Article to Understand Artificial Intelligence - AI

 

Will humans be replaced by AI in the future? 

Read the article to understand the types and development stages of artificial intelligence

If there is one technology that has revolutionized the 21st century, it has to be artificial intelligence. Sundar Pichai, the head of Google, once said: "Artificial intelligence will bring about changes in our lives and work, even more than fire and electricity." Although it is full of gimmicks, it is undeniable that artificial intelligence is revolutionizing people's lives. lifestyle. Tools and lifestyles that were previously only known in science fiction are now infiltrating our daily lives from all angles with the blessing of artificial intelligence, driving technological progress, industrial upgrading, and promoting the overall rapid development of the market economy. Therefore, it becomes very important to correctly understand the concept of artificial intelligence. This article will help you understand the definition, stages, types, and research areas of artificial intelligence.

Definition of artificial intelligence

In 1956, John McCarthy (1927~2011) proposed the term "artificial intelligence (AI)" at the Dartmouth Conference. He defines artificial intelligence as:

The science and engineering of making intelligent machines. 

Artificial intelligence can also be defined as the development of computer systems capable of performing tasks that require human intelligence, such as making decisions, detecting objects, solving complex problems, and more.

 

Stages of Artificial Intelligence

Many articles argue that Artificial General Intelligence, Artificial Narrow Intelligence, and Artificial Super Intelligence are different types of AI. In fact, more precisely, they are the three stages of artificial intelligence. 

Weak Artificial Intelligence (ANI)

Weak artificial intelligence, also known as narrow artificial intelligence. At this stage, the machine does not have any thinking ability, but only performs a set of predefined functions, such as speech recognition, image recognition, etc. It is an artificial intelligence that is good at a single aspect, similar to advanced bionics. They exist only to solve a specific specific task, and are mostly statistical data from which models can be derived. For example, AlphaGo can only play Go and cannot perform other tasks.

Examples of weak AI include Siri, Alexa, self-driving cars, AlphaGo, the humanoid robot Sophia, and more. So far, almost all AI-based systems are weak AI.

Strong Artificial Intelligence (AGI)

Strong artificial intelligence, also known as general artificial intelligence. Unlike weak AI, strong AI can deal with different levels of problems like humans, not just perform a predefined set of functions. Not only that, strong AI also has the ability to learn by itself and understand complex concepts. It is also true that the development of strong artificial intelligence is much more difficult than weak artificial intelligence. In fact, the goals held by the international mainstream academic community of artificial intelligence are limited to weak artificial intelligence. At present, few people have conducted research on strong artificial intelligence, and corresponding results have not yet been formed.

Strong AI is also seen by many scientists as an existential threat to humanity, Stephen Hawking warns:

"The development of Full Artificial Intelligence may mean the end of human civilization... Once artificial intelligence is freed from its shackles, it will continue to redesign itself at an accelerated pace. Human beings, due to the time constraints of biological evolution, cannot compete with it, and it is very likely that will be replaced."

Super Artificial Intelligence (ASI)

When weak artificial intelligence has been mostly realized and strong artificial intelligence is constantly approaching through deep learning, the concept of super artificial intelligence is ready to emerge. At this stage, the capabilities of computers will surpass that of humans. Currently, strong artificial intelligence is a hypothetical scenario depicted in movies and science fiction: machines have taken over the world.

“Artificial intelligence (and I don’t mean artificial intelligence in the narrow sense) is advancing at an unbelievable rate. Unless you have direct access to an organization like DeepMind, you have no idea how fast it is growing— It's growing at a near exponential rate. In five years (up to ten years) there will be a risk of a dangerous event."

—Elon Musk

Types of artificial intelligence

When someone asks you to explain different types of AI systems, you have to categorize them according to their function. Arend Hintze of Michigan State University classifies artificial intelligence into the following types: single-response, limited memory, mental, and self-aware. 

Reactive Machine AI

This type of artificial intelligence is the most basic artificial intelligence system, a machine that operates only on current data and only considers the current situation. That is, a reactive machine can only react, it has neither memory nor past experience to make present decisions. 

