Summary: Explains 5 questions that need to be answered in formulating artificial intelligence ethics

This article is based on the content of a speech given by Liu Feng, the author of Internet Evolution and PhD in Computer Science, at the "2017 Artificial Intelligence: Technology, Ethics and Law Seminar" jointly organized by the Institute of Science and Technology Strategy Consulting of the Chinese Academy of Sciences and Tencent Research Institute in Beijing on April 13. The title of the speech is "The relationship between the formulation of ethical rules for artificial intelligence and the direction of biological evolution"

This article elaborates on five questions that need to be answered in formulating artificial intelligence ethics, and proposes that whether a consensus can be reached on "whether biological evolution has a direction" will be the key to whether the ethics of artificial intelligence (AI) can form a norm in the future. The research report compiled based on the speech at the "2017 Artificial Intelligence: Technology, Ethics and Law Symposium" is as follows:

In layman’s terms: ethics refers to “the relationship between people and between people and nature, and the rules for dealing with these relationships.” In the thousands of years of human civilization, we still do not have a unified, standard, and clear ethical system, only some general principles recognized by most people. The ethical impact brought by artificial intelligence has become more prominent due to imperfections and controversies in ethical issues.

The Titanic (RMS, also known as the Titanic) was an Olympic-class cruise ship that sank after hitting an iceberg on its maiden voyage in April 1912. The Titanic disaster was one of the deadliest peacetime maritime disasters. More than 1,500 people on board died. Which group gets priority in life during the escape process. It has become an important issue faced during the escape. David Popnor, a professor at the State University of New Jersey and a famous sociologist, wrote in his book "Sociology": "...Unfortunately, there are not enough lifeboats. Although many people (more than 1,500) were killed, the passengers Paying attention to the social norms of 'rescuing women and children first' allowed the British public and government to "find some comfort" in the face of this huge disaster - statistics show that "69% of the women and children among the passengers survived, while only 17% of the male passengers survived."

"2012" is a disaster movie about global destruction. It was released in the United States on November 13, 2009. The movie tells the story of the protagonist and the experience of people from all over the world struggling to survive. In the face of disaster, all kinds of human conditions are revealed. In the movie, when Air Force One could only accommodate one person aboard, the President of the United States allowed the physical scientist to leave safely. When he stayed, he said, "One scientist is more important than dozens of officials."

The train is moving rapidly and cannot stop suddenly. There is a bifurcated track directly ahead, the left one is still in use and the right one has been deactivated (but it is still safe for trains to run on it). There are five children playing on the railway tracks that are still in use, and one child playing on the railway tracks that are out of service. Then the train came. According to the original route of the train, it would go to the left and hit five children. If you drive to the right, you will only hit one child. If you were a switchman, would you turn the train to the right?

A Boston Dynamics robot was attacked by a test scientist when it was moving a box, causing it to become unstable and fall. Internet viewers of the video protested, believing that the robot's rights were violated.

"Robots must not harm humans; robots must obey human orders, unless this order conflicts with the first principle; robots must protect themselves, unless this protection conflicts with the above two principles." The above are the "Three Principles of Robots" proposed by Asimov, the famous science fiction writer and "father of robotics". They have been quoted in countless science fiction novels and film and television works in later generations, and are regarded as classics.

In Asimov's eyes, robots are just cold machines, tools that serve human beings. Since it is a tool, it naturally does not need any rights. Therefore, when "protecting oneself" conflicts with "not harming humans" and "must obey human orders", the robot's most basic right to survival can be sacrificed without hesitation.

However, in recent years, in the context of the rapid development of robot technology and the increasingly close relationship with humans, the discussion on the rights of robots has received more and more attention, and scholars and the public have participated in it.

5 unavoidable questions in formulating AI ethics

1. Who should formulate AI ethics?

2. Whose perspective should we take to formulate AI ethics?

3. On what principles should AI ethics be formulated?

4. Who is AI ethics for reference?

5. Can AI be regarded as a living entity with the same rights as humans?

In the mid-19th century, Darwin founded the scientific theory of biological evolution. Darwin's theory of evolution with natural selection as its core is one of the greatest theories in the history of human science. Unified the various disciplines of biology. The influence of Darwin's theory of evolution is not limited to the field of biology. It also provides a new world view, outlook on life, outlook on the universe and methodology, which has an impact on almost all fields of science and humanities. Ethical issues including artificial intelligence will ultimately be affected by Darwin's theory.

