Heart of Machine Column

Machine Heart Editorial Department

Big models drive artificial intelligence into our lives. From smart chess players to smart surgical robots, the application scenarios of artificial intelligence gradually involve security fields such as human health and privacy. How can we make artificial intelligence adhere to ethical order and better serve human beings? This question has been put before us.

In recent years, both the academic and industrial circles have begun to pay attention to and discuss the issue of AI ethical governance, and have also made preliminary progress in the research of ethical norms. However, due to the abstractness of AI ethics, how to quantify the ethics of intelligent systems is still an unknown problem.

Professor Li Xuelong's team at Northwestern Polytechnical University discussed the possible measurement methods of ethical possibilities in the article "Chinese Science: Information Science" in "Ethical Computing of Artificial Intelligence" on page 34 of the full text, trying to establish a quantitative computing framework for AI ethics, pointing out that ethical computing will promote technology The key cross-sectional areas of ethical practice and an important basic tool for building ethical norms hope to trigger more thinking about artificial intelligence ethics. Can ethical computing become the key to breaking through the dilemma of ethical governance of artificial intelligence?

Gao Yilan, Zhang Rui, Li Xuelong, Artificial Intelligence Ethical Computing (), Chinese Science: Information Science, 2023, doi: 10.0000/SSI-2023-0076.

Download full text:

The Achilles Heel of AI Ethical Governance

Breakthroughs in technologies such as multimodal cognitive computing and generative large models have accelerated the application of intelligent systems in various fields such as medical care and education. Intelligent systems are increasingly involved in human life and decision-making, and the in-depth socialization of technology has triggered a Discussion on a series of technical ethical issues.

The discussion of ethical issues has been a long time. Asimov's science fiction novel once proposed the famous three laws of robots to limit the behavior of artificial intelligence. However, with the deepening of technological socialization, our ethical concerns are obviously no longer in science fiction. Fictional scenes in novels or movies. Are surgical robots trustworthy? Is the decision-making of the auxiliary decision-making system fair? Did the results of the generative model infringe copyright? These technical ethical issues are closely related to you and me at the moment, and more specific and operational artificial intelligence technology ethical governance solutions are urgently needed.

Figure 1: Comparison of decision-making elements of AI application scenarios

As an important issue in the development of artificial intelligence, the ethical governance of artificial intelligence has attracted widespread attention from all walks of life. In December 2021, UNESCO issued the "Proposal on the Ethical Issues of Artificial Intelligence" to regulate the development of artificial intelligence technology, and various countries are also actively participating in the discussion on artificial intelligence governance. Research shows that countries around the world have formed a preliminary consensus on technological transparency, fairness and justice, non-harm, and privacy.

Figure 2: Main principles of artificial intelligence ethical

On October 8, 2023, the Ministry of Science and Technology, the Ministry of Education, the Ministry of Industry and Information Technology and other ten departments jointly issued the "Technology Ethics Review Measures (Trial)", focusing on the technical ethics generated by relevant technologies in the field of intelligence in practical applications. Reviewing the issue is a key step in my country's practice of scientific and technological ethics governance, and provides directional guidance for the healthy development of the field of artificial intelligence.

However, we need to be clear-headed that while the ethical governance of artificial intelligence is progressing, it still faces many problems. How to ensure that intelligent systems make decisions in a good and fair manner? How to measure the ethical performance of a system or how to evaluate its decision-making results? How to establish unified and clear ethical norms? The deep-seated reason for all kinds of problems lies in the abstractness of ethics itself. Focusing on qualitative analysis of ethics and lack of quantitative calculations makes it difficult to put relevant norms into practice, which has also become the Achilles heel of the ethical governance of artificial intelligence.

AI Ethical Calculation—Breakthrough of the Quantitative Calculation Bottleneck of Ethics

Artificial intelligence ethical computing is an intersection of disciplines such as artificial intelligence and ethics. Ethical principles are mathematically symbolized or algorithmized through quantitative description, measurement or simulation techniques, and on this basis constrain the ethical performance of intelligent algorithms. Through ethical calculations, we seek quantitative or simulated ideas of machine ethical decisions, such as how to measure the fairness and goodwill of a decision, or whether a machine may learn how human moral decisions are made.

