General Artificial Intelligence (AGI) is considered a misnomer, not because current artificial intelligence systems are far from achieving general intelligence, but because they fundamentally cannot use intelligence to perform various tasks and learn like humans. Although it possesses capabilities in perception, reasoning, learning, and creativity, and can adapt to different environments and tasks, it is not limited to a specific domain, but these abilities are fundamentally different from those of humans. On the surface, current artificial intelligence systems still have many limitations and challenges, making AGI a misnomer. For example:
1. Limitations of Learning Ability
Current artificial intelligence systems require a large amount of labeled data for training and learning, and they can only learn within specific domains, unable to perform transfer learning through experience.
2. Insufficient Understanding Ability
Artificial intelligence systems face significant challenges in understanding language and context. They do not have true ‘understanding’ ability; they merely process information through pattern matching and statistical methods.
3. Creativity and Innovation Ability
Current artificial intelligence systems still cannot create and innovate like humans. They can only operate within existing data and rule frameworks, lacking the ability for imagination and creative thinking.
4. Lack of Self-Awareness
AGI also needs to possess self-awareness and self-reflection capabilities, as well as an understanding of itself and the surrounding environment. Current artificial intelligence systems cannot achieve this.
The above points highlight the real challenges that still exist in achieving true AGI, which is unrelated to the fact that current artificial intelligence systems can only perform in specific tasks and domains and cannot reach true general intelligence levels.
1. Human intelligence can roughly be divided into learning and non-learning types.
Learning intelligence refers to the ability of humans to improve and elevate their intelligence levels through learning and experience accumulation. Humans have the ability to learn, acquiring new knowledge and skills through information gathering, observation, and practice. Learning intelligence enables humans to adapt and respond to different environments and situations, continuously progressing and developing. Human learning differs from machine learning, capable of generating a range of uncertain implicit rules and orders.
Non-learning intelligence refers to the innate intelligent abilities that humans possess without needing to acquire them through learning and experience. For example, basic cognitive abilities such as perception, memory, thinking, and reasoning belong to non-learning intelligence. These abilities are innate and acquired through genetics and evolution.
Although learning and non-learning intelligence play different roles in human intelligence, they intertwine and influence each other, collectively forming human comprehensive intelligence. Human learning intelligence allows us to continuously progress and adapt to changes, while non-learning intelligence provides the foundation and framework for learning. Together, these two characteristics of intelligence drive the development of human cognition and wisdom.
2. Current mathematical tools struggle to support machine intelligence in mimicking human learning intelligence.
Current mathematics and tools indeed struggle to fully support machine intelligence in mimicking human learning intelligence. Although many advancements have been made in the fields of machine learning and artificial intelligence, achieving true human learning capabilities still faces some challenges.
Firstly, mathematics and tools rely on a large amount of data when constructing machine learning models. However, compared to humans, the data requirements for machine learning models are extremely high, often needing millions of data samples to train the models. This is impractical for many tasks and may also involve privacy and security issues.
Secondly, machine learning models are typically built on statistical methods. These methods can solve specific problems but may have limitations when dealing with complex learning tasks. For example, machine learning models still struggle with fuzzy concepts, reasoning, and creative thinking.
Additionally, human learning ability relies not only on mathematics and tools but also on rich experience, intuition, and subjective judgment. These factors are difficult to fully simulate through mathematical modeling and machine learning algorithms.
Therefore, to achieve machine learning intelligence, in addition to the development of mathematics and tools, new methods and technologies need to be explored, such as deep learning, reinforcement learning, and interdisciplinary research in cognitive science. Only by integrating various methods can we better support machine intelligence in mimicking human learning capabilities.
3. Machine intelligence cannot achieve human non-learning intelligence.
Machine intelligence may outperform humans in some areas, such as the speed and accuracy of processing large amounts of information, but it cannot achieve human non-learning intelligence. Human non-learning intelligence includes emotions, creativity, intuition, and art, which machines currently cannot simulate or reach.
Emotions are a unique human ability, encompassing the understanding, expression, and feeling of various emotions, while machines can only process emotional expression and recognition through algorithms. Creativity is a unique human ability to generate new and unique ideas and works. Machines can generate new content through learning and algorithms but lack originality and creativity.
Intuition is a crucial human ability based on experience and instinctive feelings for decision-making and judgment. While machines can simulate intuition through learning and pattern recognition, they lack human life experience and emotional thinking.
