Black Box & Explainable AI: The results of machine learning

Written by Alana Team
on November 17, 2020

Machines have a specific way of learning which is completely different from the way people learn, but that was  only developed thanks to studies about the brain and also about the human learning process.

Nowadays, our greatest challenge is to understand how machine learning can keep evolving, and contributing to artificial intelligence. Once there are different ways to teach a machine, in some cases, the results might surprise even the scientists.

Find out about  the connection between artificial intelligence and machine learning, the types of machine learning, and the concepts of Black Box and Explainable AI.

AI and machine learning: understanding the difference

Artificial intelligence is a discipline that embraces theory and practice in the creation of intelligent machines, encompassing several areas, such as philosophy, biology, mathematics, and even linguistics.

In a nutshell, artificial intelligence is described as the ability of a machine to simulate the human in some way.

Machine learning, on the other hand, is a subfield of AI and has its own set of mathematical techniques and statistics so that the program follows instructions and makes discoveries on its own, without being explicitly programmed.

WhatsApp Image 2020-11-17 at 10.07.57

The great advantage of the machine learning over the other techniques is that, in this model, there is no rigid rules created, only examples and data are provided so that the machine can identify them and learn.

Deep Learning: machine learning methods

One of the most promising machine learning methods is Deep Learning or deep machine learning.

This technique allows the machine to learn through several layers of artificial neural networks, those that mimic the biological structure of the human brain. Each layer of the neural network is composed of algorithms which are specialized for different tasks, such as facial recognition or natural language processing, for example.

What does this mean in practice? The more layers of the network, the greater the capacity for abstraction and sophistication of an artificial intelligence system.


Types of Machine Learning

Machine learning can happen in different ways, with a greater or lesser degree of interference and human supervision. There are three main methods of machine learning, each of them contains its own peculiarities.

  1. Supervised learning: happens when a specialist offer category models for the machine, in a similar process  to when a teacher teaches a human.
    Imagine that a developer inserts a photo of a dog into the machine database, and then classifies that object as a dog.
    The machine then understands what a dog is and what data of that type cannot be named otherwise. This supervision can be done by both humans and other algorithms.
  2. Unsupervised learning: allows the machine to find out  on its own what the categories of each piece of information are and can classify them based on similarities. 
  3. Reinforcement learning: happens through mistakes and successes. In this case, the machine performs random tests and, based on the experience acquired, tries to maximize the metrics presented and discover the best combinations. 
    Reinforcement learning is widely used in game development, in the creation of robots and virtual environments, as it generates excellent process optimization, although with a high computational cost.

Black Box & Explainable AI: the two extremes of the machine

The deep learning technique has peculiarities that can impact the development of artificial intelligence, and the use of many layers of decisions can make it impossible for human supervisors to interpret how the machine achieved a certain result.

This phenomenon is called BlackBox and can be harmful to some more sensitive areas, involving the need to understand data and the behavior of the machine, such as the insurance market and the health area.

One way to avoid BlackBox is to use machine learning algorithms that are simpler, but natively explainable, such as decision trees, and other models that can be traceable and transparent about the decision process.

On the other hand, there is an evolving field called Explainable AI (XAI), which aims  to bring more transparency regarding the  decisions made by machines, even if they were performed without supervision.

To learn more about artificial intelligence, listen to the Inside Alana Podcast.



personagens escrevendo




Você também pode gostar:


[Video] Leaders of the future: chat with Luiza Helena Trajano

Creating a digital business and expanding it is not a simple task, it is necessary to invest time and money in technolog...


Black Friday: the challenge of sales volume increase and customer service

Black Friday is the most anticipated large-scale promotion of the year. It is the chance that the consumers have to take...


Artificial Intelligence and the Future of Communication Between Humans and Machines

Interpreting the human language is essential for communication between humans and machines. It happens through the natur...

Receba insights sobre
Inteligência Artificial por e-mail,
na medida certa.