Darwin's theory of evolution rejected teleology (which was praised by Meyer as one of his great contributions). Darwin advocated that variation is random, using the word "nature" which is no less than God to explain the core of his carefully constructed evolutionary theory. However, Darwin also had his confusion and helplessness. He lamented that “it is very difficult or even impossible to understand that this vast and wonderful universe... is actually the product of blind chance or necessity” (2-Xie Ping. 2014. The Origin of Life—The Sublation and Innovation of Evolutionary Theory.)

Darwin's theory of evolution believes that biological evolution is not an evolution from lower to higher levels, and that humans are not more advanced than other organisms. Darwin imagined the process of biological evolution as a large tree that is constantly growing and branching. All existing organisms are located at the top of a small branch of this tree. It is difficult to say which one is more advanced. It is meaningless to compare high and low levels among biological species that exist at the same time. Evolution has no predetermined direction. There is no trunk with humans at the top of the evolutionary tree. Humans are just an ordinary branch on the evolutionary tree. From the perspective of Darwin's theory of evolution, humans are not more advanced than mice and ants, and humans do not represent the development direction of living things.

Will Darwin's theory of evolution's view that life has no direction have any new progress and breakthroughs with new technological advances? We will discuss this issue below from the research progress of brain science, Internet and artificial intelligence. It is proposed that the progress of brain science, the Internet and artificial intelligence will jointly show the direction and goal of life evolution from different angles.

Although the manifestations of living things vary widely, the brain, the core of living things, does show obvious directionality. From single cells to humans, the brain is becoming more and more complex and intelligent. After reaching the highest point of humankind, new forms of biological evolution emerged, and people united together through the Internet to evolve together.

Nobel Prize winner Australian scientist John.C. mentioned in his book "The Evolution of the Brain" that "the brain of living things evolved from the brain of fish to the brain of reptiles, then to the brain of mammals, and finally to the human brain. If we dissect the human brain, we can clearly see that the fish-like, reptile-like, and mammal-like structures are distinct in the human brain."

Conclusion of the Internet Evolution Theory: “The Internet will evolve in a direction that is highly similar to the human brain. It will have its own visual, auditory, tactile, and motor nervous systems, as well as its own memory nervous system, central nervous system, and autonomic nervous system.

On the other hand, the human brain has evolved all the Internet functions at least tens of thousands of years ago, and the evolving Internet will help neuroscientists uncover the secrets of the brain. Scientific experiments will prove that the brain also has the same search engine, the same SNS system, the same address encoding system of IPv4, and the same routing system of Cisco. . . "

The Internet of Things is the sprout of the sensory nervous system of the Internet brain, and cloud computing is the sprout of the central nervous system of the Internet brain. Industry 4.0, industrial Internet, drones, intelligent driving, and 3D are essentially the development and sprout of the Internet's motor nervous system. Big data is the basis of information in the Internet brain, and AI is the basis for the generation of Internet intelligence, wisdom, and consciousness.

When organisms have evolved to the level of humans, humans will unite together and evolve together through the Internet. And the result of this co-evolution is. The Internet, which connects humans, is structurally highly similar to the brain step by step. It continues to spread in space with the expansion of humans. If there is enough time, it can be foreseen that after enough time, space, the brain, and the Internet will merge into one and evolve into a mature Internet brain-like architecture.

The conclusion of Darwin's theory of evolution is deduced forward, that is, living things have a common ancestor, and through natural selection, the species human appeared. The conclusion of the Internet evolution theory is deduced backwards, believing that the Internet is evolving towards the brain-like architecture of the Internet through the technologies and tools created by humans. The element "people" connects the evolution of the Internet with the evolution of biology.

Determination criteria for standard intelligent systems: Any system (including artificial intelligence systems, human beings and other living systems) can be considered a standard intelligent system if it meets the following characteristics.

Feature 1: The ability to obtain data, information and knowledge from the outside world through sounds, images, text, etc. (including but not limited to these three methods).

Feature 2: Able to transform data, information and knowledge obtained from the outside world into knowledge mastered by the system.

Characteristic 3: The ability to innovate by applying the knowledge mastered according to the needs generated by external data, information and knowledge. These abilities include but are not limited to association, creation, guessing, discovering patterns, etc. The results of the use of this ability can form new knowledge mastered by oneself.

