When we hear terms like machine learning and artificial intelligence, we assume this technology is relatively recent. Many look at it as very modern innovations developed within the last decade. However, you may be surprised to learn that the history of machine learning dates back to the 1940s.
What ML is, how it works, what methods of machine learning are there, and where it’s actively used — all of this is covered in the article below.
What machine learning is
Machine Learning (ML) is a branch of artificial intelligence focusing on developing algorithms and statistical models that enable computers to learn from data, make predictions or decisions.
A simple example of a machine learning algorithm is the music streaming service Spotify. To decide which new songs or artists to recommend, Spotify’s algorithms link your preferences with those of other listeners with similar music tastes. This method, often simply called AI, is used across many services offering automated recommendations.
Another example is speech recognition software that converts voice messages into text. Or self-driving cars and driver assistance features, such as blind-spot detection and automatic brakes.
The history of machine learning
Starting off
In 1943, Walter Pitts and Warren McCulloch in their paper ‘A Logical Calculus of the Ideas Immanent in Nervous Activity’ introduced the first mathematical model of neural networks.
In 1949, Donald Hebb published ‘The Organization of Behavior’ book, linking behavior with neural networks and brain activity. It helped establish the basis for machine learning development.
In 1950, Alan Turing created the Turing Test to determine whether a computer has real intellect. To pass, the machine had to convince a person that it was also a human being.
Games and pathfinding
The first computer learning program was written by Arthur Samuel in 1952. It was a checkers game. It was used to pump up the IBM computer — the more it played, the better it studied moves to make up a winning strategy.
In 1957, Frank Rosenblatt developed the perceptron, the first neural network for computers that simulated the human brain’s thought processes.
The next major ML breakthrough came in 1967 with the Nearest Neighbor algorithm. It enabled computers to recognize the simplest patterns of image recognition. This algorithm was used to plan sales routes for traveling salesmen.
Twelve years later, in 1979, Stanford’s AI Laboratory (SAIL), founded by John McCarthy, developed the Stanford Cart, a robot that could navigate around obstacles in a room.
And in 1981, Gerald De Jong introduced Explanation-Based Learning (EBL), where a computer analyzes training data and creates a general rule to follow, disregarding irrelevant information.
Major breakthrough
In the 1990s, machine learning research shifted from a knowledge-based approach to a data-driven one. Scientists began creating programs that analyzed large datasets, learned from the results, and made inferences. In 1997, IBM’s Deep Blue shocked the world by defeating world chess champion Garry Kasparov.
In 2011, Google Brain was developed, a deep neural network that learned to detect and classify objects at the level of a cat. Next year, the X Lab created an ML algorithm capable of watching YouTube videos on its own to identify cats there.
In 2014, Facebook developed DeepFace, a software algorithm capable of recognizing or verifying people in photos with human-level accuracy.
In 2015, Microsoft created the Distributed Machine Learning Toolkit, enabling efficient distribution of ML tasks across multiple computers.
The same year, over 3,000 AI and robotics researchers, endorsed by Stephen Hawking, Elon Musk, and Steve Wozniak, signed an open letter warning of the dangers of autonomous weapons that could select and strike targets without human intervention.
In 2016, Google’s AI defeated a professional player in the Chinese board game Go, considered the world’s most challenging board game.
Present day
In 2020, during the COVID-19 pandemic, OpenAI unveiled the revolutionary GPT-3 natural language processing algorithm, capable of generating human-like text based on prompts.
Today, Machine Learning Cloud is gaining popularity. Training precise ML models requires massive amounts of data, computing power, and infrastructure. Cloud computing makes machine learning more accessible, flexible, and efficient, allowing developers to build ML algorithms faster. It is ML Cloud that helps accelerate and manage the entire ML project lifecycle.
How machine learning works
- Data collection. Information can be retrieved from such sources like databases, sensors, or the internet.
- Data preprocessing. Once collected, the data must be processed to verify its quality and eligibility for analysis.
- Model training. The algorithm must be trained to make predictions or decisions based on the input data.
- Feature selection and development. The ML model selects the most relevant features from the input data, which significantly impacts its performance.
