What I stunning learned on A.I. from my 4-years-old boy

7 min readMay 5, 2018

Understanding “what is Artificial Intelligence” is not a piece of cake.

This confusion happens because it refers to many different domains and tasks. Looking back at my little boy growing process I understand now how much this has helped me to understand how Artificial Intelligence works.

Here I want to share with you my experience.

KISS! Keep it simple, stupid!

KISS (Keep it simple, stupid!) should be a mantra. Unnecessary complexity should be avoided not only in design stuff.

This principle is valid when you have a baby, and you are teaching him his firsts words.

You have to short all long, complexes sentences. You need to reduce your vocabulary extension to those few words that your child already know. Abstract words are the worst because you have nothing to show him to understand what you are meaning.

This principle is valid even with machines.

No fancy spatial things. They are only able to do one single thing well.
Machines follow instructions. They can perform significantly but only under well-explained facts and figures.

Artificial Intelligence: a Definition by Approximation

Artificial Intelligence meaning is so confusing because it involves many sub-domains. It includes a bunch of other hype words that are, as a matter of fact, part of the process of training and learning.
Machine Learning, Deep Learning, Data Mining only to mention few of them.

Here below a scheme on some of those technologies that the A.I. ecosystem includes.

All are “AI” and, at the same time, they are not “AI.”

What does it mean?

It means that some of this words ARE NOT Artificial Intelligence because of how they work. They are algorithms that, after a lot of training, can do specific tasks.

However, at the same time, they ARE Artificial Intelligence because of, filled with even powerful algorithms, they learn.

Learning skill make the difference.

How do they learn? Doing mistake and improving by approximation.

AI algorithms are “mind tools”, not artificial minds. This implies that successful applications of AI hinge on more than big data and powerful algorithms.
Jim Guszcza, Smarter together: Why artificial intelligence needs human-centered design, Deloitte Review, issue 22

Most AIs are made of fixed rules, using the power of high performing processors. One of the most famous processors used in AI field is the NVIDIA GPU (Graphic Processor Unit).
To stick with the language of my little boy these processors “are so fast as lightning.”

GPUs are “force,” not “brain.” Machines using GPUs look like they are intelligent, but they are not. They are only able to process a massive amount of data.

Big Data is a Puzzle

Big Data represents a considerable amount of data. We can think “Data” as a puzzle of millions of pieces.

Think “Big Data” as a puzzle of millions of pieces.

You are looking for just the bit similar to that you have in your hand, that one that fits and perfectly intersects.

Do it manually, and this task may need you years and years to succeed.
Machines with powerful GPUs can do it in just a few seconds.

Ok, maybe the process is not 100% perfect. But with a good approximation, the machine will extract for you a small selection of possible right pieces. Then you will adjust.
During this process, there is no learning activity.

It is at this point that, with the right instruction (aka algorithm) the machine become able to learn.

“Machine learning is a subfield of artificial intelligence. Its goal is to enable computers to learn on their own. A machine’s learning algorithm enables it to identify patterns in observed data, build models that explain the world, and predict things without having explicit pre-programmed rules and models.” — from “Why Machine Learning Matters”

Machine Learning are machines with Artificial intelligence able to learn from data. However, these machines need training.

You adjust, and it learns. You define what is ok and what is not right for you.

Artificial Intelligence is a baby to teach

Let me explain easier with an everyday life example: the process of growing up a baby.

A newborn has not past experiences, he/she will acquire by living the life day by day.
First, he learns to recognize his mum and his dad. Thee other close relatives or family friends. He doesn’t know them yet, but, the repetitiveness of having them around, and the way in which adults refers to each other, give him clues. He is learning.

After, a few months later, he has a set of knowledge. Not only about people but also about objects.
Now he recognizes toys and different food through his senses. He is still learning.

We can say he is following a “training model.”

What does it mean? In practice, you give him various vast samples to study and understand differences. Differences are in size, colors, shape, weight, smells, touch and so on. He learns.

This learning process becomes even more evident going ahead, growing and improving the language skill.
You show, and you name it “dog.” However, there are many different types of dogs.

You show many different dogs to the child. After a while, several photos, correction, and approximation experiments later, he can recognize a dog even if you don’t specify that dog.

He has built in his mind a set of similarities. He has recognized a pattern and has received enough instructions and experience to understand by himself.

The same happens with machines.

Machines are like babies to grow and teach every day.

What does “training a machine” mean? (aka Training Model)

I want to introduce this paragraph with an interesting quote.
It comes from a great book that I want to suggest you: “The Design of Everyday Things” by Don Norman. Norman is director of The Design Lab at the University of California, San Diego.

He is an authority in user-centered design and innovation.

“Machines are pretty limited. They do not maintain the same kind of rich history of experiences that people have in common with one another, experiences that enable us to interact with others because of this shared understanding.”
― from “The Design of Everyday Things: Revised and Expanded Edition (English Edition)”

As I’ve told you how my teaching experience with my little boy was, is now time to understand machine learning training process.
You feed machines with data, and they learn.
You give to the machine a lot of dogs’ photos, and it becomes able to recognize a Siberian Husky and a Chihuahua both as dogs.

Advantages: machines fast elaborate millions of data thanks to powerful processors. The faster and powerful the processor and the quicker is the learning process as well.

Disadvantages:

Costs. Increasing the recognition capabilities require, as said, power and this have a high price.
Time (then costs again). Even if they are fast, the learning process needs time, experiments, and training. Remember? You had trained the child helping him understand when he did right and when he did wrong in recognizing the dog. You need to teach the machine in the same way, defining when it is doing right and when it is doing wrong by approximation.
Limitation. Has in the quote by Don Norman “Machines are pretty limited,” they not share knowledge and are not creative.

You can quickly teach a child different topic: cats, dogs, horses, flowers, food, and so on and so far. He can use experiences from the past to learn the pattern faster. A machine learns new models by always following the same structure. No walkthroughs or cheating method.

Learning Is A Great And Magic Never Ending Story

“We need to remove the word failure from our vocabulary, replacing it instead with learning experience.
To fail is to learn: we learn more from our failures than from our successes.”
― from “The Design of Everyday Things: Revised and Expanded Edition (English Edition)”

According to the author, Don Norman, Scientists are not worried about failures. Instead, they look for failure to eliminate errors. Scientists continuously are looking for the best way to do something, not the right way to do something.

When my 4-years-old-boy makes a spelling error, I usually repeat the word for him. Then I invite him in repeating the word using the correct spelling.

I have a French “r” so when I speak in some words, my “r” sounds so feeble that almost missing. Nobody said me the right pronunciation when I was a child and now it is impossible to change. However, my little boy pronounces it well and…he corrects me pronouncing the word properly (in reality he pulls my legs about my French “r”).

He experiments language as well as new things, doing wrong and doing well. He is learning experiences.

Failures in experiments are, as Don Norman says, “learning experiences.” You learn how NOT to do something.
The same for machines, the same for we humans.

Sources:

https://www.kdnuggets.com/2017/07/ai-deep-learning-explained-simply.html

Don Norman, “The Design of Everyday Things: Revised and Expanded Edition (English Edition)”