智能工程
上QQ阅读APP看本书,新人免费读10天
设备和账号都新为新人

Preface

After two years, I finally completed writing this book. This book was not in my original writing plan in 2015. My thinking at that time was: My research goal would be achieved after I have explained the attributes and development patterns of information and intelligence that are actually in existence, as well as the goals, requirements and logical frameworks for non-biological intelligence that surpass human intelligence. Therefore, I replaced the reference materials for the two books The Nature of Information and Principles of Intelligence on my desk with that about economics in the beginning of 2018. After a few months, I received some feedbacks from readers of these two books and subsequently felt the need to write another book on how to develop and attain non-biological intelligence. If we try to design non-biological intelligent systems whose functionality may be comparable or even superior to AGI (Artificial General Intelligence) based on the understanding, logic and theoretical frameworks depicted in The Nature of Information and Principles of Intelligence without adopting the popular methods and algorithms in the artificial intelligence field, it might be difficult to make non-biological intelligence a reality. Therefore, I changed my original writing plan and started working on the book of Intelligence Engineering. This book, together with the other two already published, would form a"Trilogy of Intelligence".

The non-biological intelligence discussed in this book is essentially machine intelligence. Their focus is modeled after human intelligence, built on existing machine intelligence, based on meaning computation, and with understanding as the premise. Such non-biological intelligence can accumulate knowledge and experience, or is capable to understand, think, judge, decide and act. This kind of machinery can cooperate with people, integrate existing machine intelligence, and develop in a mode of sustainable growth.

I had underestimated the difficulties of writing this book in the beginning, thinking that I only needed to reify the problems that have been clarified in theory and logic in the previous two books. As my research and writing went on, I found this judgment was wrong. The difficulties center around three points: First, what kind of system framework can realize all the functions and requirements of the non-biological intelligence defined in Principles of Intelligence. Second, how the transformation to an intelligence processing model of meaning-understanding-thinking-action can be made with the software and hardware developed using the traditional IT model. Third, at which level of granularity the achievability of non-biological intelligence should be described.

For the first question raised above, Figure 1.5 of this book provides a general outline of the answer. The framework of non-biological intelligence must support all its activities in the process: cognition, tasks, living and control. The same framework must meet the eight principles and requirements of non-biological intelligence: autonomy, diversity, development, livability, interaction, structurality, inheritability, and entirety. One possible framework that meets these requirements consists of the 11 function systems presented in Figure 1.5. All its components, from basic microprocessors to functions aggregated at all levels above, both have autonomous independence and are under the effective management and control of the corresponding function assemblage and the entire entity. The 11 function systems are: perception, description, connection, memory, learning, processing, tasks, resources, living, interaction and control.

The two requirements of autonomy and livability are mainly realized through four function modules: resource module, living module, thinking module and control module. Resources in conjunct with the livability function ensure the normal operation of intelligent systems. Awareness and thinking functions analyze and differentiate the risks intelligent systems face, controlling decision-making and supervising emergency responses. Autonomous independence is assigned to all components of the intelligent system in order to avoid systemic risks and improve livability. The inheritability requirement is mainly reflected in the first three phases of the life cycle of intelligent systems: initiation, assignment, and cultivation. It continues thereafter in the learning and interaction processes. The development requirement is satisfied through the learning module. Learning is the main source of development while other function modules evolve via summarization in their own conducts or enhance their own functionalities via learning. The interaction requirement is primarily satisfied through the interaction function module, meeting all intelligent system perception, learning, task, resource and living functions'needs for interaction. The two contradictory yet consistent requirements of structurality and entity are essentially met through the overall coordination among the internal autonomous structures and collaborative modules of the 11 function systems.

The four intelligent processes of cognition, tasks, living and control cover all the key elements, functions and conducts of intelligent systems. The completion of each process requires the support of 11 function systems, and each plays a role in the entire process of the intelligent system life cycle. This book lays out the structure, functions and interrelationships of the 11 function systems based on the needs of these four processes. This book also makes four special plans for discussing the functions and completion of these four processes, which constitute the key to reading and applying the book.

