Machine Heart Column
Machine Heart Editorial Team
Large models driven by artificial intelligence are entering our lives, from intelligent chess players to intelligent surgical robots, and the application scenarios of AI are gradually involving safety areas such as human health and privacy. How can we ensure that AI adheres to ethical orders to better serve humanity? This question has now been placed squarely before us.
In recent years, both academia and industry have begun to pay attention to and discuss the issue of AI ethical governance, and initial progress has been made in the research of ethical norms. However, due to the abstract nature of AI ethics, how to quantitatively measure the ethics of intelligent systems remains an unknown challenge.
Professor Li Xuelong’s team at Northwestern Polytechnical University discusses possible measurement methods for ethics in the article “Ethical Computation in Artificial Intelligence” published in Chinese Science: Information Science, covering 34 pages, attempting to establish a quantitative computing framework for AI ethics, pointing out that ethical computation will be a key interdisciplinary field to promote the practice of technological ethics and an important foundational tool for constructing ethical norms, hoping to inspire more thoughts on AI ethics. Can ethical computation become the key to breaking through the dilemma of AI ethical governance?
Gao Yilan, Zhang Rui, Li Xuelong, Ethical Computation in Artificial Intelligence (Artificial Intelligence Ethical Computation), Chinese Science: Information Science, 2023, doi: 10.0000/SSI-2023-0076.
Full text download:
https://www.sciengine.com/SSI/doi/10.0000/SSI-2023-0076
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 healthcare and education, leading to an increasing involvement of intelligent systems in human life and decision-making, triggering a series of discussions on technical ethical issues.
The discussion of ethical issues has a long history. Isaac Asimov’s science fiction novels proposed the famous three laws of robotics to limit AI behavior, but with the deepening of technological socialization, our ethical concerns are clearly no longer confined to fictional scenarios in novels or movies. Can surgical robots be trusted? Are the decisions made by decision support systems fair? Do the outputs of generative models infringe copyright? These technical ethical issues are directly related to us today, urgently requiring more specific and actionable ethical governance solutions for AI technology.
Figure 1: Comparison of Decision Elements in AI Application Scenarios
As an important topic in the development of artificial intelligence, AI ethical governance has attracted widespread attention from all sectors. In December 2021, UNESCO released the Recommendations on the Ethics of Artificial Intelligence to regulate the development of AI technology, while various countries are also actively participating in discussions on AI governance. Research shows that countries around the world have reached a preliminary consensus on technical transparency, fairness, non-harm, privacy, and other aspects.
Figure 2: Major Ethical Principles of Artificial Intelligence
On October 8, 2023, ten departments including the Ministry of Science and Technology, the Ministry of Education, and the Ministry of Industry and Information Technology jointly issued the Interim Measures for the Review of Scientific and Technological Ethics, focusing on the ethical review issues arising from the practical application of technologies in the intelligent field. This is a key step for China in the practice of scientific and technological ethical governance, providing directional guidance for the healthy development of the AI field.
However, we need to be aware that while progress is being made in AI ethical governance, many problems still remain. How can we ensure that intelligent systems make decisions in a benevolent and fair manner? How can we measure the ethical performance of a system or evaluate its decision-making results? How can we establish unified and clear ethical norms? The deep-rooted cause of these problems lies in the abstraction of ethics itself. The focus on qualitative analysis of ethics, lacking quantitative computation, makes it difficult to implement relevant norms, which has also become the Achilles’ heel of AI ethical governance.
AI Ethical Computation – Breaking the Bottleneck of Quantitative Ethical Calculation
AI ethical computation is an interdisciplinary field that intersects artificial intelligence and ethics, quantitatively describing, measuring, or simulating ethical principles through mathematical symbolization or algorithmization, and constraining the ethical performance of intelligent algorithms based on this. Through ethical computation, we seek to quantify or simulate the ethical decision-making of machines, for example, how to measure the fairness or benevolence of a particular decision, or whether machines can learn human moral decision-making processes.
Based on the different levels of ethical cognition and autonomous decision-making of intelligent systems, ethical computation is divided into two paradigms: high-level ethical cognition and low-level ethical cognition.
Figure 3: Paradigms of AI Ethical Computation
2.1 High-Level Cognitive Ethical Computation: Norming AI Intent
High-level cognitive ethical computation aims to construct ethical reasoning modules, enabling computers to learn and imitate human moral decision-making mechanisms, thus norming the moral decision-making intentions of highly autonomous intelligent systems.
The trolley problem is a classic ethical dilemma that has long troubled the development of autonomous driving systems. There are no one-size-fits-all choices for such dilemmas; different moral decision-making contexts and philosophical perspectives (consequentialist ethics, deontological ethics, virtue ethics) can lead to differentiated decisions. At this point, introducing high-level ethical computation into the system can calculate feasible machine decisions based on philosophical assumptions or human decision-making experiences, thereby norming the intentions of AI systems.
Figure 4: Schematic Diagram of the Trolley Problem
High-level cognitive ethical computation faces challenges due to the complexity of human ethical decision-making motives and diverse decision-making scenarios, and the requirement for machine decisions to be interpretable also poses difficulties for this approach. Nevertheless, it still helps to understand the mechanisms of human ethical decision-making and may assist in effectively controlling more autonomous machines.