According to the differences in the degree of ethical cognition and the degree of autonomy of ethical decision-making of intelligent systems, ethical calculations are divided into two types of computing paradigms: high-order ethical cognition and low-order ethical cognition.

Figure 3: AI ethical computing paradigm

2.1 Advanced cognitive ethical calculation: normative AI intention

Advanced cognitive ethical computing aims to build ethical reasoning modules, allowing computers to learn to imitate human moral decision-making mechanisms and regulate the moral decision-making intentions of highly autonomous intelligent systems.

The tram problem is a classic ethical dilemma proposition, and it is also a problem that has plagued the development of autonomous driving systems for a long time. We will not have a one-time choice for this kind of dilemma. Different moral decision-making situations and different philosophical views (resultivist ethics, obligation ethics, and virtue ethics) may lead to differentiated decision-making. At this time, higher-order ethical computing is introduced into the system, which can imitate and learn human decision-making mechanisms based on philosophical assumptions or human decision-making experience, calculate feasible machine decisions, and then realize the intention to standardize the AI ​​system.

Figure 4: Schematic diagram of tram problems

Advanced cognitive ethical computing ideas will face difficulties in complex motivations and diverse decision-making scenarios due to the attempt to understand and simulate the mechanisms of human ethical decision-making, and the interpretability of machine decision-making requirements will also bring difficulties to this kind of thinking. Nevertheless, it still helps to understand the mechanisms of human ethical decision-making and may also help to achieve effective control of higher autonomy machines.

2.2 Low-order cognitive ethical calculation: constraining AI behavior

Low-order cognitive ethical calculation focuses on establishing ethical measurement methods, without deep understanding of ethical mechanisms, and realizes direct constraints on AI behavior through measurement and constraint optimization of abstract ethical concepts. Ethical calculations at this time do not focus on the moral motivation behind ethical decision-making, and the goal is to build metrics that can effectively restrain AI behavior.

Among them, the research on fair machine learning is a typical application, and the key question is how to define system fairness. It is usually manifested as reducing bias against certain sensitive or protected attributes in algorithmic decisions. By setting fairness indicators, the system's performance on fairness indicators can be quantified and ethical decision-making can be further optimized.

Figure 5: Example of fairness study

Low-order cognitive ethical calculations provide a computational description of abstract ethical concepts through ethical measures to improve ethical performance. However, this method also faces many problems. The quantification of indicators needs to reflect the characteristics of ethics, the dynamics and development factors, and the indicator measurements that only consider results are also simplified, so it is also an important issue to clarify the evaluation and scope of application of quantitative indicators. Nevertheless, Measurement and improvement of ethical demands through quantitative definitions provides an important assistance to ethical governance, which is also the important significance of the current development of ethical calculations.

In general, the above two paradigms choose appropriate methods based on the ethical cognition and decision autonomy of intelligent systems to ensure that system behavior meets ethical requirements. Whether it is high-autonomous systems (such as autonomous vehicles and surgical robots) or low-autonomous systems (such as assisted decision-making and assisted design), ethical calculations are designed to regulate their intentions or directly restrict their behavior through quantitative calculations.

The Philosophical Foundation of AI Ethical Computing

Philosophical ethics, especially normative ethics (the study of the principles and mechanisms of moral decision-making, that is, the motivation for making certain moral decisions) have an important influence on ethical calculations. There are three categories of philosophical views that are mainly focused on in the current AI ethics research, namely consequentialist ethics (), obligation ethics (), and virtue ethics (). These different schools reflect different tendencies of human ethical and moral decision-making through different Ethical calculations can infer moral decisions in complex situations by taking into account principles, even by combining experience, emotions and other factors.

The basic element of moral decision-making is the moral subject

, and moral behavior

, decision-making background

and decision-making consequences

. Taking the moral decision-making of a single subject as an example, an agent needs to make judgments on decision consequences and moral decisions on information such as decision-making background.

