God does not play dice. Artificial Intelligence does it.

“God does not play dice with the universe” is one of Albert Einstein’s most famous quotes.

Einstein referred to quantum mechanics, a branch of physics that studies the behavior of very small particles (at the atomic and subatomic level), which he never accepted.

The most surprising thing about quantum mechanics is that it proposes that at the atomic level everything is governed by probabilities, it is not possible to calculate any state with certainty.

And this is a totally opposite approach to that of classical mechanics, which is deterministic, that is, it maintains that if all the data is available and the laws that govern it are known, any result is measurable and predictable. For every result, there is previously an explainable cause (cause-effect). An example is the formula for Newton’s second law that we all know: Force = Mass x Acceleration.

“Traditional” computing is based on this determinism. If you know the rules (laws, or formulas), you can write them in computer code. What is called an algorithm. And if you have all the data the algorithm needs. What is known as the variables. You can calculate a result accurately. No room for error.

It is what they used in the Tokyo 2020 Olympics to program a robot that threw basketball shots.

He made it 2.020 times in a row. Not a single fail.

After all, the parabolic shot formula is well known (Newton’s Laws). It is not easy to do the calculations because there are many variables involved. In addition to the known and fixed ones such as the height and distance of the basket, gravity (which depends on altitude), air density, atmospheric pressure, air friction, weight, and shape of the ball also influence and even the rotation of the Earth, among others.

Knowing the formula, and having all the data, we can calculate an accurate result.

And what happens if we don’t have all variables, or we don’t know the formula?

This is when Artificial Intelligence (AI) comes into play, and as with quantum mechanics, the most we can hope for is to calculate the probability of an event occurring.

This is what happens when we want to predict the behavior of our customers. There is no rule to guess it exactly. But what we do have is lots and lots of data about the behavior of other customers in the past. We have the Data, but we don’t know the Formula.

Using Artificial Intelligence, we can try to find out the rule. Although in the end, the most we will get is an approximation, a probability.

And how is it done?

In Unsupervised AI models, one of the techniques used is Clustering. We group the data into sets. Randomly. And we study these groups to identify any “insight” that makes sense. How are the data from the same group similar? How are they different from the other group?

And if we don’t find a relationship or meaning (a pattern). Then we’ll roll the dice again.

That is, we will again randomly group the data into sets, and analyze them again. And so, we will repeat random grouping (rolling the dice) until we find some sort of rule, relationship, or pattern between the groups.

This Unsupervised AI technique is used by companies, for example, to create better profiles of customers and their purchasing habits, and thus be able to send them more personalized commercial messages that improve the probability of purchase.

This is the case of product recommendations. With the thousands and thousands of data they have about customers and their previous purchases, Artificial Intelligence groups them until it finds a “possible” relationship between 2 different groups.

Simplifying a lot, a case would be that of a “group A” made up of customers who buy mid-high range sports equipment and consume nutritional supplements. And, on the other hand, we would have “group B” made up of customers who buy organic food and consume wellness and health treatments.

It is possible to semantically analyze the two types of consumption and reach the conclusion that both groups, although they buy different products, have common interests related to health, so that the products and services that a group buys have a greater probability (not certainty) of that are also of interest to the other group.

It is the usual phrase that appears from “Other users like you have also bought …”, or “You may also like it”.

Needless to say, these analyzes are much more complex, with more groups and more variables.

Unsupervised AI is also used for other uses, such as fraud detection. The logic is similar, segment customers into different groups and identify some kind of rule between them. In this case, it would be to detect a difference in behavior between the group that makes correct use of the credit card, and another group in which the use of the card is out of the ordinary, which is why the suspicion of fraud arises. in progress.

In “traditional” computing we know the rules and we have all the data, and we get accurate results.

We use Artificial Intelligence when we don’t have all the data, or we don’t know the rules (or only part of them), and the result we will obtain is always a Prediction, that is, the probability of something happening. But never security.

And this lack of certainty was what made Albert Einstein so uncomfortable.