A classic example of a reactive machine is the famous IBM chess program Deep Blue. In May 1997, it beat world champion Garry Kasparov 3.5:2.5, becoming the first computer system to beat a world chess champion within the standard time limit.

Deep Blue seems to have human intelligence, but in reality, it only focuses on the status quo on the chessboard and makes decisions among possible moves. Unlike experienced humans, it has no concept of the past, other than abiding by the rule that repetitions cannot occur three times in a row.

Current artificial intelligence either has no concept of the world, or only has an extremely limited and specific concept of the specific tasks it performs. The innovation of Deep Blue is that it abandons the idea of ​​broadening the range of possible moves that a computer might consider. Instead, the developer found a way to narrow its "perspective", evaluate the outcome of certain moves based on it, and stop pursuing some possible moves. Similarly, AlphaGo can't evaluate all moves, but it takes a more sophisticated approach than Deep Blue: using neural networks to evaluate game moves. 

This type of AI cannot function outside the professional field and is easily fooled. They cannot become part of the world through interaction. Obviously, they are only the most basic artificial intelligence systems, and they do not meet our future expectations of artificial intelligence systems, that is, they can actually interact with humans and even react to the surrounding environment.

Limited Memory AI

Limited memory, as the name suggests, refers to AI that can make informed decisions by studying past data. This type of AI has ephemeral or temporary memory that can be used to store past experiences and evaluate future behavior.

Self-driving cars are this type of artificial intelligence, using recently collected data to make instant decisions. For example, using sensors to identify pedestrians crossing the road, steep roads, and traffic signals to make better driving decisions can help prevent traffic accidents.

However, the historical information of self-driving cars is too short-lived to store it in an "experience bank" like an experienced human driver. This type of AI doesn't build comprehensive "representations." It doesn't remember its own experiences and learn how to respond to new situations.

Theory Of Mind AI

Theory of mind, psychological term, is the ability of humans to understand the mental states of themselves and those around them. This theory originated from philosophy. After entering the field of psychology, it gradually became one of the research focuses of cognitive psychology and neuropsychology. Theory of mind is the key to the formation of human society, through this theory, people can better understand the inner dynamics of social interaction and interaction. Just imagine, if humans cannot understand each other's motives and intentions, then communication and cooperation with each other will become extremely difficult or even impossible.

This type of artificial intelligence is more advanced than the first two, and it plays an important role in psychology, mainly in the development of the "emotional intelligence" of machines.

If AI systems can really walk side by side with humans, they must be able to understand that everyone has thoughts and feelings, understand human expectations, and adjust their behavior accordingly. This is our current AI and future AI. important difference between.

Self-aware AI

This is the final step in the development of artificial intelligence: building a system that can form a self-representation. Geniuses like Elon Musk and Stephen Hawkings have been warning us about the evolution of artificial intelligence. Let's pray that we never reach this state of artificial intelligence. Because, in this state, the machine will have its own consciousness, it can be aware of itself, know its own internal state, and can also predict the feelings of others. Currently, self-aware artificial intelligence is still far away from us. However, in the future, it is still possible to reach the stage of super artificial intelligence.

Artificial Intelligence Research

Artificial intelligence can solve real-world problems through the following technologies:

• Machine Learning
• Deep Learning
• Natural Language Processing
• Robot
• Fuzzy Logic
• Expert System

Machine Learning

Machine learning is the science of using machines to interpret, process, and analyze data to solve real-world problems. Its roots can be traced back to a checkers program designed in 1952 by IBM's Arthur Samuel (known as the "father of machine learning"). There are three types of machine learning, namely supervised learning, unsupervised learning and reinforcement learning.

There is no doubt that machine learning helps humans overcome the bottlenecks in knowledge and common sense that we believe will hinder the development of human-level artificial intelligence, so many people see machine learning as the dream of artificial intelligence.

Deep Learning

Deep learning is the process of implementing neural networks on high-dimensional data to gain insights and form solutions. Deep learning is an advanced field of machine learning that can be used to solve more advanced problems, and it is the logic behind face authentication algorithms for virtual assistants like Facebook, self-driving cars, Siri, Alexa, etc.

