How easy it is to learn AI?
Believe it or not, explaining AI is relatively simple, even a 10-year kid could easily understand it. In fact, there is no need to unravel complex math formulas to understand how an AI processes information.
In this post, I am going to explain to you in the least possible number of words how an AI is made. At the end of the post, you will be able to conceptualize your AIs as well.
We start from data
Every time I hear an explanation on how AI works, everyone begins with the dummiest possible example: dogs and cats. The data picked for the example could not be worse. In fact, cats and dogs are physical entities, but how do you convert them into data. I am assuming people are talking about images, however, the conversion of an image into pixels is too complex for a beginner to imagine. You would require a 3-dimensional matrix that holds RGB information.
To correct this problem, I am going to start in a much simpler fashion: Imagine you have an excel file at your disposal (technically this would be considered cross-sectional data). Each column corresponds to a variable, while each row to a different object. in this case, a wine dataset.
We split the data
There are several kinds of AIs. Because I wish to minimize your learning effort, imagine that AI is built to do one thing: PREDICT!
To make a prediction, you will need to pick the data that you wish to predict in the future. For this purpose, we are going to split the data into features (data that act as predictors) and labels (data you wish to predict).
The entire splitting process is explained with this image: orange columns are features, violet columns are labels:
We train the model
Now we are ready to create a model. This action is called training a model and consists of finding the link between features and labels (what I call rules). The AI undergoes this process (as you can see from the outlined architecture):
Once the AI has been trained, we have at our disposal what we call a model. While the training process is expensive, using a model to make predictions is cheap and can be used without limits.
We make a prediction
To use the model we just need a new dataset with some features. The model, now that has learned to rules, can try to estimate the labels (wine name).
For example, a wine in stock that is still, sparking and is just 1 year old may likely be Champagne!