network has learned, it can freeze its weights and work in memory or execution mode. Google uses this type of algorithm, for example, for image searches.

There is no single definition of what Deep Learning is. In general, when we talk about Deep Learning, we are talking about a kind of Machine Learning algorithms based on neural networks that, as we have seen, are characterized by cascading data processing. The input signal propagates through the different layers, and in each one of them, it undergoes a non-linear transformation that extracts and transforms the variables according to certain parameters (weights or thresholds).

There is no established limit to the number of layers a neural network must have in order to be considered Deep Learning. However, it is considered that deep learning emerged in the 80’s, from a neuronal model of between 5 or 6 layers, the neocognitron, created by the Japanese researcher Kunihiki Fukushima. Neural networks are very effective in identifying patterns.

A very striking example of Deep Learning application is Google’s joint project with the Universities of Stanford and Massachusetts to improve the natural language processing techniques of an AI type called the Recurrent Neural Network Language Model (RNNLM).

It is used for automatic translation and creation of subtitles, among other things. Basically, it builds sentences word by word, based on the previous word.

What is Natural Language Processing?

The most important resource that the human race possesses is knowledge or information. In today’s information age, the efficient management of this knowledge depends on the use of all another natural, industrial and human resources.

Throughout human history knowledge is mostly communicated, stored and managed in the form of natural language – Greek, Latin, English, etc. The present era is no exception: knowledge continues to exist and is created in the form of documents, books, articles, although these are stored in electronic or digital form. The great advance is that in this form, computers can already be an enormous help in the processing of this knowledge.

However, what is knowledge for us – human beings – is not knowledge for computers. They are files, a sequence of characters, and nothing else. A computer can copy such a file, back it up, transmit it, and erase it – like a bureaucrat who passes papers to another bureaucrat without reading them. But you can’t look for answers to questions in this text, make logical inferences about its content, generalize and summarize it – that is, do everything that people normally do with the text. Because you can’t understand it.

To combat this situation, much effort is devoted, especially in the most developed countries of the world, to the development of science that is responsible for enabling computers to understand the text. This science, depending on the practical versus theoretical approach, the degree to which comprehension is expected and other aspects has several names: natural language processing, word processing, language technologies, computational linguistics. In any case, it is about processing the text by its meaning and not as a binary file.

The general scheme of most systems and methods involving language processing is as follows:

* First, the text is not processed directly but transformed into a formal representation that preserves its characteristics relevant to the specific task or method (e.g., a set of letter strings, a database table, a set of logical predicates, etc.).* Then, the main program manipulates this representation, transforming it according to the task, searching in it for the necessary substructures, etc.* Finally, if necessary, changes made to the formal representation (or the response generated in this way) are transformed into natural language.