It's more than 60 years that artificial intelligence has emerged as a real technology from the world of sci-fi movies and stories and gone through many ups and downs. It's natural that along this path, artificial intelligence sometimes has become a hot topic of news and sometimes also been neglected. However, today, with the explosive development of areas like big data, high-powered parallel processing, and advanced neural algorithms, we see a new resurgence in artificial intelligence, and many major technology players struggle to take the lead in this path.
While there are different forms of AI in various areas, it's safe to say that today machine learning is the most widely used mechanism for achieving artificial intelligence.
What is machine learning?
As mentioned earlier, machine learning is one of the branches of artificial intelligence. Rule-Based engines, evolutionary algorithms, and Bayesian statistics are all examples of tools for reaching artificial intelligence. Many early AI programs such as IBM's Deep Blue (which defeated Gary Kasparov in 1997) were rule-based and dependent on human programming. On the other hand, machine learning is a tool by which computers can teach themselves, and set their own rules. In 2016, Google's DeepMind could beat the champion in Go through the use of machine learning. DeepMind did this by training itself on a large data set of expert moves.
Different types of machine learning are:
- Supervised learning: Here, the trainer provides the computer with certain rules that connect an input (for example, a feature of an object such as "soft") to an output (the object itself, such as cotton).
- Uncontrolled Learning: In this method, the computer receives inputs and is left alone to discover patterns.
- Reinforcement learning: Here, a computer system continuously receives inputs and constantly improves its intelligence. For example, you can think about a driverless car receiving inputs about the road.
A vast amount of data is required to train algorithms for machine learning. First of all, training data should be labeled (like the GPS location attached to a photo). Then, it turns to classification. This happens when the features of the Intended object are labeled and sent into the system along with a set of rules that lead to a prediction. For example, "red" and "round" are two inputs to the system that leads to the output of "apple". Also, a learning algorithm can be left alone to create its own rules that can be applied when it is provided with a large set of the objects. For example, when algorithm encounters a group of apples, it figures out that they all have the same features like "round" and "red".
Machine learning in many cases involves "deep learning". Deep learning is, in fact, a subset of machine learning and uses layered algorithms that form a network to process information and reach predictions. The point that distinguishes deep learning is that the fact that the system can learn on its own, without human training.
Important machine learning events
The fact is, machine learning became very popular in the 1990s, but later this popularity dropped, until recently that we see its return to hot headlines. Some of the significant events in this period are:
- Google Brain was created in 2011, a deep neural network that was able to identify and categorize objects.
- The DeepFace algorithm was introduced in 2014. This algorithm could identify individuals from a set of photos.
- Amazon launched its machine learning platform in 2015. Microsoft also introduced a Distributed Machine Learning Toolbox this year.
- The Google DeepMind program, called AlphaGo, in 2016, defeated Lee Sedol, the world champion in the complex game of Go.
- In 2017, Google announced that its machine learning tools can recognize objects in photos and understand speech better than humans.
Why is Machine learning important?
Aside from the tremendous power that machine learning has shown to beat humans at games like Jeopardy, chess, and Go, it has numerous practical applications. For example, machine learning tools are used to recognize faces of people in photos, translate messages, and find locations around the world that have certain geographic features. IBM Watson helps doctors make decisions on cancer treatment. Cars use machine learning to gather information from the surrounding environment. On the other hand, machine learning is at the center of fraud prevention efforts. For example, it has been proven that the combination of unsupervised machine learning and human experts can act very accurately in detecting cyber threats.
Eventually, almost every organization that wants to use its information to gain insights, improve relationships with customers, increase its sales level or remain competitive in a specific area, will rely on machine learning. This technology has applications in various fields of government, business, and education, and almost anyone who wants to make informed predictions and has a large enough data set can use machine learning to achieve its goals.