Machine learning is a subfield of artificial intelligence that studies the development of algorithms that can learn patterns from given input data and make predictions based on them.
The first machine-learning model, capable of evaluating each checker side’s winning chances, was programmed in 1950 by Arthur Samuel, an American programmer and one of the fathers of artificial intelligence (he coined the term “machine learning” in 1959.
In today’s world, machine-learning solutions are becoming an increasingly important part of our everyday lives. They are applied in various sectors like natural language processing, computer vision, speech recognition, email filtering, agriculture, and medicine.
In this article, you can find a complete picture of the machine learning world, from what it is to how AI works, from ML real applications to its limitations, and from its history to its future.
So, don’t waste any time and jump straight to the explanations.
Table of Contents
What machine learning is (definition)
Machine learning (ML) is a subfield of artificial intelligence that studies the development of algorithms that can learn patterns from given input data and make predictions based on them.
These “human-like intelligent” systems are called models and after a process of training, when the model “learns” patterns from the information we have gathered, it is ready to be validated (to estimate its accuracy) and make predictions.
Machine learning history
Even though the first machine-learning model, capable of evaluating each checker side’s winning chances was programmed in 1950 by Arthur Samuel (one of the fathers of artificial intelligence who also coined the term in 1959), in 1947 Alan Turing, famous computer scientist, mentioned computer intelligence in a public speech in London, saying, “What we want is a machine that can learn from experience,” and that the “possibility of letting the machine alter its own instructions provides the mechanism for this.”
Only a year later, in 1948, he introduced some basic concepts of artificial intelligence, like a model composed of artificial neurons that emulates the brain’s thinking process (actual neural network).
In 1950 Turing invented the Turing Test, a way to identify if a system is “intelligent” or not based on whether a human, during a conversation, can identify if it’s a machine or a human being.
By the early 1960s an experimental “learning machine” with punched tape memory and trained by humans through reinforced learning, called Cybertron, had been developed. It could analyze sonar signals, electrocardiograms, and speech patterns.
The interest and fascination with pattern recognition continued into the 1970s. In 1981, a paper was presented on the use of teaching strategies to make an artificial neural network learn to recognize 40 characters from a computer terminal.
What programming language is used in machine learning
The programming language favorite for data analytics, data science, and machine learning is Python. The latter is a famous high-end language developed in the 80s by the Dutch programmer Guido van Rossum.
Python is widely used because of its:
Large collection of libraries and packages
Python is a language full of libraries that allow programmers to not start from scratch their projects and improve their efficiency and productivity. There are also many modules about ML and data science.
- Working with textual data – use NLTK, SciKit, and NumPy.
- Working with images – use Sci-Kit image and OpenCV.
- Working with audio – use Librosa.
- Implementing deep learning – use TensorFlow, Keras, and PyTorch.
- Implementing basic machine learning algorithms – use Sci-Kit- learn.
- Want to do scientific computing – use Sci-Py.
- Want to visualize the data clearly – use Matplotlib, Sci-Kit, and Seaborn.
Readability and simplicity
Python is considered one of the most readable programming languages by many, probably because of its simple syntax and indentation. There are a lot of functions named in English and it has fewer signs (parenthesis, commas…) to use than other languages. This will make your codes easier for you and your colleagues to understand and work on (clean code benefits)
Flexibility
Python is optimal for programmers because it supports numerous programming styles, each one useful in different contexts. It supports the procedural, functional, object-oriented, and imperative style of programming allowing machine learning experts to work comfortably on what approach they prefer.
Artificial intelligence vs machine learning vs deep learning
Artificial Intelligence
Is defined as Artificial Intelligence a machine that can mimic human behaviors and cognitive skills for problem-solving or learning, such as speech recognition, decision-making, and translation.
There are three types of artificial intelligence, one more developed and powerful than the other in the direction in which they are displayed:
ANI (Artificial Narrow Intelligence)
An intelligent machine that can do only a specific and limited number of tasks that it is programmed to do, and it has no “consciousness” of what is outside the field in which it operates.
AGI (Artificial General Intelligence)
An intelligent machine that can understand and perform a wide range of instructions, but it can also adapt to new situations and decide “how to move” based on past experiences, just like humans.
ASI (Artificial Superintelligence)
A superhuman intelligence machine with cognitive abilities much superior to ours that overcomes every aspect of human intelligence.
Machine learning
Machine learning is a subfield of artificial intelligence that studies the development of algorithms that can learn patterns from given input data and make predictions based on them.
