In this blog post, we’ll examine the changes brought about by artificial intelligence technologies, such as deep learning, their future potential, and the impact they will have on human society.
What are artificial intelligence and deep learning?
There is an issue that is consistently discussed not only in South Korea but around the world: the shortage of jobs. However, even among humans, competition for jobs is fierce, and for a long time, there have been concerns that a new competitor has emerged. The potential competitors threatening human jobs are automated devices, most notably robots. It is easy to understand that robots equipped with engines and electric motors can generate greater power and move faster than humans. However, quite a few people still question whether these machines can truly match humans in their ability to independently gather and analyze information to make decisions when faced with an entirely new problem.
The field that provides answers to these questions is Artificial Intelligence (AI). And at the heart of current advancements in AI technology lies deep learning. We need to take an interest in deep learning. This is because we can only prepare for the future by understanding what AI—which may soon become a major competitor to humanity—excels at, the principles by which it learns, and the limits of its potential development. Ultimately, we must first understand what this potential competitor is, what it excels at, and how it acquired those capabilities. We have entered an era where the saying, “Know yourself and your enemy, and you will win every battle,” applies to artificial intelligence as well.
To understand deep learning, we must first examine its overarching concept: artificial intelligence. Artificial intelligence generally refers to systems designed to perform intellectual activities such as human learning, reasoning, judgment, and problem-solving. Conceptually, artificial intelligence is sometimes divided into strong and weak AI. Strong AI refers to artificial intelligence capable of general thinking and problem-solving at a level similar to that of humans; however, it has not yet been realized. In contrast, weak artificial intelligence refers to current AI systems designed to perform specific tasks. The AI featured in the movie ‘The Terminator’ is often cited as a prime example of strong artificial intelligence, while technologies such as robot vacuum cleaners and voice assistants fall under the category of weak artificial intelligence.
Furthermore, artificial intelligence can be categorized into Symbolic AI and Connectionism based on its implementation methods. Symbolic AI is an approach that seeks to implement intelligence based on rules and logic, and it has played a significant role in fields such as natural language processing (NLP). In contrast, connectionism is an approach that draws inspiration from the neural network structures of humans and animals, utilizing artificial neural networks—composed of numerous interconnected artificial neurons—to enable learning.
Deep learning, as the name suggests, refers to deep learning. It is a field of machine learning in which artificial neural networks are structured in multiple deep layers to enable machines to autonomously learn features from data. In other words, deep learning is a technology based on artificial neural network structures that enables machines to learn from data and perform recognition, inference, and judgment on their own. Its most significant feature is that machines can discover patterns by learning from vast amounts of data without humans having to manually input every single criterion for judgment. For this reason, deep learning is regarded as a core technology driving the advancement of artificial intelligence today.
How does deep learning work?
The principles of deep learning, which are based on artificial neural networks, fundamentally stem from the mechanisms of artificial neural networks. Developed by drawing inspiration from the structure of the human central nervous system, an artificial neural network consists of numerous artificial neurons interconnected with one another. Each node receives and processes input values before passing them on to other nodes; this process repeats to produce the final result.
The process of transmitting values is not random. Depending on the characteristics of the input data, different weights are assigned to each connection, and the extent to which information is transmitted varies according to these weights.
Each node functions like a single function; multiple functions combine to form a larger function, and as this process repeats, the final function that constitutes the entire neural network is completed. When an input value is provided, the neural network undergoes this computational process to produce a final result.
In an artificial neural network, learning can be defined as the process of finding the functions and weights that produce optimal results. By repeatedly learning the relationship between inputs and outputs using previously accumulated data, the network is enabled to make increasingly accurate judgments. For example, if the input contains information such as “blue,” “clouds,” “sun,” “airplane,” and “sparrow,” the neural network can analyze the relationships among these elements through a vast amount of training data and ultimately derive the result “sky.”
Where is deep learning being used, and how will it evolve in the future?
