Artificial Intelligence has become what everyone talks about and not everyone understands.

When we talk about Artificial Intelligence, people already believe that it is something that works for everything, that in a short time there will be humanoid robots everywhere and that we will have to be careful, because in the future it may make the world of interaction inflexible.

Where is AI at today and what is it for?

I am sorry to disappoint those entrepreneurs who think that soon they will be able to hire an Artificial Intelligence to do all the work for them, the reality is far from this.  What has increased a lot is the computing power in relation to their costs.  Added to this is the ability to mount systems on servers in the Cloud with enormous capacities.

Today there are several fields where AI far exceeds conventional computing capabilities. 

There are experimental developments and machines in large companies and universities that are providing promising results.  Today many companies and institutions use AI in any of the fields where it is appropriate or combine these utilities to provide a solution.

What really is artificial intelligence?

"Artificial intelligence (AI) is the part of computing that deals with the design of intelligent computer systems, that is, systems that exhibit characteristics that we associate with intelligence in human behavior: understanding language, learning, reasoning, resolution of problems, etc. " (Barr and Feigenbaum, 1981)

In simple terms: Artificial Intelligence (AI) refers to systems or machines that mimic human intelligence to perform tasks and that have the ability to iteratively improve from the information they collect.

AI contains many subfields, some very broad. Some of the different areas of tremendous AI growth over the last few years:

Like most scientific fields, there are both theoretical and applied areas.  The theory seeks to examine and identify ideas about the representation of knowledge, learning, systems of rules, etc. It has almost become a branch of mathematics and computer science.

On the other hand, the approach of applied engineering is to solve real-world problems using artificial intelligence techniques.  Most are more interested in the practical application side of AI than the theoretical.  Nowadays a programmer doesn't need a Ph.D. to implement artificial intelligence solutions, but like any engineering problem, you need to understand enough to know what tools to use to solve what kind of problem, and then how to use those tools.

Daily use and potential of Artificial Intelligence

What is a neuronal network?

Neural networks are a fundamental pillar of Artificial Intelligence.

Neural networks is not something new. They date back to the 1940s and 1950s, when the first concepts began to be published. However, they were never very successful, there were not enough computing resources available to get good results.

The initial idea arises from the intention of imitating the brain. The human brain contains about 100 billion neurons, and its "computing power" resides in the connections between those neurons.

Neural networks are computer programs that mimic neurons and their connections. The basic element is the neuron, you work with a more or less complex model and mathematics comes into play.

With artificial neurons, computer programs are created that link them together and form neural networks.

Where is the true "magic" of neural networks?

Well, nothing more and nothing less than they learn to do things.

The more powerful and better assembled neural networks are, the more complex things they can learn.  Basically neural networks learn by training. Give them examples of things and they will learn. Let's say, for example, that we want them to learn to add.  To help you learn to add we give you examples and tell you which results are correct.

2 + 2 = 4 | 1 + 5 = 6 | 4 + 4 = 8

Once trained we can give you two numbers that were never entered in the training and the neural network has learned to add.

Artificial Intelligence, where it is really powerful.

Natural Language Processing (NLP)

Known by its acronym in English (NLP: Natural Language Processing). This line of Artificial Intelligence works to try to make sense of human language and the words that make up language.

Language treatment can be tricky. Many applications or Internet services have long used the NLP, for example when making voice dialing with a mobile phone, social networks to determine what types of content are shared, there are endless software that use it.

Some examples are:

In this example, from the company Bitext, you can see how positive, negative and neutral feelings are located.

Face identification (vision)

When an AI looks at someone, they are not looking at her like you and me.

Machine Learning

Another category within AI is Machine Learning.

We are all used to rule-based systems whether they are simple or complex. One of the problems with rule-based systems is that they don't tend to scale very well and can become very complex when the nuances of decision-making come into play.

Machine learning is a huge improvement on rule-based systems - instead of having to create new rules each time, we provide a different option for the system to consider. We give him new data and we say: "Here is something new, not seen before, but it is similar to X".

While rule-based systems take the “data-> action” approach. Machine Learning works in a different way. At its core, the system works by taking data, comparing it to a 'model' (of something), and if the model matches, then it takes an action.

Let's look at an example. Let's say we define a 'blueberry pie' as light in color, small, rounded on top, with dark chunks (the blueberries) mixed in all over. Okay, that seems easy.

Now, keeping in mind the description of the cake, look at the image below.  Why is it not easy to distinguish a cake (muffin) from a Chiguagua?

Machine Learning Categories

There are three broad categories in Machine Learning. These are Supervised Learning, Unsupervised Learning, and Reinforced Learning, and each has its place and things that it is good at.

Supervised learning

This is the easiest to understand. We give the algorithm a series of examples.  In our example we train him with several types of fruits, telling him what value each one has in numbers and then he will identify the fruit according to the predictive model that has been created.

Unsupervised learning

This type of algorithm can often be seen in identifying data sets.  This is where we give the algorithm a large amount of data and tell it which parts of the data are interesting. We could say, for example: here is a large CSV file of Malaga home purchase data for the last five years, group the data into regions with groups of properties in certain price brackets.

Reinforced learning

In the latter area, rather than providing examples to the algorithm (as in supervised learning), we provide you with a method that you can use to examine and quantify your own learning performance by giving you a reward indicator of some kind.  If we wanted the algorithm to learn to play a game, we could start the game by giving it 5 points, and if it does something right, give it another point, if it does something wrong we take it away.
The algorithm can rectify, remembering the plays that gave it points and not repeating the plays that took it away.

Uniting capabilities. This is how AI becomes more powerful.

In the same way as in traditional software, different technologies come together to provide a solution to the end user.  The same is usually done in AI. If we consider building a robot that is moving and aware of its surroundings, it is done using a combination of different AIs:

On their own, these AI technologies are interesting and useful, but when they work together, they present something more powerful than the sum of its parts.

footer_logo
© 2022, Green Data Center Guide. All Rights Reserved.