First, the cognition process is for the purpose of understanding, centering around the autonomous development of intelligent systems, and via the path of gradual overlays. All the functions of intelligent systems, including cognition development, are based on cognition. "It is wise and honest to hold what we know and admit what we don't know. " This famous quote from Confucius 2500 years ago is the philosophical foundation for this book. What this book follows in practice is a saying of another Chinese sage-Xun Zi: "Without making single steps, how could it be a thousand miles. " Non-biological intelligence is to be realized by using the Internet as its platform and interaction function as its medium to apply what is achieved in human intelligence in its development, word by word, action by action, scenario by scenario, and subject by subject. The development of non-biological intelligence will also employ large-scale autonomous parallel computation, overlaid onto the memory of the intelligent system to form an intelligible and usable memory that far exceeds that of one human being, gets verified, and reaches perfection via interaction and usage. This is the principle for designing the function systems of non-biological intelligence relating to cognition in the present book.

Second, the task process is based on the accumulation of cognition. Thus, it only undertakes what it can accomplish, and does not take what it cannot do. The core of intelligent systems'cognitive process is to accumulate common sense and basic knowledge. When reaching a certain level, it will continue to further explore the types of tasks it can undertake till achieving specialization in those areas. The so-called specialization implies that the intelligent system can solve all possible problems in the domain after experiencing all the possible scenarios of such problems, undergoing all possible processes of solution-seeking, and becoming certain of such solution seeking processes. Even if there exists some uncertainty, it will not affect the reliability of such problem solutions. In other words, cognition that centers around tasks aims to exhaust each problem spaces and solution spaces of particular tasks. It is certain that the task process will only assume tasks with known solution paths. We know that it is impossible to exhaust each problem spaces and solution spaces of particular tasks. But we can draw such a conclusion from the abstract or pure logic perspective. As time passes, the problem spaces and solution spaces for most or the majority of jobs in the economic society can be exhaustively explored. One person is able to know several hundred or over one thousand people. That person can work skillfully in several dozen or more than one hundred different scenarios. An assembly line worker, a salesperson, a financial planner, or a manager can only handle a limited number of types of problems and their variations. As for an intelligent system, it starts with a finite set of problems. However, the intelligent system can exhaustively explore more and more types of problems with the support of its immense, tireless, rapidly iterative, mutually independent yet collaborative cognition function. It will surpass ordinary workers at the certain stage of its development, go on to surpass a group of people, or even outperform most people. Ultimately it will possess new characteristics or reach new heights that no one has done before.

Third, the living process is established when an intelligent system becomes independent and autonomous in society and can assume social responsibilities. It accumulates knowledge, experience, skills, facts, and data uninterruptedly and autonomously. It must fully take control of its own livability at a certain point of its life cycle.

Finally, the control process is the key to the formation, development and role-playing of all processes of non-biological intelligence. An intelligent system is composed of billions and trillions of autonomous components. While these autonomous components are different in type, function, and level, they have their own and mutual control as well as coordination functions, which form the basis and important constituents of the control process. What is more important in controlling an intelligent system is to control it entirely. This is the responsibility of the control function system. There are two significant modules in the control function. One is awareness and thinking. It continuously monitors operations and environments of the intelligent system, analyzes and determines if there exists any risk or matter which is overlooked locally by the control function system. The other significant module in the control function is overall decision-making, control, coordination and management in intelligent systems.

This book now explains the second difficulty in realizing non-biological intelligence with related functions and rules of the six function systems: How the transformation to an intelligence processing model of meaning-understanding-thinking-decision making-control can be made with the software and hardware developed using the traditional IT model.

In fact, data analysis systems and business intelligence systems, through rigorous data control, data formats, data dictionaries, and designated processing functions, can derive expected and meaningful results from numbers (or symbols) and offer explanations to the results based on data structures, functions and data dictionaries in traditional database management systems. This demonstrates that meaningful results can be derived from symbols or the system can give its"understanding", as it can be explained, of specific fields and processing results in the database in this context. The limitation of this practice is that meaningful results cannot be generalized or accumulated cross-system beyond the scope and depth of the system. This example also sends us an important message: We can derive meaning from symbols using rigorous structures that are corresponding to their intrinsic connotation along with processing and explanatory rules. Although some other processing modules can also derive meaningful results from symbols being processed and in which whose meanings are implicitly imbedded, they cannot be considered intelligence if not explainable. The principal prerequisite of intelligence is that it can be explained.