Feature 4: Able to feed back the data, information and knowledge generated by the system to the outside world or transform the outside world through sound, image, text, etc. (including but not limited to these three methods).

The establishment of the standard intelligence model refers to the von Neumann architecture. The von Neumann architecture consists of five parts: calculator, logic control device, memory, input system and output system. By comparing the differences between the standard intelligence model and the von Neumann architecture, it can be found that the von Neumann architecture can supplement two parts. Through this addition, we can represent people, machines and artificial intelligence systems in a clearer way.

The first supplement is the innovation and creation function, which is the ability to discover new knowledge elements and new rules based on existing knowledge, enter them into the memory for use by computers and controllers, and interact with the outside world through input/output systems. The second complement is an external knowledge base or cloud storage that enables knowledge sharing, while the external storage of the von Neumann architecture only serves a single system. Therefore, extending the von Neumann architecture can form a new architecture

Direction of Evolution - Hierarchy of Intelligent Systems

The basic characteristics of level 0 intelligent systems exist in theory, but such artificial intelligence systems do not exist in reality.

The basic characteristics of the Level 1 system of intelligent systems are systems and objects that cannot interact with human testers.

Level 2 systems of intelligent systems, whose basic characteristics are the ability to interact with human testers and the presence of controllers and memory.

The basic characteristic of the Level 3 intelligent system is that in addition to the characteristics of the Level 2 system, the programs or data contained in its controller and memory can be upgraded or added without being connected to the Internet.

In addition to the characteristics of level 3 systems, the basic characteristics of level 4 systems of intelligent systems are that they can share information and knowledge with other intelligent systems through the network.

The most basic characteristics of the fifth-level system of intelligent systems are the ability to innovate and create, identify and identify the value of innovation and creation to the intelligent system, and apply the results of innovation and creation in the development process of the intelligent system.

The most basic characteristic of the sixth-level system of intelligent systems is that as time advances and approaches infinity, the input and output capabilities, knowledge mastering and application capabilities of intelligent systems that continuously innovate and create new knowledge will also approach infinity.

Inspired conclusions drawn from brain science, the evolution of the Internet and the hierarchical classification of intelligent systems

The evolution of the brain, the evolution of the Internet, and the classification of intelligence levels in intelligent systems all show obvious directionality. For these three fields, what they have in common is that they carry the knowledge and wisdom of life and are constantly evolving and improving.

The improvement and evolution of knowledge and wisdom are the core of biological evolution. From the perspective of knowledge and wisdom, we can judge the direction of biological evolution and the level of biological evolution.

The expansion rate of the knowledge base of biological populations is the focus of biological evolution. The knowledge base of other organisms has stagnated and reached a dead end. Therefore, there has been no further change for tens of millions of years, and it has become increasingly lower in the earth's life cycle.

In the past 10,000 years, human beings have continued to expand and accelerate in knowledge and wisdom, and have made further huge leaps due to the invention of the Internet and artificial intelligence, thus gaining dominance over the natural competition on the earth.

The direction and goal of biological evolution

The direction of biological evolution is to continuously expand the knowledge base of the population and the ability to transform nature. The goal is to achieve omniscience and omnipotence over the entire nature (life circle, universe). The level of creatures is divided according to their knowledge base capacity and ability to transform the world.

The impact of evolutionary direction on AI ethical issues and answers

1. Who should formulate AI ethics?

Formulated by the biological species with the most knowledge and wisdom

2. Whose perspective should we take to formulate AI ethics?

Standing in the position of the biological population with the richest knowledge and the most powerful wisdom

3. On what principles should AI ethics be formulated?

Formulate AI ethics based on principles that are beneficial to expanding the knowledge and wisdom of the population where the rules are set

4. Who should AI ethics be used for reference first?

For reference by managers (governments, entrepreneurs), researchers (scientists), developers (programmers and product managers) of the population (life that makes ethical rules)

5. Can AI be regarded as a living entity with the same rights as humans?

AI cannot yet be regarded as a living entity with the same rights as humans. It shares part of human knowledge and intelligent functions, but it cannot replace the most important creativity and creative evaluation value. More importantly, AI cannot determine its own evolutionary direction and goals, and it has no motivation to evolve correctly. Its evolutionary motivation comes from humans and it is still a tool of humans.

