In this blog post, we explore the technologies and elements required to create strong AI capable of autonomy, emotion, and reason.
From self-driving cars, AlphaGo, and fingerprint recognition devices to the self-thinking, self-acting robots seen in movies, all are machines based on artificial intelligence. As AI technology rapidly advances, it has become deeply embedded in various sectors of modern society, and its applications are becoming increasingly diverse. The various smart devices and automation systems used in our daily lives are designed based on AI, with the goal of making human life more convenient. However, fully autonomous robots that think and act on their own, as seen in movies, differ significantly from the AI technology we currently possess. This is because real-world AI is primarily limited to processing predetermined data and carrying out given tasks. So, what elements need to be added to transform simple machines into robots that act independently? To understand this, we need to examine the types of AI and the direction of technological development in greater detail.
First, let’s examine the difference between narrow AI and general AI, which represent the most significant distinction in the stages of AI development. The fingerprint recognition machines and self-driving cars mentioned earlier fall under narrow AI. Narrow AI refers to AI that lacks self-awareness and can only act according to how it was programmed by humans. It consists of algorithms designed for specific purposes and is used solely to perform specific problems or tasks. For example, the well-known AlphaGo also falls under narrow AI; it was designed as an algorithm specialized for making moves in the game of Go. While it can perform the specific task of playing Go exceptionally well, it is completely useless in any situation outside of Go. In this way, narrow AI is optimized only for specific tasks and is utilized solely for limited purposes, much like a tool.
Narrow AI is further subdivided into general-purpose narrow AI and general-purpose narrow AI. General-purpose narrow AI simply performs tasks within a specific field. For example, tasks such as identifying spam emails or detecting cancer cells in medical images are examples of general-purpose narrow AI. On the other hand, general-purpose narrow AI is a more advanced form that includes learning capabilities. As a result, it can learn to replicate a recipe it has never attempted before by observing someone else prepare it, and it can take over and perform simple tasks. General-purpose AI can be considered the initial stage of transition from AI to strong AI. At this stage, the ability to learn new information and adapt to situations—going beyond simple repetitive tasks—is required, making machine learning and deep learning technologies crucial.
In contrast, entities like the Terminator, the robots in *I, Robot*, and JARVIS from *Iron Man*—which we often see in movies—are examples of strong AI that act with self-awareness. Strong AI refers to artificial intelligence that possesses a sense of self and is capable of making its own judgments and taking action. The goal of strong AI is to achieve a level of intelligence comparable to that of humans. Such AI, which encompasses not only reason but also emotion, possesses autonomy, allowing it to make decisions and respond to situations on its own without specific instructions. The most significant difference between strong AI and weak AI is the presence or absence of a self; if a self exists, the machine can act based on its own goals and judgments.
Developing strong AI requires a level of reason and emotion comparable to that of humans. While many people do not question whether AI should possess reason, there is much debate over whether it needs emotion. Reason is essential for analyzing given data and making judgments, but emotion is a necessary element for understanding complex human social contexts and emotional situations beyond simple analysis. Current AI is gradually acquiring the ability to analyze and make judgments based on data, but it is still difficult for it to mimic the complex emotional states of humans. For example, AI systems such as chatbots or virtual assistants can mimic basic emotions, but they do not actually feel emotions; they merely display predetermined responses. However, since the interaction between reason and emotion enables better decision-making, there is a view that both of these elements are necessary for the development of strong AI.
The necessity and feasibility of developing strong AI are currently the subject of extensive debate in both academia and industry. If we were to develop strong AI, it could bring about significant changes in human society. In particular, the assistance of AI capable of making sophisticated decisions that surpass human judgment is expected to have great social value in fields such as healthcare, education, law, and the environment. For example, in an aging society, if AI makes medical judgments and diagnoses, the accuracy and efficiency of treatment would increase, thereby improving the quality of healthcare services. Furthermore, if AI performs analysis and decision-making to address environmental issues, it could significantly contribute to moving toward a sustainable future.
To evolve weak AI into strong AI, we must first develop the rational judgment capabilities of current weak AI to a much higher level. Expert systems help machines process large amounts of data and understand various situations based on conditions, but they still fall short of human higher-order thinking. Through machine learning—technology that enables machines to create their own rules and analyze new data—computers can acquire more flexible thinking abilities. Genetic algorithms can also serve as a powerful tool in this regard. Rather than seeking a single specific answer, genetic algorithms focus on finding the optimal solution among possible alternatives, mimicking the way humans adapt to diverse situations.
Endowing strong AI with emotions is also a critical challenge. Human emotions are highly complex and subjective, so there are limitations to simply programming the AI to react with specific emotions based on a given situation. Currently, emotions are mimicked by programming the system to replicate predetermined emotions based on specific situations, but this is difficult to consider a true expression of emotion. To develop AI capable of genuine emotion, we must gain a deeper understanding of the human emotional system and research models that reflect the various elements of emotion. If emotions are endowed upon AI, it can elicit more human-like responses, making communication between humans and AI more natural.
If we endow weak AI with reason and emotion in this manner, we will eventually be able to create strong AI. Once such technology is perfected, strong AI will transcend the role of a mere tool to become a partner capable of collaborating with humans to solve various problems. With future technological advancements, strong AI is highly likely to become a reality rather than a distant prospect.