Based on such analyses, let's assume that we carry out the following steps. Transform obtained information to perceived objects (symbols) in a structural and explainable way;Convert these symbols with the uniform, understandable symbols and labeling schema of intelligent systems;Maintain the entire structure of perceived contents and objects determined by the perceptor, static spatial relationships of all kinds, and dynamic rules plus spatial relationships for all objects that are detailed in the time series;Transform further the previously perceived objects of the same type and previously described symbols of the same kind;Represent all the relationships perceived this time via connection;Compare the basic units of the perceived objects with existing memory and express explicitly all the relationships. The perceived objects, saved through the method described above, have been transformed to the understandable and usable memory of intelligent systems. This process has three key phases: perception, linkage and memory.

The key step is the control of perception. All external objects, documents, images, audio and video files must be perceived by one perceptor only. What is perceived by one perceptor is represented via the description function using symbols that can be understood by intelligent systems. Using only one perceptor is the approach to meeting this requirement, such as one perceptor for Chinese characters;one kind of sound wave features constitutes one perceptor while sound is a time series of sound wave segments;a combination of individual basic shapes and colors functions as the perceptor for images while videos are comprised of series of images. Billions or tens of billion of logic perception microprocessors and a smaller number of collaborative physical perception microprocessors are directly connected to the description processors for object identification. The key elements of the identified objects and their connections, after description, are transferred to memory. These key elements and connections, together with other possible related connections in the memory, become understandable and usable memory units for all function modules of intelligent systems. The task process, in a certain sense, is the reverse process of the cognition process. When tasks are perceived, they enter the designated area for tasks and match with proper execution microprocessors. Task execution is completed by calling up physical resources, memory units and determined processing microprocessors through these execution microprocessors. The living and control processes are also completed in a similar way. All functional processes or intelligent activities of intelligent systems, starting with perception and using special structures, complete the transformation from symbol to meaning, storage to memory and computation to understanding.

The last difficulty in having non-biological intelligence is essentially a technical and workload issue as the finer the granularity, the greater the workload, and the higher the forward-looking requirements in theory and practice. In the beginning of this book writing, I planned to use 28 chapters to complete it at a granular level similar to a conceptual design for non-biological intelligence. Later I found that the workload was too huge and it was also unnecessary because the purpose of this book is not for implementing specific projects of non-biological intelligence. Rather, I only seek an explanation of its achievability. I thus increase the granular level as much as possible till I can show the achievability of non-biological intelligence—the purpose of this book. In my roughest estimate, building a general non-biological intelligent system is a project that generally requires a funding of hundreds of billions of dollars or trillions of yuan, 6-8 years of time, a R&D team of several thousand members along with a large number of Internet-based volunteers in its peak development time. Without any precedent, discussions at a fine granular level would lack basis. Therefore, this book does not cover specifics and I have removed descriptions of logical processes in examples for realizing major functions of non-biological intelligence in the original writing plan. This change in my writing plan considers not only the book length issue and its consistency with the previous two books but also not limiting my thinking and imagination about realization models and variety of fine points for non-biological intelligence.

Some people may say that as the non-biological intelligent system is too large and complex, it will be suffocated according to the second law of thermodynamics. Other people may think that such a system will collapse due to the complexity certainly existing in its computation. These are actually thoughts from different perspectives. All the processes of non-biological intelligence process meanings while understanding forms the basis of meaning processing which overlays on understanding. These processes start with 1+1+…until individual logical objects, physical entities, functional units, and information records reach perfection;different-level aggregations composed of the afore-depicted individual objects reach perfection;and systems composed of afore-described individual aggregations reach perfection. At this point, 1+1 can be easily achieved at Level-M (millions), Level-G, or even Level-T. This+1, +1, …process can be done continuously for achieving the capability of understanding and performing external tasks. This capability will gradually surpass the ability of one person, several people, tens of thousands of people, and at some point, surpass the ability of all mankind.

