Raise your hand who haven't ever thought "What if..." for the idea of having real intelligent machines among us. Or, still, who haven't ever seen a movie staring human-like robots?
Humans have been fascinated for so long about the idea of having machines performing tasks only we could make, that the first time an Artificial Intelligence (AI) was put into the screen was in 1927. However, it was only during the World War II that we had real studies beginning on this particular field - and that is almost a hundred years from now!
It was basically because of this urging will to prove that we could create a computer able to think, communicate, and act like a human being, that we now have the ultimate advances we see on our daily life. Smart assistants (like Siri and Alexa), self-driving cars, translators, voice and facial recognizer, disease mapping, plan autopilot, credit card fraud prevention, personalized online marketing, and so many others are all everyday examples of AI been used. But these, what we have, these are way far from be at the finish line.
As technology earns its own place on the market, researchers raised many other terminologies to support the AI continuous evolution. Machine Learning (ML), Deep Learning (DL), and Data Science (DS) are some of the terms we noticed were added on the AI's toolbox. As they have been mentioned a lot in recent years, it's time for us to understand how they differ from each other, but also how they complete each other.
Artificial Intelligence
Everything that make it possible for a machine to present signs of what we call human intelligence can be told as AI. That is, the concepts of AI were created to name all the effort we have been making to make sure that this obsession we have under the question "Can machines think?" becomes less and less an unreality.
AI is nothing but our desperate need to prove ourselves that we can create a computer that is not only able to look like a human, but also to act, speak, learn, and think like we do. So it's not only having a cold computer program calculating formulas and giving you the results anymore, it is also having this computer learning from experiences and exhibiting signs of intelligence by itself.
Today, this is still a very abstract concept, since the only type of AI we have is a goal oriented. It means that it is created to be smart at a very specific task, but it doesn't have a proper conscious of what it's been doing.
A computer would deserve to be called intelligent if it could deceive a human into believing that it was human. The Turing Test. Alan Turing, 1950
However, studies have been also trying to achieve a more general type of AI, where machines will be able to have its own set of understanding, interpreting, and acting; becoming unpredictable, and indistinguishable from a human being in some given situation.
Although this is still not part of our reality, many advances in technology have been substantial for people to start believing that we could, one day, achieve this sort of intelligence. But for now, having computers performing specific tasks and showing, sometimes, better results than a human being... This could be already called madness a few years ago.
Machine Learning
image by author
It was when studies started to get really deeper into discovering how to make machines learn like humans, that the first concept of ML was introduced. In 1959, this field appears as an evolution of the computating learning theory and the pattern recognition studies.
Machine learning is the field of study that gives computers the ability to learn without being explicitly programmed. Arthur Samuel, 1959
Scientists no longer wanted to only work on a trial and error system. They wanted to improve their ability to reach out a result that could be acceptably accurate. For that to happen, they saw the need to explore into the real meaning of learning; and so, later on, they come up with this concept that a human learning process could be put into some logic means through an algorithm, and explained to a machine. This machine would be then able to reproduce this "logic", they called the "learning process", and improve their results by retrieving errors as feedback. This so-fancy concept is nothing but what we already well-know as learning through experience.
Many types of learning were developed on later years. But for now, we just need to understand that ML is a study that introduced the idea that we can make machines learn from experience - from data - without actually explicit programming their learning process and behavior.
Only by explaining the machines how to recognize patterns, researchers created a way to make computers develop a sufficient intelligence to take back their own errors and use them to improve their responses. This is Machine Learning. It is a subfield of AI, since it is the tool used to achieve many of the artificial intelligence applications we can see today.
Deep Learning
image by author
To be able to make the computer learning theory works with more complex kind of data, and to write the pipeline for the learning process as an algorithm that could interpret things like images or sounds, scientists did an awesome job recreating the human brain engine by simulating our neurons.
I bet that sounds complicated, huh? Let me make it simpler.
In human beings, neurons are the basic structure responsible for carrying messages to and from our brain. Without them, we wouldn't be able to see, to hear, to feel, to taste, or to walk. It means that our brain needs to be entirely connected to our body in order for us to be able to do whatever thing we might want to do. And this "connection" is only possible because we have neurons - and many other structures - to carry electric impulses throughout our body. But machines don't have this connected body we do. So that's where the scientists came up with this insight.
"Maybe, we need to simulate a neuron!"
From that moment, when we found out that our brain worked just like our personal engine, researchers begin to create what they called the artificial neuron.
Not all ML algorithms are powerful enough to work with any kind of information. Let's think of this in terms of our body. We may say that it's easier for us to understand that if we feel hungry we need to eat, than learning how to read when we still don't know what letters are for.
Similarly, some ML algorithms only work when the information we provide them are very organized and structured, just like our body is organized to know that eating is the natural action for when we feel hungry. This little issue brought up scientists the discussion about how to make machines understand data that are not so organized and structured, like images and audio files.
The artificial neurons were definitely a huge leap to the ML algorithms evolution. They are the concept of Deep Learning as it is.
DL is a class of ML algorithms that use concepts way more complex, including multiple layers of data processing, to extract features from unorganized and non-structured data, that is, raw data. In other words, DL can be seen as an evolutionary complement of the ML algorithms, that has been used to bring us even closer to the initial idea of having general artificial intelligence created. Now, with the advances of DL algorithms, we can not only learn from basic set of data, but also break very complex data into small levels, and work with these smaller information into layers that will, at the end, bring all the processing together to come up with final features.
But you might be asking yourself "Where does this Data Science thing go?"
Data Science
Now that we have all these technologies being created, it's time for us to think about its usability for good. And if you just though of DS as one of the ways to take advantage of these tools, you get it right, pal!
Data Science is an interdisciplinary field that involves a wide range of subjects, such as mathematics, statistics, computing processes, algorithms, business understanding, data analysis, and so on. The main goal of a data scientist is to extract valuable knowledge from information - called data -, in order to provide actionable insights in a broad range of application domain.
To be even clearer, DS is a field that can make use of the ML and DL concepts and tools to come up with insights that will turn into solutions for you.
But who might be you?
You could be anyone who need to solve a problem and who have a set of information, but still don't have a clue on how to use it. A data scientist will take her or his ability to work on big sets of data to manipulate, organize, and interpret them, and finally provide you with a clean image of what you need to turn your decision making process easier.
Summary
image by author
And to put all these information into one place, here is a final picture of the relationship among all these terminologies.
Briefly, AI can be said as everything and every effort that make it possible for a machine to present signs of what we call human intelligence. Also, ML is a tool that we presented for the AI that allowed machines to learn by being exposed to external information. This way, we can understand that ML is part of the AI world.
Digging a little bit deeper, we have DL that came as a complement to the ML algorithms to solve some restrictions when working with more complex type of data. That said, we can understand that DL is part of ML, being considered as a subclass within the whole package of ML algorithms.
Finally, DS is well-known as a field of study that is able to take advantage of all of these tools, and combine them along with some mathematical, statistical, and analytical knowledge to extract valuable information for the most varied domains.
Here you can have some more valuable information:
Toward Data Science (EN)
A História da Inteligência Artificial (PT-BR)