The Titanic allowed women and children to leave because they would be more helpful to the future of mankind and could better expand and inherit human knowledge and wisdom. In the 2012 movie, the president chooses to die, leaving young geophysical scientists to live. Because scientists have more knowledge and wisdom to contribute to solving human crises. In the switchman's dilemma, it is necessary to analyze in detail which side may contribute more to the future knowledge and wisdom of mankind.

However, there still needs to be a balance between the interests of the population and personal interests. How the ethical setting strikes a balance between populations and individuals is also related to the urgency of the scene. For example, in an extreme environment like human extinction in the 2012 movie, the distribution of group interests and individual interests should be different from ordinary scenes where the degree of crisis is not strong.

It is necessary to form a consensus on the direction and goals of biological evolution, and then formulate ethical rules for artificial intelligence (AI).

If there is no consensus on whether biological evolution has a direction, there will be many difficulties in formulating ethical rules for artificial intelligence (AI). For example, if humans and animals are on the same level, how to deal with the relationship between humans and animals when formulating rules? If there is no evolutionary direction, there will also be problems about who should be protected first in formulating ethical rules for artificial intelligence (AI). Without determining the evolutionary dynamics and goals, the relationship between artificial intelligence systems and humans cannot be handled.

At the same time, the nature of emotion and consciousness is also related to the direction and goal of evolution. For example, happiness, excitement, and happiness are often related to the reactions of organisms that comply with the evolutionary trend. Sadness, anger, and disappointment are also often related to reactions that go against the evolutionary trend. If the evolutionary motivation and evolutionary goals are not determined, it is impossible to judge whether the emotions of the artificial intelligence system are real emotions.

Important issues that need to be discussed in depth

The driving forces behind the evolution of AI are scientists, entrepreneurs, programmers, and product managers.

What is the driving force behind the direction and goal of evolution in humans and organisms?

Can the evolutionary power of biology be applied to the development of AI? If successful, the AI ​​threat theory will have a basis for realization.

The setting of AI ethics will undergo major changes. If theoretically proven to be impossible, AI will never be able to fully surpass humans.

Appendix 1: Definitions of knowledge and wisdom

Data is an abstract representation of the quantity, attributes, location and interrelationships of objective things using conventional keywords, so that it is suitable for preservation, transmission and processing in artificial or natural ways in this field.

Information is a data flow that is timely, has a certain meaning, is logical, processed, and valuable for decision-making.

Knowledge is valuable information that has been evaluated and reviewed, and then deposited into the human knowledge base and structured.

Wisdom is the ability of human beings to analyze, compare, and deduce solutions to problems arising from the movement of the material world based on existing knowledge and based on the information obtained. The result of the application of this ability is to unearth valuable parts of information and make them part of the existing knowledge architecture.

Appendix 2: Classification of intelligent systems

For level 0 systems of artificial intelligence systems, their basic characteristics exist in theory, but such artificial intelligence systems do not exist in reality. In the hierarchical rules extended by the extended von Neumann architecture, some combinations can be made, for example, information can be input, but information cannot be output; or information can be output, but information cannot be input; or innovation and creation can be made, but the knowledge base cannot grow. For these cases where the corresponding system examples cannot or cannot be found in reality, we will uniformly classify them as "level 0 systems of artificial intelligence systems", which can also be called "special types of artificial intelligence systems".

For level 1 systems of artificial intelligence systems, their basic characteristic is that they cannot interact with human testers. For example, there is a thought called animism, which believes that everything in the world has a soul or natural spirit. A tree or a stone has the same value and rights as a human being. Of course, from a scientific perspective, this view can only be regarded as conjecture or philosophical thinking. From the perspective of the grading rules of "whether it can interact with testers (humans)", because objects such as stones cannot interact with humans, they may have an internal knowledge base that can innovate knowledge, or can interact with other stones, but to human testers it is a black box and cannot be understood. Therefore, objects and systems that cannot interact with testers (humans) can be defined as "Level 1 systems of artificial intelligence systems". Examples that meet Level 1 classification include stones, sticks, iron blocks, water drops, etc. Objects or systems that cannot interact with humans.