Deep learning
Deep Learning is a subset of ML based on the neural network model, made of different layers of artificial neurons that interact like the ones in our brains. Deep learning models are also scalable, which means that they can handle larger and larger amounts of data. The adjective “deep” refers to the numerous layers in a neural network system and therefore its complexity.
Machine learning “teaching” approaches
Unsupervised learning
The goal of an unsupervised learning model is to discover patterns, relationships and structures in no labeled data (without an output), that may be hidden from the human eye.
Some common tasks in unsupervised learning include clustering, where the algorithm groups similar data points together, and dimensionality reduction, where the algorithm reduces the number of features in the data while preserving important information.
Supervised learning
A large amount of data consisting of a set of labeled inputs and outputs is given to the model, and its task is to figure out the relationship between the two. Unlike unsupervised learning, this approach is so-called because it needs external intervention in data scheduling.
The types of supervised learning algorithms are regression, classification, and active learning.
Classification algorithms have a restricted set of outputs they can predict, like a medical diagnosis model, which can only express whether the patient experiences a certain disease or not.
Regression algorithms don’t have limited outputs but sometimes numerical ranges. A house prediction model is a good example of regression.
Active learning
Active learning models use continuous feedback from humans to always updatd their database and consequently evolvd.
Reinforcement learning
A computer program interacts with a dynamic environment in which it has a certain goal (such as driving a vehicle, playing a game against an opponent, or making business decisions). During the learning process, the program is given feedback related to a system of rewards, which serve to make it progress over time. If the action taken is correct, the model gets a reward, if not, a punishment.
Let’s say we are doctors and we have 3 different models useful for our work. An unsupervised learning model can be useful for us to analyze patient data and find similarities, such as characteristics of age or sex that are often found in infected patients, all without the need for outside help.
Next, we train a supervised learning model, which is more specific and through the frequency of features found in patient data can determine whether or not someone is infected.
Finally, we can use a reinforcement learning model to decide what measures to recommend to patients. The latter receives continuous feedback on the effectiveness of drugs, dosages and treatments on patients. By comparing the outcomes of the predictions with patient data, our model creates a plan suitable for each person.
Machine learning models
What is a ML model
A machine learning model is a program that can learn from and generalize unseen data and make predictions based on them.
Some ML models
Decision tree
A decision tree is a supervised learning model that uses a tree-like structure to make predictions. The model structure is editable and made of:
- Conditions the algorithm extracts from the data are used to split the dataset into further smaller datasets, called nodes.
- branches, the possible choices or values that the attribute can take.
- leaves that represent the final predictions or decisions.
Random forest
A random forest is a model that uses many decision trees and makes a prediction by averaging the predictions of each component tree. It generally has much better predictive accuracy than a single decision tree.
Support vector machine (SVM)
A support vector machine is a supervised machine learning classification model that aims to find a hyperplane that best separates different classes of data points in a high-dimensional space.
The hyperplane is determined by maximizing the margin between the closest data points, known as support vectors, of each class. This allows the model to make accurate predictions for new data points by classifying them based on which side of the hyperplane they fall on.
Artificial neural network
A neural network (ANN) is a supervised deep learning model made of different layers of artificial neurons that interact like the ones in our brains.
These neurons, also called units, are wired together through channels, that hold a value, called “weight”. Each unit takes an input value, which can be from the training data or another neuron, multiplies it by the connection’s weight, and adds a value called bias, different from each neuron.
The value obtained (xw +b), is passed through an activation function that determines if it will be transmitted to the next layer of neurons.
During forward propagation, Data go from the input layer to the hidden layers, where they are processed and passed to the output layer. The output is confronted with the actual value with a loss function.
The loss function evaluates a loss score and based on it, the weights and biases are adjusted during backpropagation.
Machine learning real-life applications
Machine learning systems are used in numerous sectors, here are a few examples:
Healthcare
Machine learning is revolutionizing healthcare by diagnosing diseases through image recognition models, enabling personalized medicine based on individual genetic profiles, facilitating drug discovery, and improving patient outcomes through predictive analytics.
Finance
In the financial sector, machine learning is employed for fraud detection, credit scoring, algorithmic trading to optimize investment strategies, implementation of customer service chatbots for efficient client interactions, and market prediction based on past trends.
Retail
Machine learning applications in retail encompass demand forecasting, personalized product recommendations, inventory management optimization, and the prediction of customer churn to enhance customer retention strategies.
Manufacturing
In manufacturing, machine learning is utilized for predictive maintenance, ensuring optimal equipment performance, maintaining quality control standards, optimizing supply chain processes, and improving production scheduling efficiency.