The non-biological intelligent system is designed in such a way that all its components are in the mode of autonomous and independent operations, which not only matches the characteristics of intelligence, but also reduces the complexity of the system. Individual functional processes and intelligent conducts of non-biological intelligent systems, consisting of a gigantic number of microprocessors and intelligent units that can operate independently while following the overall coordination, perform all the functions of intelligent systems.

There are in total ten chapters in this book. Chapter 1 is about building non-biological intelligent systems. This chapter provides a general outline for the present book and serves a transitional role by recapitulating the major conclusions in The Nature of Information and Principles of Intelligence as well as introducing the logic, framework and thinking in Intelligence Engineering. This chapter, first of all, further elaborates on the what and why of non-biological intelligent systems, and summarizes the basic requirements for them based on Principles of Intelligence. Then, it puts forward the systemic framework for meeting the requirements and considerations for accomplishing the important steps outlined in this book. Do note that Section 1.1 to Section 1.5 of this chapter depict the requirements for non-biological intelligent systems by adhering to the logic and conceptual framework in The Nature of Information and Principles of Intelligence whereas Section 1.6 and Section 1.7 are written according to the logic and conceptual framework presented in this book. There exist conceptual inconsistencies between these two parts that are primarily reflected in the division of components at the lowest level. The smallest logical units of information, functions, and processing in the previous sections are merged to form a microprocessor in Section 1.7 and subsequent chapters. The reason for this alternation is because a microprocessor constitutes the smallest components in aggregation from the engineering point of view.

Chapter 2 is on perception. This chapter is the longest and with the finest granularity among all the chapters in that it covers the basic, core and important parts of non-biological intelligence ranging from traditional symbol processing to meaning processing, and from the Turing computation model to intelligence computation model transformation. Starting from the perception function of non-biological intelligence, this chapter introduces the perception function system based on perception microprocessors and the system of rules. It also discusses attainment methods for perceiving different types of objects and subsequent processing. The perception function has three specific designs. First, one physical or logical perception microprocessor only identifies a specific object. In addition, the specific perception microprocessor identifies not only the smallest, identifiable objects but also aggregated objects designated by memory units of intelligent systems. Second, the perception microprocessor, serving as an encyclopedia of the intelligent system, can connect with all related contents and complete the entire feedforward processing. Any newly entered object can be fully integrated into the existing memory. Third, the perception, description and memory of an object can become via connection or directly a logical microprocessor for function differentiation.

Chapter 3 discusses description. Perception is the first key step for realizing the transformation of symbols to meanings while description is for establishing a uniform function system that intelligent systems can understand and use. Regardless of which format and from which functional process, all information in the memory must be processed by the description microprocessor to ensure the consistency and meaning features of all the stored information of intelligent systems. This chapter specifies the definition, functions and realization mechanism for description. Description is based on a uniform symbol system and rules for representation. The "understanding" of the described object by the description microprocessor is based on onlyness. One description microprocessor only handles objects one perception microprocessor or other function systems need to describe and transmitted by one microprocessor. The one-to-one mechanism ensures description correctness which, together with the exhaustive characteristics of connection, will change from strict transformation to real understanding.

Chapter 4 covers connection, memory and understanding. This chapter introduces the connection and memory functions on the basis of perception and description, ending the understanding-based meaning processing and cognition processes of intelligent systems. Connection is a fundamental function of intelligent systems that links all components, functions, and activities: It is through connection that all the relationships among components are expressed, functions are represented and called, and the process of activities is also a connection process. Just like the brain, each neuron has an average of 1, 000 synapses for connecting with other neurons and for performing functions by representing information through the structures formed via connections. One descriptive memory unit of intelligent systems also contains a huge number of connections, many of which exceed the order of thousands and may reach the order of millions at maximum. Memory is the result of perception, description, and connection as well as forms the foundation for learning and growth of non-biological intelligence. Different types of memories, through given and developing rules, tirelessly pursue perfection and become triggers for learning. Understanding is the result of the cognition process. This chapter explains understanding, the core secret of human intelligence, and how intelligent systems attain it.

