As a scientific research field that attempts to study, imitate, and expand human intelligence, artificial intelligence technology has been accompanied by profound technical ethics debates since its birth. With the breakthroughs and rapid implementation of machine learning and other related work in recent years, ethical issues have become increasingly prominent and have forced the academic community and society to begin to face the ethical governance challenges of this technology. Although preliminary progress has been made in normative research on ethical governance, there are still many difficulties in implementing its governance practices, and ethical practices have shown a tendency to gradually lag behind the needs of technological development. Therefore, establishing an ethical governance practice plan that matches the evolving artificial intelligence technology and realizing a positive interaction between governance theory and governance practice will be a key issue in the future development of the field of artificial intelligence. The abstract nature of ethical governance theory makes it difficult to implement the current ethical principles of artificial intelligence. Artificial intelligence ethical computing (AI) will be an important solution to meet this challenge. This study clarifies the importance of ethical computing by exploring the practical necessity and development possibility. Based on relevant research, the research scope of ethical computing is given. The degree of cognition of ethical mechanisms and the degree of autonomy of system ethical decision-making are divided according to the calculation process. Two types of research paradigms of high-order cognition and low-order cognition of ethical computing are established, and three levels of calculation of ethical measurement, ethical decision-making and ethical reasoning are abstracted according to their calculation stages. This ethical calculation framework can sort out the current ethical computing applications. This article takes ethics embedding and fair machine learning as examples to illustrate the research characteristics and technical methods of the two types of research paradigms. On this basis, it further discusses and constructs an ethical governance system with ethical computing as the core, analyzes possible solutions to resolve ethical governance dilemmas through ethical computing, and makes prospects for the development of artificial intelligence ethical computing.

