#2 - Understanding AI: Human, All Too Human

Continuing from the previous article...

The question I get asked the most after all these years working with AI is: How can I understand artificial intelligence without an engineering background? How can a business person grasp this complex and numbers-filled world? To answer this, let's delve into the fundamentals of AI itself: mathematics.

As humans began expressing thoughts through symbols, mathematics became our primary tool for abstract thinking. Interestingly, those who are most drawn to abstract concepts often feel the most disconnected from them nowadays. In this series of articles, I propose a reconciliation for those who, due to a fragmented education system influenced by the specialized focus and social division of labor during the industrial revolution, found themselves marginalized from this way of thinking, lacking mathematical support for their abstractions.

Consequently, we've witnessed an unprecedented separation of mathematics and philosophy into distinct paths, whereas they used to be intertwined in our analysis of the world.

Artificial intelligence compels us to revisit these foundations because, ultimately, AI is a manifestation of human abstraction. I dare say, even though it may challenge traditional views, that few things in technology are as deeply human as artificial intelligence, which is why it captivates us—a reflection of ourselves.

In AI, you don't need to know how to solve equations unless you want to become a machine learning engineer or a researcher in the field. However, to truly understand how AI works, it's essential to grasp certain mathematical concepts. Understanding, in this context, means being able to work with the abstract ideas surrounding these concepts and apply them to problem-solving.

The first concept we must grasp is that artificial intelligence can be likened to a grand equation with millions or even billions of x's and y's in search of a value. When we consistently determine the probabilities of y (output) values for each x (input), we say that the AI has learned. Training an AI, therefore, involves exploring millions or billions of combinations to find the y for a given x using one or more core equations—a mathematical conductor that guides this numerical symphony. We call this conductor equation algorithm.

To gain a better understanding of this process, we need to revisit something we learned long ago in school: functions.

What is a Function?

A function can be seen as a rule that takes input values and produces corresponding output values. The input values are referred to as the domain, typically associated with the variable 'x,' while the output values are called the range, associated with the variable 'y.' In simple terms, a function allows us to manipulate input values to obtain specific output values.

Your brain just recalls the school days saying, "No!!! "

But let's understand it in a different way: A function can also be understood as what happens to x if I change y? Or in a recipe, how much water should I add if I increase the flour? Or what should be the quantities of the other ingredients in the cake if I add a little more milk? Or what force should be applied to throw a baseball after running to another point in the stadium? All of these are functions. Inputting data to obtain an output, y and x working together every single day.

Another practical example but now in Machine learning field can be image recognition tasks, where functions are employed to classify images into different categories or recognize specific objects within an image. For instance, in a problem of facial recognition, a function is designed to analyze facial features such as eyes, nose, and mouth.

The function takes pixel values of an input image and applies to the "conductor equation" to find which Y's determine a specific face. By learning the underlying patterns, the function can accurately identify individuals or recognize specific objects within images.

Functions are represented using notation such as 'f(x).' The 'f' denotes the name of the function, and the '(x)' indicates the input variable. For example, the function 'f(x) = 2x' can also be written y=2x.' Similarly, we can have various functions, like 'g(x)' or 'h(t),' depending on the context.

Using the last example we can have a function for each pixel and represent them as P1(x) for the first pixel , P2(x) for the second and so on now think about what happens with a 4K imagine with 8.3 million individual pixels.

To evaluate a function at a specific value, we replace the input variable with that value and apply the given rule. For instance, if we want to find 'f(2)' for the function 'f(x) = 2x,' we substitute 'x' with '2' and calculate the result. Following the order of operations, we simplify the expression to find the corresponding output value. In this case, 'f(2) = 4.'

Hey, do you understand that abstract concept you saw at school better now? Every day, you use functions. There's an f(x) on your TV that decides every point on the screen. There are thousands of f(x) to rotate the screen on your phone, and an f(x) to determine the zoom level based on finger movement.

Our lives are filled with hidden mathematical functions in technology, and artificial intelligence is no different. But how big can all of this be? And how does AI manage to understand things on its own? How does it know how to reach the right value by itself? Well, those are topics for future articles. I hope you enjoyed it, and if you didn't understand any point, please leave a comment. Take care.