Chapter 5 focuses on learning and interaction. For the intelligent system, learning is a basic conduct that never stops throughout its life cycle. The trigger for learning comes from the needs of each function system while these needs come from two main sources. First, each function system pursues its own internal motivation for perfection. The rules for maintaining this motivation are implanted at the time of initiation and assignment, and stimulated during the time of cultivation. As long as the function system does not reach perfection, it will trigger the learning process. Second, if each intelligent conduct in its execution cannot ensure the reliability of its results within a specified range, learning is triggered. Even if the task execution is stopped, the corresponding learning will not cease until the perfection of problem spaces and solution spaces are reached. Interaction includes all connections, physical or logical, between the intelligent system and the outside world. The physical connection guarantees the necessary contact between the intelligent system and the outside world, whereas the logical connection assures that the intelligent system interacts with the outside world in a manner and applying rules familiar to interaction parties. This chapter elucidates how to learn and interact as well as how to accomplish both.

Chapter 6 is devoted to intelligent system computation modes and processing functions. The processing function system fulfils all the logic of intelligent systems or, in other words, the information processing requirements. This chapter has three main themes. First, it describes intelligent system computation modes and the composition of the processing function system. Second, it analyzes how the software gradually transforms from learning and growth starting in the initiation, assignment, and cultivation phases into its own capabilities of operation, maintenance, and development. Third, it presents the main computation types and functions intelligent system needed, depicts computation methods for these functions, and demonstrates their differences from traditional calculations.

Chapter 7 introduce concerning resource and task function systems. The resource function system is the simplest among the 11 systems discussed in this book although it is a prerequisite for the living and development of intelligent systems. This chapter illustrates principal resource types and how intelligent systems obtain them. It further indicates how intelligent systems take over from human experts, and gradually in their development become in charge of autonomous management, maintenance, and even provisions and modes for exploiting some resources. The tasks in this chapter specifically refer to one type of tasks, from all kinds of intelligent ones summarized in Principles of Intelligence, that are designated externally for intelligent systems to accomplish. Tasks serve as the purpose and basis for the existence of intelligent systems. All processes and phases of intelligent systems center around task execution. This chapter describes the composition of the task function system, task types, and the general process of task execution. It also expounds on methods for addressing some major problems. The most important principle in this context is that intelligent systems can only carry out tasks that have been completed in the past and are submitted externally. As intelligent systems already have experience in successfully completing such tasks, they only need to activate and run microprocessors retained from previous operations and the newly submitted tasks can be accomplished. In the case of new tasks, intelligent systems can utilize the retained microprocessors that have completed similar tasks in simulation during the cultivation or learning processes. Even if the newly accepted tasks are different from those in simulated executions, these differences can be addressed through logical processes with a high degree of certainty. Customers'requirements for the certainty of results can also be met as a result.

Chapter 8 is on living, thinking, control and autonomy. This chapter covers the living and control function systems of intelligent systems, focusing on the composition and realization of autonomy—the core factor among the three essential elements of intelligence. The differences between human intelligence and non-biological intelligence are systematically analyzed in Principle of Intelligence. Possessing autonomy is a prerequisite for the emergence of non-biological intelligence. Without autonomy, a system can only be a tool for non-biological intelligence no matter what algorithms are adopted and how powerful the computation is. The chief functions of autonomy include the awareness and conducts for self-protection, the domination of growth and development, and the independence of social entities. This chapter describes the composition and realization of autonomy in three aspects: living, awareness and thinking, and control. The living system ensures all resources the intelligent system needs in routine operations and that all of its functions can be performed normally. Furthermore, this assurance function should be gradually transferred from human experts to related functions governed by the living function system of non-biological intelligence. Intelligent systems'awareness and thinking are different from that of people. They are not emotions and desires, but a mechanism established for timely discovering and analyzing failures of various parts, potential risks, and global decision-making problems of intelligent systems. This mechanism should be determined and handled by the control function. The control of intelligent system is composed of various autonomous and independent functional components and global control functions while the distributed control functions are coordinated by the global control functions.