For the level 2 system of artificial intelligence system, its basic feature is the ability to interact with human testers, the existence of a controller and memory, that is, the system described by the von Neumann architecture. Therefore, many household appliances are called smart appliances, such as smart refrigerators, smart TVs, smart microwave ovens, and smart sweepers. Most of these systems have a characteristic, that is, although they have more or less control program information inside them, once they leave the factory, their control programs cannot be updated, they cannot be upgraded, and they cannot automatically learn or generate new knowledge. For example, in a smart washing machine, whatever button people press will activate the function of the washing machine. Its functionality remains unchanged from the time of purchase to the time of damage (except for malfunctions). This kind of system can interact with human testers and users, conforms to the characteristics described by the von Neumann architecture, and its control program or knowledge base will not change from the time of its birth. This kind of system can be defined as a "second-level system of artificial intelligence systems". Examples include daily sweeping robots, old-fashioned household refrigerators, air conditioners, washing machines, etc.

For the level 3 artificial intelligence system, its basic feature is that in addition to the characteristics of the level 2 system, the programs or data contained in its controller and memory can be upgraded or added without being connected to the Internet. For example, home computers and mobile phones are commonly used smart devices, and their operating systems can often be upgraded regularly. For example, the operating system of a computer can be upgraded from .0 to .0, and the operating system of a mobile phone can be upgraded from .0 to .0. The internal applications of these devices can also be continuously updated and upgraded according to different needs. In this way, the functions of home computers, mobile phones and other devices will become more and more powerful, and more and more scenarios can be handled. In addition to home computers, many household appliances and robots have also begun to have interfaces, and system upgrades can be performed through external devices. This type of system is significantly more adaptable than Level 2 intelligent systems. This kind of system can interact with human testers and users, but cannot interact with other systems through the "cloud". Its control program or knowledge base can only accept USB, CD-ROM and other external devices for program or information upgrade. It can be defined as a "Level 3 system of artificial intelligence systems". Examples include smartphones, home computers, stand-alone office software, etc.

For the level 4 system of artificial intelligence system, in addition to the characteristics of the level 3 system, its basic characteristics are that the most important thing is that it can share information and knowledge with other intelligent systems through the network. In 2011, the European Union funded a project called , which aims to allow robots to share knowledge over the Internet. Helping robots learn from each other and share knowledge will not only reduce costs, but also help robots improve their self-learning and adaptability, promoting their faster and larger popularization. These capabilities of cloud robots improve their adaptability to complex environments. In addition to the functions of Level 3 systems, this type of system also has an important additional function, that is, information can be shared through the cloud. Therefore, this type of system can interact with human testers and users, and can conduct information interaction through the "cloud" to upgrade programs or information. However, all the information of this type of system is obtained directly from the outside, and it is unable to independently, innovatively and creatively generate new knowledge internally. This kind of system can be defined as "the fourth level system of artificial intelligence system". Examples include Google Brain, Baidu Brain, cloud robots, B/S (/, browser/server) architecture websites, etc.

For level 5 systems of artificial intelligence systems, the most basic characteristics are the ability to innovate and create, identify and identify the value of innovation and creation to human beings, and apply the results of innovation and creation in the development process of mankind. When we expanded the von Neumann architecture, we added innovative knowledge modules to the original von Neumann architecture, which is an attempt to incorporate humans into the expanded concept of artificial intelligence systems. Human beings can be regarded as special "artificial intelligence systems" built by nature. Different from the first four levels, the biggest characteristic of human beings and other living creatures is that they can continuously innovate and create, such as discovering gravity and the periodic table of elements, writing new novels, creating new music, paintings, etc., and then disseminating and sharing them through articles, letters, telegrams, and even the Internet. Continuously innovating and creating, and being able to identify the usefulness of innovation and creation to ourselves, allows humans to occupy the intellectual commanding heights in the earth's ecological environment. Therefore, this kind of system can interact with human testers and users, can innovate and create new knowledge, and can interact with information through articles, letters, telegrams, and even "clouds" such as the Internet. This kind of system can be defined as a "Level 5 system of artificial intelligence systems." Humans are the most prominent example of Level 5 AI systems.

For the sixth-level artificial intelligence system, the most basic feature is that as time advances and approaches infinity, the input and output capabilities of the intelligent system that generates new knowledge will be continuously innovated and created, and its ability to master and apply knowledge will also approach infinity. According to the Christian definition of God as "omniscience and omnipotence", it can be seen that intelligent systems are constantly innovating With the creation and continuous accumulation of knowledge, intelligent systems represented by humans will eventually achieve the state of "omniscience and omnipotence" in enough time. From this perspective, whether it is the concept of "god" in Eastern culture or "God" in Western culture, from the perspective of intelligent system development, it can be seen as the evolutionary state of intelligent systems (including humans) at future points in time.