Technology
The technology sector benefits from machine learning in various ways, including natural language processing (NLP) for virtual assistants, image and speech recognition technologies, and the implementation of recommendation systems in online platforms.
Telecommunications
Telecommunications companies leverage machine learning for network optimization, predictive maintenance of infrastructure components, and customer churn analysis to enhance service quality and customer satisfaction.
Automotive
The automotive industry integrates learning into autonomous vehicles to improve safety, traffic prediction, and optimization to improve overall transportation efficiency and develop automated driving systems.
Energy
Machine learning applications in the energy sector include predictive maintenance for equipment, optimization of energy consumption patterns, and fault detection in critical infrastructure.
Agriculture
Agriculture benefits from ML through crop yield prediction models, precision farming techniques that optimize resource usage, and the detection of pests and diseases for timely intervention.
Education
In education, machine learning provides personalized learning experiences and simple assistance to students and automates grading processes to reduce assessment procedures and ensure more time for teachers to focus on youngsters’ growth.
Government
Governments utilize machine learning for fraud detection in public services, security, and surveillance applications, predictive policing strategies, and the optimization of public services through data-driven decision-making.
Marketing and advertising
In the field of marketing and advertising, machine learning is useful in targeted advertising campaigns, customer segmentation for customized marketing strategies, and in general improving the efficiency of advertisements.
Machine learning problems and limitations
Training resources
The biggest obstacle to training a model is creating a suitable dataset for its training, which in cases of very sophisticated machines like ChatGPT can weigh as high as 570 GB, and the time and power that training a large model requires. In 2023 big tech companies like Amazon, Google, and Microsoft invested over $6 billion in training AIs.
The black box theory
The “black box theory“ poses another significant challenge to artificial intelligence development. It refers to a situation where the machine’s process of making decision output is entirely opaque, meaning that even the coders of the algorithm can’t fully understand the pattern that the program extracted out of the data.
This can cause a lot of problems and in the past has led to tragic incidents, when in 2018 a prototype of a self-driving car ran over a pedestrian, or when in some cases the Bing AI chatbot generated hostile and offensive prompts.
The future of machine learning
Although artificial intelligence technologies will certainly be part of our tomorrow and the ML market value is expected to grow by 1270% from 2022 to 2029 (from $30 billion to $410 billion), the future of this field is still unknown.
Machine intelligence is the last invention that humanity will ever need to make
Nick Bostrom, philosopher at the University of Oxford
Future trends
ML experts have recognized some trends that can shape the future of this interesting and life-changing field.
1. Increased speed with quantum computing
The development of quantum computers is believed to contribute to the industry by making model training processes much faster.
2. The creation of big models
The creation of huge models with knowledge and skills in numerous areas of work and daily life.
3. More adaptable algorithms
Scientists will no longer need to rewrite new algorithms from scratch for each of their projects. Instead, they will be able to integrate their work into the new systems, along with the user data.
4. Loss of importance of codes
With the evolution of numerous libraries, in the future, it will be less and less common to write long and tedious code. The focus will shift more toward understanding mathematical algorithms and how they make decisions than to writing the program itself.
5. The rise of reinforcement learning
As reinforcement learning is the best method of instructing a model on how to make correct decisions, it will be of utmost importance in the future, precisely because it will help us to make small, everyday decisions and also businesses to choose the right marketing or investment strategies.
Coexistence between man and machines: will AI replace humans?
Following the latest advancement in ML people are rightly asking if artificial intelligence will replace us humans in every of our functions.
Although a future in which machines will have dominion over the world is dystopian and highly unlikely, people are worried about the stability of their jobs.
In a pessimistic view of the future, the constant development of ML has created algorithms that overcome every aspect of human intelligence, leading to the disappearance of most jobs. Unemployment is skyrocketing and the social divide is increasingly evident.
In a positive view of the future AI helps human workers be more productive and contributes to technological, scientific, and medicinal development.
In both forecasts, the jobs affected will be reasoning jobs. Compared to the latter, the time when AI will have the skills needed to perform manual jobs will be longer, while professions involving emotions and human interactions will never disappear.
A baby learns to crawl, walk and then run. We are in the crawling stage when it comes to applying machine learning
Dave Waters, CEO of AI startup Nara Logics
Conclusion – machine learning
In this post, we mainly discovered that machine learning is a subfield of artificial intelligence born in 1959 that deals with the development of intelligent machines called models.
The latter works by giving data to algorithms that can understand patterns in them.
AIs are employed in many working areas, and in the future the machine learning field will be increasingly important in every aspect of our life.
If you gained useful knowledge from this article consider subscribing to my newsletter to keep informed about the fastest-developing and most important industry in the world.