Natural Language Processing

Natural language processing is the science of extracting insights from human natural language in order to communicate with machines and grow businesses. It is also one of the oldest, most researched, and most demanding fields in artificial intelligence. Any attempt to develop intelligent systems ultimately seems to have to address the question of what form of standards to use to communicate. For example, verbal communication is often preferred over communication using graphical or data-based systems.

The foundations of natural language understanding were established in the 1940s and 1950s using finite automata, formal grammars, and probability. However, in the 1950s and 1960s, early attempts to use machine translation of languages ​​proved fruitless. In the 1970s, the current trend was towards symbolic and random methods. After entering the 21st century, with the rise of machine learning, natural language processing ushered in new breakthroughs, and promoted the application of existing methods such as stochastic processes, machine learning, information extraction and question answering. For example, Twitter uses natural language processing technology to filter the language of terrorism in tweets, and Amazon uses natural language processing to understand customer reviews and improve the user experience.

Robot

AI bots are artificially intelligent entities that act in real-world environments, producing results by acting responsibly. This field is closely related to artificial intelligence in computational geometry and vision. Currently, in robotics, especially in embedded systems, we can see many manifestations of artificial intelligence, including search algorithms, logic, expert systems, fuzzy logic, machine learning, neural networks, genetic algorithms, planning, and even games. The humanoid robot Sophia is a good example of artificial intelligence in robotics.

Fuzzy Logic

Fuzzy logic is a calculation method based on the principle of "reality", rather than modern computer logic in the usual sense, namely Boolean logic. In other words, the results we get are often not black and white, positive or negative, but results "to a certain extent." For example, a robot may encounter obstacles on its way to a goal, but the robot must persist in achieving the goal. In other words, the robot's world is not only discrete, it also depends on certain "degrees of freedom", certain properties with varying degrees of variation, rather than just producing "on" or "off", "yes" or "no" "the result of.

Fuzzy theory was proposed in 1965 by Lotfi Zadeh (1921~), Zadeh did not initially imagine that fuzzy logic could be used in industrial processes for engineers to control and "smart" consumer products. Later, Mark Hopkins discovered the application of fuzzy logic in various fields, including economics, agriculture, aerospace, nuclear science, biomedicine and so on. In fact, fuzzy logic has already achieved a wide range of applications.

Expert System

An expert system is an intelligent computer program system that contains a large amount of knowledge and experience at the level of experts in a certain field, and can use the knowledge and problem-solving methods of human experts to deal with problems in the field. That is to say, an expert system is a program system with a large amount of specialized knowledge and experience. It applies artificial intelligence technology and computer technology to reason and judge according to the knowledge and experience provided by one or more experts in a certain field, simulating human experts. In order to solve complex problems that require human experts to deal with, an expert system is a computer program system that simulates human experts to solve domain problems. For more than 20 years, the research of knowledge engineering, the theory and technology of expert systems have been continuously developed, and their applications have penetrated into almost every field, including chemistry, mathematics, physics, biology, medicine, agriculture, meteorology, geological exploration, military, engineering technology, law, In many fields such as business, space technology, automatic control, computer design and manufacturing, thousands of expert systems have been developed, many of which have reached or even exceeded the level of human experts in the same field in function, and have been produced in practical applications. huge economic benefits.

Expert systems use if-then logical notation to solve complex problems. They do not rely on traditional procedural programming. The expert system usually consists of six parts: human-computer interaction interface, knowledge base, inference engine, interpreter, comprehensive database, and knowledge acquisition. Among them, the knowledge base and the reasoning engine are separated from each other, which is unique. The architecture of the expert system varies with the type, function and scale of the expert system. The basic workflow of the expert system is that the user answers the system's questions through the human-machine interface, and the inference engine matches the information input by the user with the conditions of each rule in the knowledge base, and stores the conclusions of the matched rules in the comprehensive database. Finally, the expert system will draw the final conclusion and present it to the user.