Improvements in massive data, basic computing power, intelligent algorithms, dedicated hardware and other aspects have prompted the booming development of the artificial intelligence industry. Various technologies continue to reshape human living habits and production methods. The application breadth and depth of related technologies continue to expand, giving rise to various cross-integration fields including AI for , quantum machine learning research, etc. New technologies emerge one after another and new applications are constantly implemented, bringing unprecedented opportunities to the development of the field of artificial intelligence. However, at the same time, technological ethical anxiety is also increasing. Effective ethical governance of technology is urgent. What ethical issues are faced with the current development of artificial intelligence technology? What are the difficulties in ethical governance? What technical means can assist ethical governance? This article proposes to build a technical tool system for ethical governance with artificial intelligence ethical computing (AI, hereafter referred to as ethical computing) technology as the core. What is ethical computing? By which technologies are ethical computations performed? How to build an ethical governance system through this technical means? The discussion will be carried out later. Let’s start with the ethical issues faced, taking the following three typical application scenarios as examples. (1) Artificial intelligence technology has brought major changes to the medical and health field, mainly in two forms of technical support: virtual branch and physical branch. The virtual branch refers to the artificial intelligence algorithm that can use big data to mine potential medical auxiliary information, including protein response prediction, drug prediction, psychological rehabilitation auxiliary treatment, etc. The physical branch refers to medical service robots supported by various artificial intelligence algorithms. These include robot companions used to take care of critically ill patients, assistant doctors in surgeries and even surgeons. In these important application scenarios related to human life and health, ethical issues are particularly prominent. This involves how to protect the privacy of stakeholders from infringement, and also involves the difficulty of how to ensure that decisions do not cause damage to the lives and health of the audience. For example, in the application scenarios of artificial intelligence surgeons, How to define whether a surgical accident is an accident? Who bears the responsibility for the accident? Such issues involve various complex ethical subjects, and the ethical issues behind them are often very complex. Is it possible to use technical methods to assist in defining these complex ethical issues? (2) Self-driving cars are expected to improve transportation efficiency and reduce the probability of traffic accidents. However, due to their high degree of autonomy, this scenario also involves very complex ethical issues: How do self-driving machines make moral decisions and how do they determine liability for accidents? How can society quantify the ethical principles that guide machine behavior? These have become issues that urgently need to be responded to. Relevant research conducted an extensive social experiment to address the major challenges in this field, collecting 40 million ethical decisions from 233 countries and regions and conducting statistical analysis, pointing out that moral choices show a unified trend from a macro perspective, but there are still inherent conflicts in moral decision-making, interpersonal differences, regional and cultural differences, etc. In order to reduce conflicts and clarify differences, this field also needs more ethical dialogues and more precise technical quantitative standards. (3) Computer-aided decision-making is also one of the important applications of artificial intelligence. The algorithm's effective mining of massive data allows it to deeply explore historical decisions and learn decision-making factors. The improvement of automation in decision-making methods can greatly improve decision-making efficiency, but the bias and discrimination hidden behind it are also worrying. Typical scenarios include systems for recidivism probability estimation, automatic screening of recruitment resumes, Advertising push, etc., these artificial intelligences used for decision-making will have a significant impact on society and individuals, and the transparency and fairness of their decision-making mechanisms need to be improved urgently. Not only in decision-making systems, but also in natural language processing, computer vision and other types of research, there are problems of acquiring and amplifying historical biases. How to improve these problems requires more exploration. At the same time, relevant proposals put forward for assisted decision-making scenarios also pointed out that the application of artificial intelligence in decision-making may lead to changes in organizational culture and personal behavior. Therefore, it is also necessary to develop practical evolutionary indicators of the impact of artificial intelligence technology to measure its benefits and long-term and short-term impact on decision-making stakeholders. The above are ethical issues that exist in some classic application scenarios. The application of new technologies is constantly posing new challenges to ethical research. For example, the generation of large models represented by AI painting has recently triggered discussions on copyright disputes and plagiarism issues. The following article will discuss the progress of ethical computing, Problems caused by such large models will also be briefly discussed. It is not difficult to see that current artificial intelligence technology has caused a wide range of profound social problems, and will face more complex technical ethical dilemmas with the development of technology, so there is an urgent need to provide effective solutions. In fact, the discussion of artificial intelligence ethical issues has already occurred as early as 1960, and artificial intelligence ethics research has accompanied the development of technology. However, early research mainly revolved around ethical theory, which was divorced from specific application scenarios and highly abstract. Therefore, How to effectively consider ethical factors in actual application scenarios and construct practical ethical governance solutions is an important research topic. Regarding ethical governance issues, in recent years, various national organizations have made important efforts and put forward many technical ethical demands including explainability, fairness, privacy, etc. Relevant industry development norms have been formulated. Although many legal norms and initiatives have been proposed, the practice of ethical governance still faces many difficulties due to the ambiguity and difference of these abstract indicators. This article believes that concretizing ethical norms, quantifying ethical demands and providing technical support for ethical norms is the key to breaking the situation, that is, eliminating abstract ambiguities through ethical calculations and providing quantitative features, which is expected to make up for the differences between theory and practice. However, the key issue facing ethical calculations is: Can abstract ethical concepts be calculated? How to do the calculation? How should computing serve ethical governance practices? This article will start from the development status and dilemmas of artificial intelligence ethics research, and point out the necessity of ethical computing research. By exploring the research and development of ethical computability, we will discuss the possibility of ethical computing and define the concept of ethical computing. Further focusing on the issue of how to calculate, we will summarize the existing computing methods and computing paradigms of ethical computing, and give examples of representative ethical computing technologies. Finally, we will also analyze the relationship between computing and ethical governance.

The overall discussion idea is shown in Figure 1, which completes the explanation of ethical computing from computing system to ethical governance system. Specifically, the second section summarizes and discusses the realistic background and computing history of ethical computing research. It explains the current development status of ethical research, points out the development pain points of ambiguous ethical theory and limited practical feasibility in ethical governance, points out the important practical significance of ethical computing, and analyzes the research background of computational ethics on this basis. Section 3 gives the specific definition of ethical computing in this article. The research goals, research methods and research paradigms of ethical computing are summarized, and two types of representative work are given. Fair machine learning is used as a case to give examples of ethical computing. Section 4 constructs an ethical governance system based on ethical computing, and finally in Section 5, the outlook for artificial intelligence ethical computing and ethical governance is given.

Expertise and easy viewing