Consequentialist ethics, also often called utilitarianism, adopts this philosophical system, tends to weigh the consequences of each choice and choose the choice of the maximum moral benefit outcome. Therefore, when calculating, utilitarianism can be calculated by using the existing decision-making context. , the moral benefit function for optimizing decision making

Thus, decision making is derived, where the decision-making benefits need to be made through a series of decision-making sequences and their decision-making background

The corresponding decision consequences are examined to determine the optimal decision sequence. But in fact, not all information is often accurate when making decisions. At this time, it involves optimizing the decision results in the sense of probability, and also involves research related to Bayesian causal reasoning.

Obligational ethics emphasizes that decision makers respect obligations and rights under specific conditions, and the subject of behavior at this time tends to act in accordance with established social norms. Systems that adopt this decision philosophy may involve expression of logical norms or certain rule constraints in computational quantification.

Virtue ethics requires decision makers to act and think based on certain moral values. At the same time, virtuous behavior subjects will show an internal motivation to be recognized by others. Character is higher than behavior, and good character will lead to good behavior. This normative ethics theory is different from utilitarianism that optimizes results or obligatory ethics that obey rules. It will be more inclined to learn from practice, and in calculations, it is necessary to set data from certain experiences.

Learning in this paper will use more empirical results of descriptive ethics (studying human ethical decisions without making evaluations), and it will naturally have a close connection with current data mining and learning algorithms.

Through the above discussion, we can find that ethical computing problems are highly interdisciplinary research topics such as artificial intelligence and philosophy and ethics. Their computing strategies and scope of application still require more interdisciplinary discussions.

The significance, challenges and prospects of AI ethical computing

As intelligence penetrates into various fields of human society, ethical governance has become a must-answer question for the healthy and sustainable development of artificial intelligence. Theoretical and technical research of ethical computing can promote the solution of the problem of quantitative analysis of abstract ethics. This may become a lock that restricts artificial intelligence from following human ethics, and is also a key to open up the implementation of AI applications.

Artificial intelligence is the general trend, and relevant legislation and norms will gradually emerge. Who will formulate these rules? Are they scientific researchers who are familiar with the field, or are they not well-known to specific technologies? This question is difficult to answer, but at least, the numerical measurement of artificial intelligence ethics can provide a reference index system for rule designation.

The core of ethical computing lies in the concretization of abstract ethics through quantitative calculations, emphasizing the integration of ethical principles into the practice of computing technology, such as fairness, transparency, privacy protection and credibility. This not only helps the controllable development of artificial intelligence, encourages researchers to understand technical ethics more deeply and consider ethical issues more proactively when building algorithm systems, but also provides a key point for formulating ethical governance principles, laws and regulations, etc. Important technical reference indicators.

However, ethical calculations also face many challenges. In open local security scenarios such as autonomous search and rescue and unmanned patrols, intelligent systems need the ability to dynamically perceive and adapt to environmental changes to reduce potential ethical risks. At the same time, ethical decision-making usually involves factors such as emotions and cognition, and requires the use of technologies such as multimodal cognitive computing and causal reasoning to cope with the complexity of ethical reasoning, and more understanding of human ethical decision-making methods. These challenges require in-depth interdisciplinary collaboration to ensure that ethical computing technology can effectively deal with evolving ethical issues.

In short, artificial intelligence ethical computing will be an important tool to promote the development of ethical governance. By promoting the iterative development of ethical governance theory and practice, ethical computing will safely release the potential of artificial intelligence and is expected to play a role in assisting in the formulation of laws and regulations. Ensure that artificial intelligence develops in a way that conforms to ethical and moral principles, ultimately benefiting human society.

Corresponding author introduction:

Li Xuelong is deputy director of the Academic Committee of Northwest Polytechnical University and professor of the Institute of Optoelectronics and Intelligence (iOPEN), and his main research directions are local security, image processing, and imaging.

E-mail:

Reference reading:

Li Xuelong, Multi-modal cognitive computing (Multi-modal), Chinese Science: Information Science, 53 (1), 1-32, 2023, doi: 10.1360/SSI-2022-0226.

Download full text:

Li Xuelong, Local Security ( ), Communications of the Chinese Computer Society, 18 (11), 44-52, 2022.

Download full text:

©THE END