Art is a domain unique to humans, including the creation and appreciation of forms such as music, painting, and literature. Although machines can generate artistic works through learning and algorithms, they lack genuine emotion and inspiration.
In summary, while machine intelligence can surpass humans in some aspects, human non-learning intelligence is unattainable for machines. These non-learning intelligences are vital components of human culture and existence, distinguishing humans from machines.
4. The inability to form values is a key issue for machine intelligence.
Although machine intelligence can extract useful information from data through learning and reasoning and perform specific tasks, to impart value to machine intelligence, ethical, moral, and social value factors need to be considered and introduced.
Value is a subjective concept, so determining what value is appropriate for machine intelligence presents a challenging issue. For example, how should an autonomous vehicle make decisions in dangerous situations? Should passenger safety be prioritized, or should minimizing harm to other pedestrians take precedence?
There are also more complex issues, such as bias and discrimination present in machine intelligence in design and recommendation algorithms, as well as how to balance privacy protection with data utilization. These issues require comprehensive consideration of various values, such as fairness, privacy, transparency, and freedom.
Therefore, to address the value issue of machine intelligence, multiple dimensions need to be considered and discussed, along with the formulation of corresponding principles and norms. This requires the participation and joint efforts of society as a whole, including government, academia, businesses, and the public. Only then can machine intelligence truly bring positive impacts to humanity and align with human value systems.
5. The development of general intelligence faces three major bottlenecks: technological, biological, and social.
The technological bottleneck is reflected in the need for artificial intelligence systems to possess higher computing power, more advanced algorithms, and more effective data processing methods to achieve more complex and intelligent functions.Moreover, the ‘calculating’ of general intelligence not only includes computational ability but also involves strategic (planning) ability, which refers to the capability of intelligent systems in handling complex problems, reasoning, and decision-making. Current artificial intelligence technology often faces challenges due to algorithm design, data quality, and model interpretability when addressing various real-world complex problems. With continuous technological advancements and deeper research, we can expect greater breakthroughs in intelligent systems’ strategic abilities, enabling them to better tackle various complex challenges.
The biological bottleneck mainly reflects our limited understanding of the cognitive mechanisms of the human brain. Achieving a similar level of intelligence requires deeper research in neuroscience and cognition.Specifically, this includes:
1. Understanding the complexity of the human brain and cognitive systems
The human brain is a highly complex organ, and its working principles are not yet fully understood. The interactions between neurons in the human brain are very complex, and our understanding of these interactions is still limited. To develop truly general intelligent artificial intelligence systems, a deeper understanding of how the brain works is needed, and these principles must be applied to computer systems.
2. Processing unstructured information
Humans can easily process unstructured and ambiguous information, but this presents a greater challenge for computers. For example, in the field of natural language processing, understanding the semantics, grammar, and context of human language is a huge challenge for computer systems. Although some progress has been made, many difficulties remain in enabling computer systems to possess human-level language understanding capabilities.
3. Solving complex problems
General intelligence requires the ability to solve various complex problems, which necessitates breakthroughs in knowledge and skills across different fields, including reasoning, planning, and decision-making.
The social bottleneck includes the integration of artificial intelligence systems with human society, for example, cultural differences, ethical and moral considerations, and privacy protection, all of which are significant factors affecting the development of artificial intelligence. Specifically, it involves:
1. Understanding and responding to social cultures
Considering the differences in behavioral norms and values across different cultural and social backgrounds, a general intelligent system needs to adapt to and respect different cultures. For instance, a customer service robot intended for the global market needs to understand and respond to users from different cultural backgrounds, which requires a deep understanding and handling of cross-cultural communication.
2. Addressing ethical and moral issues
The development of general intelligence must consider ethical and moral issues, including privacy protection, fairness, and safety, to ensure its development and application meets ethical standards and societal expectations. Considering ethical and moral issues in artificial intelligence decision-making is a complex task. For example, autonomous vehicles need to be able to handle moral dilemmas, such as how to make decisions in the event of an unavoidable accident. This requires integrating ethical and moral principles into the design and decision-making processes of artificial intelligence systems.
3. Interacting with humans
General intelligence needs to effectively interact and communicate with humans, which requires addressing challenges in natural language understanding, emotion recognition, and emotional expression to ensure a good user experience and human-computer interaction.
Overcoming these bottlenecks requires interdisciplinary collaboration and continuous innovative efforts. Only by achieving breakthroughs in technology, biology, and society can general intelligence move towards more mature and comprehensive development.