Original article was published by Nechu BM on Artificial Intelligence on Medium
The 3 things I learned from Artificial Intelligence about life
When you realize your life is managed by algorithms the only way to get the control back is to think like an algorithm.
As a child I felt like an explorer discovering the fascinating world around me, any new activity was an adventure with a rush of adrenaline. I was curious and seeking new experiences and challenges. I felt the king of the world.
As I grew up, I started to understand that the world is way more complex and sophisticated, than I thought. My decisions started to become more relevant and have a greater impact, decisions like going to University, moving to a new city or choosing the people I surround myrself with. Everyone encounters a period in their lives in which insecurity and a lack of confidence are present. This is the moment we define our characters. This encompasses a belief in ourselves, our values, abilities and knowledge. There can be two outcomes, one is staying in a state of insecurity and hesitation and the other, growing and developing our self-esteem. This is not a onetime thing; it can happen at different stages of our lives.
The rules of the game have changed and there is a third new outcome to take into account, you delegate all your decisions to your smartphone. Everything starts with a subtle move when you neglect your orientation and prefer Google Maps to move from one place to another, it continues with a recommendation from Amazon on what book to read next and finishes with the person you will date recommended by a dating app algorithm. Little by little we delegate more and more decisions to the Artificial Intelligence algorithms in our apps, and lose the ability to think for ourselves and make our own decisions, we lose our character. When I realized that for all these day to day decisions I was dependent on an algorithm, was when I realized I no longer had control over my decisions.
The interesting fact is that, before smartphones we had already been able to master fire, build pyramids, and cross oceans in sailing boats. Now we don’t even know how to cross our city without Google Maps. How did all of this happen? I do not have the answer to that question, but I can tell you how I took back the rains of my own life, without relying on an algorithm. All you need, is to learn from your enemy, just think like an algorithm. In this article we will explore 3 concepts I learned while developing Machine Learning algorithms and interpolate them into our everyday life.
1 — Objective Function
Machine learning models are defined to solve specific tasks like classification, prediction, clusterization … When we first create a Machine Learning model it isn’t very good at the task it is supposed to solve, we need to train the model so it learns. For example, if you want an AI capable of predicting the stock market movements, you will feed the model with a ton of past data on the stock market so it learns hidden patterns that make share fluctuate. How does the model know it is improving in the decision making or getting worse? The objective function. The role of this function, as it names implies is to calculate if the prediction done by the model is achieving the objective it was designed for. It measures the difference between the prediction and the expected result, to learn based on this outcome. There are multiple types of objective functions, deciding which one to choose is crucial since it will define the direction to go.
Many people lack an objective function in their lives, they have good capabilities and skills (the model) but they miss out on measuring the progress. Even though they are approaching their goals, they are not using the proper objective function, so they consider all their past actions as errors. Just because you don’t have an expensive car, nice holidays on a remote beach or a position in a firm with a cool name, does not mean you are not being successful, it just means you don’t measure success in that way. Success and accomplishment is different for every person, defining your objective function and sticking to it is key, don’t let other people’s results affect your own.
Still don’t understand very well what I mean by objective function? Have you ever had a moment in your life where you felt that all the things that motivated you in the past and fired that spark within you are no longer interesting or motivating? This turning point, when you realize that what you thought was worthy and valuable no longer has meaning and the things that were bland and did not interest you have become your major priorities and inspiration. This is the moment when you change your objective function, you change the way you measure success.
If you are not able to identify your objective function you might feel lost and disoriented, encountering frustration and stress. Humans, as machine learning models, need to know where they want to go so they can realize if with their actions they are approaching their objectives.
Alice: Which way should I go?
Cat: That depends on where you are going.
Alice: I don’t know.
Cat: Then it doesn’t matter which way you go.”
Lewis Carroll, Alice in Wonderland
2 — Trial and Error
Machine learning models designs are widely shared, you can find many references to type of structures, layers, parameters … What is more difficult to find is these same structures already trained. Even though companies share their knowledge of the best models, they hardly ever share the trained models. What is the difference between the structure of a model and a trained model? The learning phase. This step is crucial and needs a vast amount of resources (primarily data and computing power). There is no magic behind an algorithm, there is not one talented design algorithm able to make the best decision since it was first designed. If the model has not been trained, it is almost useless. Then, what happens in this training phase? How does the model learn so it can predict so well? Based on trial and error. I will say it again, trial and error. The first prediction the model does will always be bad, it will be far from the expected result. Thanks to the objective function, the model will then measure the error and make a new prediction. Every time the model fails, it comes closer to the expected result, this happens over and over again untill it reaches the expected result.
Can you imagine a machine learning model saying after its first prediction ‘I am very bad at this, I want to quit’ or ‘I have been trying over and over and still do not get the results I want’. It does not, failing is the opportunity to learn, predicting and measuring the error is the moment to adjust your decision and approach the desire results. The model needs to fail in order to learn. Once you realize that you can’t be good at something the first day, you need to try and fail so you can learn, from this point on you will be able to progress.
In the early nineties, if you were a successful businessman in New York you would have been a member of a private social club. Those private clubs where a place where the most talented and rich people would congregate and network. Joining them was very difficult and exclusive, but it was also a huge opportunity since it increased your social status and gave you the option to join new business opportunities, exclusive to club members. Most of the clubs allowed you to join based on your background, the university you attended, your family name or your bank account. However, one of the most exclusive and unique clubs would not take into consideration any of the previous characteristics, their rule was way simpler than that. ‘You need to succeed and go bankrupt in at least 2 businesses, once you success on your third one you can join’. They valued failing because they knew it was the source of learning. It also showed resilience and the will of the individual to start over again, no matter what had happened before. It showed that success was built upon hard-work and difficult decisions instead of a lucky move. Their principle was very simple ‘Failing is the way’.
3 — Best Model
As we have seen, we start by designing the model, choosing an objective function and then training the model. If you only know how good your model is after training, how can you know which shape the model should have? How do you know which objective function you should choose? You don’t know, the more you work on a certain problem or domain your intuition of what could work better improves, but you never know for certain until you train and measure the prediction. In consulting, when a client requests a model to classify documents, we start testing 5 models with out of the box parameters, this allows us to have a sense of the model that might work best. Then, we continue with the best 3 models or the models that might have the highest road for improvement. we focus on optimizing parameters and having a better understanding of the data, finally we test the 3 models again. From this we can measure the performance of the models and based on the relevant KPIs pick the best one. Once we have the results we will see it is ‘obvious’ which model is the best performing one.
There is not one best model for a certain task, just as there is not one best path for life. One of the sentences I like the most is ‘you cannot connect the dots looking forward you can only connect them looking backwards’. We think that our destiny is already designed, that there is just one way to achieve our dreams and goals, that there is a path somewhere and that once we find it everything will be solved. If we don’t find it, we feel lost and disoriented. Don’t feel stress or anxious if you don’t find the best path, I have good news for you, it does not exist. Just like we work on different models and we keep multiple options open, try to live your life this way, keep your mind open and look for new challenges, opportunities can come in many forms and shapes.. It might seem like you wonder and lack direction, but one day you will look back and you will be able to connect all the dots. Every decision and step you have taken will appear to have been magically ordered and structure in one path. If you pick the right objective function and properly apply the training phase, you will be able to develop your ‘best performing model’.
Even though machine learning models rely heavily on mathematics and don’t care much about success, we still can extract key takeaways to apply in our lives to achieve our objectives. I would like to hear from you, is there any other lesson you have extracted from Machine Learning models we could apply to our lives? Leave your answer in the comments!