In the first episode of Introduction to Artificial Intelligence and the differences between Strong AI and Weak AI, we started talking about basic concepts and the evolution of technology over the years.
Continuing this season of Inside Alana Podcast, the second episode seeks to leave science fiction and go to reality, unraveling the differences between the way machines "think" and acquire knowledge VS the human mind.
It is daring to say that machines think like humans because for them to act, someone must feed them data.
On the other hand, as we talked about in episode one, the development of artificial intelligence focuses on studying two aspects of the human brain: anatomy and how learning works.
Thus, we can say that the machines have an artificial brain that is developed from the attempt to replicate the functioning and organization of the neural networks of the human brain.
To understand how this is done, it is first necessary to learn the following concepts:
- The difference between artificial intelligence and machine learning
- Types of machine learning
- What are artificial neural networks?
AI x Machine Learning
According to Marcellus, CTO of Alana AI, artificial intelligence is "the science and engineering of creating intelligent machines", and there are several ways to simulate this, such as the machine learning method.
The method is the application of statistical and mathematical techniques so that a program follows instructions and makes discoveries on its own, that is, the human creates ways for the machine to learn on its own.
Machine learning encompasses the use of algorithms that can lead the program to discover for itself what action to take, which is a recipe to be followed by the program to feed artificial intelligence.
Types of Machine Learning
In general, there are three ways to teach a machine, and each one with its level of complexity:
- supervised learning
- unsupervised learning
- learning by reinforcement
1. Supervised Learning
In this model, the human provides a series of examples of what the machine should learn and what it should not.
As Marcellus exemplified, to create a human face classifier, we can provide thousands of photos of people to the algorithm so that it learns to differentiate human features from other things.
It is called supervised precisely because someone needs to follow the learning process, and this is done with the examples provided for the machine.
2. Unsupervised Learning
This method is when the human lets the machine discover on its own what are the types of examples. They are algorithms used to create groups from similarities and differences.
In this model, examples are not provided, only the raw data and it is up to the machine to find patterns that differentiate the information and, finally, to create classification groups (cluster).
3. Learning by Reinforcement
It works based on errors and successes because the algorithm performs tests to understand what is successful and what is not.
It is the learning based on experience, where the machine understands its errors and looks for a correct approach in the next attempt. What happens is that the algorithm has a goal and a metric, and performs tests to know how close it is to reach the goal.
Artificial Neural Networks
Artificial neural networks are commonly presented as interconnected neuron systems, which can compute input values, that is, they simulate the behavior of biological neural networks.
Marcellus used handwriting recognition as an example, which would work as follows:
- The neural network is defined by a set of input neurons that can be activated by the pixels of an input image;
- The data acquired by this activation of neurons are passed on, weighted, and transformed by a function determined by the network designer;
- This process is repeated until, finally, an output neuron is activated and determines which character was read.
These Artificial Neural Networks (ANN) are the basis of deep learning, which is the most used machine learning technique with the best results.
Deep learning is based on Artificial Neural Networks, which imitate the biological structure of the brain.
The depth of learning happens because multiple layers of neural networks are used to perform the data classification.
For example, each layer has a specific feature to learn (such as curves and edges) in image recognition, so each will focus on its specific topic and get deeper into it. To better understand, it is worth listening to our second episode in full.
We can conclude that machines imitate in a very basic and superficial way, as a human thinks, but definitely, they do not think like a human.
Listen to the Inside Alana Podcast on the platform of your choice and stay on top of everything about artificial intelligence.