- Model evaluation and optimization. After training, the model is evaluated to determine its accuracy against desired criteria.
- Deployment and monitoring. Once successfully trained and evaluated, the model is deployed in real ML applications.
Types of Machine Learning Methods
Supervised machine learning
The model is trained on labeled data with ready answers. The algorithm must choose the correct answer to a hypothesis, and humans monitor the result.
Unsupervised machine learning
The model is trained on unlabeled data. The algorithm identifies patterns and features that differentiate objects, making it suitable for predictions and automated data cleaning, language recognition, and understanding.
Semi-supervised machine learning
Both labeled and unlabeled data are used, with the algorithm handling the remainder of the labeling based on set parameters. This is useful for processing large files.
Reinforcement learning
The algorithm learns to use a trial-and-error method, based on a point system where actions are rewarded. This is similar to a game where players earn bonuses for correct actions.
What is Deep Learning?
Deep Learning (DL) is a subtype of AI and ML that uses multilayered artificial neural networks to achieve high accuracy in tasks such as object detection, speech recognition, language translation, and more.
Artificial neural networks continually receive algorithms and data, allowing for more efficient learning. The more data, the better the learning process. Over time, the neural network encompasses more levels, enhancing its effectiveness.
Unlike ML, DL requires high-performance equipment, powerful graphics processors, and video cards, which can be addressed through NVIDIA GPUs.
GPUs help researchers and data analysts speed up training from weeks to hours. Instead of purchasing expensive equipment, training can be conducted in the cloud.
Where Machine Learning is Used
- Social networks. For instance, Facebook tracks user actions, likes, comments, and time spent on specific posts. ML learns from this behavior and suggests friends and pages of interest.
- Retail. Product recommendations are one of the most popular applications of ML. Sites use ML and AI to track user behavior based on previous purchases, search information, and cart history, then recommend products.
- Healthcare. ML analyzes data from various sensors to assist doctors in real-time diagnosis and treatment. Researchers are developing ML solutions that detect cancerous tumors and diagnose eye diseases.
- Financial services. ML technology helps investors identify new opportunities by analyzing stock market movements, evaluating hedge funds, or calibrating financial portfolios. It also helps detect high-risk loan clients and reduces signs of fraud.
- Manufacturing. ML helps companies improve logistical decisions, including asset management, supply chains, and inventory control.
Pros and cons of machine learning
Advantages:
- Efficiency. ML can automate repetitive tasks, leading to an efficiency boost and allowing a person to focus on more complex and creative tasks.
- Consistency. Machines can perform tasks consequently, without getting tired, which causes fewer errors compared to manual processes.
- Insights based on data. ML algorithms can analyze big datasets to detect patterns and ideas that human can miss.
- Adaptiveness. Models can adapt to new data, improving with time.
- Error reduction. ML can get more precise and accurate than the traditional methods, reducing mistakes in such fields as medical diagnostics or financial forecasting.
Disadvantages:
- Data quality. ML models efficiency is highly dependent on data quality and quantity. Corrupted data could lead to inaccurate models.
- Privacy. Operating large massives of data, especially with personal information, leads to concerns on privacy and security.
- Expenses. ML model training can be resource-intensive, requiring huge computing power and time.
- Experts. Without expert help, it is hard to interpret results and remove uncertainty.
The future of machine learning
Machine learning’s future involves significant applications in quantum computing, enabling faster data processing and enhancing algorithms’ ability to analyze and draw conclusions from datasets.
Software applications will become more interactive and intelligent, thanks to ML-powered cognitive services like visual recognition, speech recognition, and language comprehension.
ML will integrate increasingly with the Internet of Things (IoT) and smart systems, allowing IoT devices to analyze live data, making smart homes and cities more efficient.
The future of ML is associated with highly personalized AI applications. From healthcare providing treatment plans tailored to genetic codes to educational systems adapting learning to individual student needs, ML will achieve levels of personalization previously unattainable.
Machine learning systems will have more robust frameworks where human feedback is integral to the learning process. This approach ensures that ML systems remain aligned with human values and can adapt to scenarios where humans make the final decision.