Based on this operating principle, deep learning is currently being utilized in various fields. First, looking at the field of automatic speech recognition, most speech recognition systems today operate based on deep learning technology. Smartphone voice assistants, real-time voice translation, voice search, and automatic caption generation have all seen dramatic improvements in performance alongside the advancement of deep learning. In the field of speech recognition, deep neural network models utilizing large-scale data have demonstrated much higher accuracy than traditional methods and have established themselves as the current standard technology.
Deep learning is also being utilized as a core technology in the field of image recognition. In particular, the field of computer vision began to develop rapidly after Geoffrey Hinton’s research team achieved performance that significantly outperformed existing machine learning techniques using a deep learning-based model at the 2012 ImageNet competition, a large-scale image recognition contest. Since then, the importance of deep learning has grown significantly across various fields—including image recognition, object recognition, facial recognition, medical image analysis, and autonomous driving technology—and its performance continues to improve.
In addition, deep learning is being utilized in a wide variety of fields, including new drug development, toxicity prediction, customer relationship management, financial services, manufacturing process optimization, autonomous driving, generative AI, and robot control. In particular, with the rapid advancement of generative AI technologies—including large language models (LLMs)—the scope of deep learning applications is expanding far beyond what it was before.
Limitations and Future Prospects of Deep Learning
So, what does the future hold for deep learning? While current deep learning systems demonstrate outstanding performance, they also have several limitations. For a long time, some researchers have pointed out that deep learning is not yet capable of fully replicating human thought.
Cognitive scientist Gary Marcus has assessed that deep learning has limitations in its ability to understand causality or engage in logical reasoning, and that it also struggles to comprehend common sense and abstract concepts. He also pointed out that its ability to adapt to new environments or learn from limited amounts of data falls short compared to humans. These criticisms remain significant challenges in AI research today.
On the other hand, many experts hold a high regard for the potential of deep learning. In fact, over the past decade or so, deep learning has achieved remarkable advancements in various fields, including speech and image recognition, as well as natural language processing, medicine, and scientific research. Recently, with the rapid development of generative AI, it has reached a level where it can converse naturally with humans, write text, generate images, and solve complex problems.
These changes demonstrate that AI technology is rapidly spreading not just in the near future, but already in the present.
Market size also supports this trend. The global AI market has maintained a very rapid growth trajectory in recent years, and global market research firms project that this high growth rate will continue in the future. Furthermore, governments and global corporations around the world are continuing to make massive investments in AI research and development and related industries, making it highly likely that the advancement of deep learning technology will continue.
What should we prepare for in the age of artificial intelligence?
To summarize the above, deep learning can be described as a core technology that enables machines to learn from vast amounts of data to classify and identify information, as well as perform reasoning and decision-making. As long as advancements in big data technology, improvements in computing power, and ongoing research into artificial neural network algorithms continue, deep learning is highly likely to evolve further.
In fact, today’s artificial intelligence has advanced to the point where it not only perceives visual and auditory stimuli but also understands and generates language, and processes various forms of information in an integrated manner. If these technologies become even more sophisticated, it is highly likely that AI will collaborate with humans or take over certain tasks in a much broader range of fields in the future. Consequently, the prediction that work methods in many professions will change and that some jobs will be automated is becoming a reality.
However, the emotional and empathetic capabilities, ethical judgment, creative thinking, and ability to form social relationships—which are unique to humans—are still considered areas that artificial intelligence will find difficult to replace. Furthermore, as long as the reliability and safety of artificial intelligence are not fully guaranteed, the human role is likely to remain crucial in fields such as medicine, law, politics, and education, where a high level of responsibility and ethical judgment is required.
Advances in artificial intelligence and robotics will bring significant changes to society and industry as a whole in the future. These changes may present new opportunities while also posing new challenges. Therefore, individuals must continuously develop new competencies to adapt to the evolving technological environment, and governments and businesses must also actively respond to these changes through improvements in education and institutional frameworks. If these preparations are made collectively, we will be able to reap the full benefits of artificial intelligence while minimizing its negative social impacts.
