Machine learning is the art of analyzing and learning from big data to solve problems, and is an evolutionary form of artificial intelligence. It is revolutionizing a wide range of industries and is having a profound impact on society as a whole. Advances in machine learning will change the structure of existing jobs, create new types of work, and play a role in policy decisions.
What is machine learning?
The rise of AlphaGo in 2016 was one of the first events that made people realize that machine learning, especially deep learning, had already overtaken humans in learning methods that we had only heard about at conferences. Realizing the power of machine learning, governments and companies have also begun to develop a Korean version of AlphaGo. As the development of the Internet has led to a hyper-connected society where people and things are closely connected, it is necessary to understand the concept, types, and prospects of machine learning, which has emerged as a new mainstream in the IT field.
Concept and Features of Machine Learning
To understand machine learning, we must first understand the concept of big data. Big data generally refers to the storage of information that has a differentiated scale in terms of velocity, volume, and variety that is incomparable to traditional data. Machine learning is the most practical way to understand, analyze, and derive new meaning from big data.
Machine learning is differentiated from data mining in terms of the data it targets: while the latter targets structured data in a typical format, the former can be applied to more flexible data. For example, if you’re mining data about animals, and you’re dealing with pre-categorized categories such as species, gender, and age, machine learning is not bound by these categories and can add new ones, such as size and weight, as needed.
Machine learning is also similar to artificial intelligence (AI) in that it solves problems on its own, but it goes a step further than traditional AI methods in that it can react and adapt to additional data. And while traditional AI emphasizes deterministic algorithms with conditional statements and logical expressions, machine learning is more akin to statistics and accounts for the stochastic side of things.
Applications of machine learning
Machine learning is driving innovation in a variety of fields. For example, in healthcare, it’s being used to analyze pathology data for early detection and prediction of diseases like cancer, and in finance, it’s being used to predict stock prices, detect fraud, and make personalized financial product recommendations. In manufacturing, machine learning is used to improve the efficiency of quality control and maintenance operations; in transportation, such as self-driving cars, it analyzes road conditions and helps find the best route in real time; and in customer service, chatbots are evolving to answer customer questions more accurately and quickly with machine learning.
Types of machine learning
Machine learning can be broadly categorized into supervised, unsupervised, and reinforcement learning.
Supervised learning refers to the labeling of data, or the assumption that there is a correct answer. For example, identifying whether a new piece of data is a car by looking at different pictures of cars and finding common features of cars is a result of learning that there is already a correct classification for a car, and we can improve our performance in classifying whether it fits that correct classification.
Second, unlike supervised learning, unsupervised learning is based on probabilistic models that don’t have a right answer. It only tells us what data is relevant in close proximity to each other, but it leaves it up to human interpretation as to what concept a cluster of close data represents. For example, unsupervised learning of the phenomenon that private conversations on social media decrease whenever there is a social event in the media may result in a very low association between Argentina and South Korea, but a high association with Uruguay, which is not a result of unsupervised learning itself, but can be attributed to geographic proximity or rivalry, depending on the researcher’s interpretation.
Finally, reinforcement learning is inspired by the interconnected network structure of neurons in the human brain, and is said to be able to solve problems efficiently by sharing, evaluating, and improving the results of each algorithm. Although it has a long history, it has only been practically practiced since 2006 and is currently the hottest topic. In particular, deep learning, which is a representative of reinforcement learning, is not supervised learning in which the concept of an object is given in advance, but through continuous learning of image data, it has succeeded in distinguishing which photo corresponds to a cat by itself, and this distinction exceeds human limits in that it not only identifies a face consisting of eyes, nose, and mouth, which can be seen at human eye level, but also finds features down to the lowest unit, the pixel, which is invisible to humans.
The social impact of machine learning
Advances in machine learning are not just a technical achievement; they are having a profound impact on society at large. In the labor market, increased automation and efficiency are leading to the decline of traditional jobs and the creation of new ones. For example, automation of production lines is reducing the number of repetitive tasks, while new occupations such as data analysts and AI specialists are emerging. The application of machine learning techniques to politics and economics is also making data-driven decision-making more important, which has the potential to increase transparency and efficiency in policymaking, but also risks increasing social inequality through misuse of data or biased algorithms.
The future of machine learning
As the algorithms of machine learning become more advanced, humanity will be able to coexist and collaborate with artificial intelligence to make great strides in solving upcoming problems. While it has been mainly applied to engineering problems, in the future it will be applied to more diverse and complex social problems, emphasizing the importance of human interaction with machine learning. Furthermore, researchers are moving beyond implementing machine learning using computers to develop computers that are physically suitable for machine learning. Deep learning, after all, is inspired by the connections of the neurons that make up the human brain, so it is likely to develop in a similar way.
