Petit traité d’Intelligence Artificielle industrielle pour les nuls
Welcome to this series of blog posts about Artificial Intelligence applied to the industry. The reasons why I decided to write about this subject are diverse. The two most important reasons are:
1) In my industrial practice, I am often told that Artificial Intelligence is an overused topic whose final understanding is limited, although we know that the subject is now inevitable. My goal is therefore to show you that AI is science rather than shamanism (Disclosure: this is a simple example… the author has absolutely no bias about shamanism or other types of magic 😉 ;
2) There are currently 3 types of documents related to Artificial Intelligence :
a) Documents filled with incomprehensible mathematical formulas unless you have 3 postdoctorates behind the tie (Disclosure: this is a simple example… the author has absolutely no bias about PhDs and other high-level degrees😉;
b) Documents filled with thousands, if not millions of lines of code incomprehensible if you are not a programming expert (Disclosure: well… you know the principle now😉 ;
c) Documents that promote the great future opportunities for data scientists;
The average person, you may ask me? Simply forgotten my friends….
These are the reasons why I am starting to write this series of blog posts today. This is my humble participation in the demystification of Artificial Intelligence, more particularly in an industrial context.
A brief message to Artificial Intelligence specialists: Please do not be offended by what may seem outrageously simplified in these posts. It may simply be a hint that this humble writing is not for you😉
Artificial Intelligence – Supersimplified definition
All right, all right, all right… here are some explanations!
1) Artificial Intelligence is an extremely vast field. It includes several disciplines such as Machine Learning, rule engines, etc.
2) As a general rule of thumb, when we talk about using predictive models to help us make decisions, we are talking about Machine Learning (we will see why this name a little later). This is probably the most widely referred to subclass of Artificial Intelligence. Machine Learning will be discussed in most of this series of blog posts.
3) Neural networks are a subclass of Machine Learning. This is a discipline that is currently the subject of intensive research and interesting developments. This is a technology that attempts to mimic the behaviour of human neurons in order to significantly increase the predictive power of Machine Learning models.
We will see in more detail (simplified) how Machine Learning works in the next article. We will see how the creation of decision-making models has changed drastically with the advent of Artificial Intelligence.
However, before I leave you, I would like to answer two questions that are commonly asked to me: What is a Machine Learning model? How does it work?
Well, here it is… a Machine Learning model can be presented as a simple file.
And if we look at its content now…
Code, just code! The model in this example is generated by a well-known online Machine Learning platform named BigML.
The model file to which this piece of code belongs is used to detect in advance which customers of a telephone company are likely to cancel their subscription within X months.
How does this model work now?
In a simplified way, we first provide him with the data describing a client, the model then performs a series of processing operations, then finally it sends us back as an answer: the client will leave within X months, or the client will not leave within X months. Magical, isn’t it?!?!!!
Well, I think that’s enough for today. I don’t want to overheat your Natural Intelligence
Next week, we will see a crucial aspect. It’s nice to have a model that makes predictions, but you still have to create that model! That is what we will see.