Chapter 9 depicts the life cycle of intelligent system. This chapter presents, from initiation to termination, the six phases of non-biological intelligent system. Initiation comes from design and development;assignment relates to formation;cultivation is the realization of functions;growth reflects capability growth relying mainly on learning;performing tasks means to start fulfilling social responsibilities;duplication and termination signal the heredity and closure of intelligent system. The life cycle of intelligent system features the gradual handover process of its dominant subject from the R&D team in early days to the non-biological intelligence itself in late time, which once again highlights the importance of the development environment for intelligent system.

Chapter 10 is Epilogue: Raising the Curtain for the Creation of Non-biological Intelligence That Surpasses Human Intelligence. By recapitulating the major findings of The Nature of Information, Principles of Intelligence and Intelligence Engineering, I would like to reiterate that physics laws and most mathematical tools are not applicable to information and intelligence. Yet, it is a historical responsibility for mankind to create non-biological intelligence that surpasses human intelligence. This is also a requirement necessary for the continuation of mankind and global civilization. Because of the enormous achievements and paradigms that almost become the consensus of researchers in the fields of information, intelligence and artificial intelligence, physics, mathematics and the epistemological frameworks behind them have led to the three fields of information, intelligence and intelligent system construction engineering. These fields are however independent, not following the laws of physical movement and not suitable for using major mathematical tools. That explains why we cannot develop basic theories and make breakthrough progresses in research on non-biological intelligence. It is the most fundamental task for us to change our cognition models and approaches to what is emerging with new development patterns.

My grandson was 8 months old when my book writing was close to completion. I was watching him crawling on the floor. When he started crawling, the harder he tried, the farther he was away from the destination he wanted to reach. But, in a few days, he could crawl effortlessly towards the direction of his destination by coordinating his hands, feet, belly and head in a perfect manner. No one teached him how to crawl in this process. Even if he were taught, he would not be able to understand the teaching. He had no one to imitate either. Nor can other people or animals demonstrate to him how to crawl. This example makes me more certain of intelligence heredity and the necessity of initiation and assignment. In fact, mature intelligence is muscle intelligence and cerebellum intelligence while brain intelligence is immature and attained via learning, trial and error. The intelligence development of intelligent system has to transform brain intelligence into cerebellum or muscle intelligence through learning before intelligent systems can perform intelligent tasks.

This book discusses the realization mechanism, processes, and key points of non-biological intelligence or machine intelligent systems that are gigantic, complex and without any precedent or consensus. It is impossible to cover everything on intelligence engineering in this book in addition to what is left out by the present author intentionally or unintentionally. This book does not elaborate on the engineering specifics of non-biological intelligence. Rather, it only outlines in general the achievability of implementing major components of intelligent systems.

The Nature of Information and Principles of Intelligence consider the natural attributes of information and intelligence. After completing the trilogy of intelligence, the current author will proceed to explore the social and economic attributes of information and intelligence in order to lay the foundation for an economic theory that examines socioeconomic activities after information, a specific resource, and intelligence, a distinctive technology, are introduced and integrated. This book is written based on information structures, manifest information structure completeness and other related contents presented in The Nature of Information as well as the formation and development of intelligence key components, logic, computation architecture and non-biological intelligence discussed in Principles of Intelligence. Because of that, I did not repeat what is already covered in these two books. The readers are advised to refer to related contents in these two books when reading this one in order to gain a better understanding of it.

Frankly speaking, I know there are still many drawbacks in this book when I completed its writing. Many difficulties exist in discussing non-biological intelligence engineering when no such R&D projects have been carried so far. However, there is always the first time and the first person to eat crabs if a non-biological intelligence engineering project is to be conducted. We would only be able to discover problems and correct mistakes when we do our exploration of non-biological intelligence. It is based on this attitude that I concluded the revision of the current book and submitted my manuscript to the publishing house, hoping to set the"ball"rolling.

I would like to thank Prof. Heting Chu, who translated the Preface and Contents during the special Spring of 2020.

I would like to thank Mr Jiuru Liu, the Editor in chief of the Publishing House of Electronics Industry and his colleagues for their dedicated work in getting this book published. Without their efforts, this book would not have been available to readers at such a speed and in such a fine form.

Xueshan Yang

March 29, 2020