2.2 Low-Level Cognitive Ethical Computation: Constraining AI Behavior
Low-level cognitive ethical computation focuses on establishing ethical measurement methods without requiring an in-depth understanding of ethical mechanisms. By measuring and optimizing abstract ethical concepts, it achieves direct constraints on AI behavior. At this point, ethical computation does not concern the moral motives behind the system’s ethical decisions; the goal is to construct measurement indicators that can effectively constrain AI behavior.
Among them, research on fair machine learning is a typical application, with the key issue being how to define system fairness. This usually manifests as reducing bias against certain sensitive or protected attributes in algorithmic decisions. By setting fairness indicators, we can quantify the system’s performance on fairness metrics and further optimize ethical decisions.
Figure 5: Example of Fairness Research
Low-level cognitive ethical computation provides a computational description of abstract ethical concepts through ethical metrics to improve ethical performance. However, this method also faces many issues. The quantification of indicators needs to reflect the characteristics of ethics as a dynamic and developing factor, and only considering outcome-based metrics also oversimplifies the situation. Therefore, clarifying the evaluation and applicable scope of quantification indicators is also an important issue. Nevertheless, quantifying definitions to measure and improve ethical demands provides important assistance for ethical governance, which is also the significant meaning of the current development of ethical computation.
In summary, the above two paradigms select appropriate methods based on the ethical cognition and decision-making autonomy of intelligent systems to ensure that system behavior meets ethical requirements. Whether for highly autonomous systems (such as autonomous vehicles and surgical robots) or low-autonomy systems (such as decision support and design assistance), ethical computation aims to norm their intentions or directly constrain their behavior through quantitative computation.
Philosophical Foundations of AI Ethical Computation
Philosophical ethics, especially normative ethics (which studies the principles and mechanisms of moral decision-making, i.e., the motives for making certain moral decisions), has a significant influence on ethical computation. The main philosophical perspectives currently focused on in AI ethics research include three categories: Consequentialism, Deontology, and Virtue Ethics. These different schools reflect the different tendencies of human ethical and moral decision-making. Through different principles, and even considering a combination of experiences and emotions, ethical computation can infer moral decisions in complex situations.
The basic elements of moral decision-making are moral subjects
, moral behaviors
, decision contexts
, and decision outcomes
. Taking the moral decision of a single subject as an example, the agent needs to judge the consequences of decisions based on context information and make moral decisions.
Consequentialism, often referred to as utilitarianism, tends to weigh the consequences of each choice and select the option that maximizes moral benefit. Therefore, in calculations, utilitarianism can optimize the moral benefit function of decisions within existing decision contexts, thus deriving decisions where the decision benefits need to be examined through a series of decision sequences and their corresponding contexts
to determine the optimal decision sequence. However, in reality, not all information is accurate during decision-making, which involves optimizing decision outcomes in a probabilistic sense and may also relate to Bayesian causal reasoning research.
Deontological ethics emphasizes that decision-makers respect obligations and rights under specific conditions, thus the behavior subject tends to act according to established social norms. Systems adopting this decision philosophy may involve expressing logical norms or certain rule constraints in quantitative calculations.
Virtue ethics requires decision-makers to act and think based on certain moral values, and moral agents with virtue exhibit an intrinsic motivation recognized by others. Character is above behavior; good character leads to good behavior. This normative ethical theory differs from utilitarianism, which optimizes outcomes, or deontological ethics, which adheres to rules, and is more inclined to learn from practice. In calculations, it needs to learn from certain empirical datasets
, utilizing more descriptive ethics (which studies human ethical decision-making without evaluating it) and is closely related to various data mining and learning algorithms.
The Significance, Challenges, and Prospects of AI Ethical Computation
As intelligence penetrates various fields of human society, ethical governance has become a necessary question for the healthy and sustainable development of AI. The theoretical and technical research of ethical computation can help solve the challenges of abstract ethical quantitative analysis, which may become a lock constraining AI to adhere to human ethics, and also a key to unlocking the application of AI.
Artificial intelligence is an unstoppable trend, and relevant legislation and norms will gradually emerge. Who will formulate these rules? Researchers familiar with the field or groups that do not sufficiently understand specific technologies? This question is difficult to answer, but at least, the numerical measurement of AI ethics can provide a reference indicator system for rule formulation.
The core of ethical computation lies in quantifying and concretizing abstract ethics, emphasizing the integration of ethical principles into the practice of computational technology, such as fairness, transparency, privacy protection, and trustworthiness. This not only helps the controllable development of artificial intelligence and encourages researchers to understand technological ethics more deeply and consider ethical issues more proactively when constructing algorithm systems, but also provides crucial technical reference indicators for formulating ethical governance principles, laws, and regulations.
However, ethical computation also faces many challenges. In open-field security scenarios such as autonomous search and rescue and unmanned inspections, 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 often involves emotional and cognitive factors, requiring multimodal cognitive computing and causal reasoning technologies to tackle the complexity of ethical reasoning, as well as a better understanding of human ethical decision-making processes. These challenges require in-depth interdisciplinary collaboration to ensure that ethical computation technology can effectively address the evolving ethical issues.
In conclusion, AI ethical computation will serve as an important tool to promote the development of ethical governance. By facilitating the iterative development of ethical governance theory and practice, ethical computation will safely unleash the potential of artificial intelligence and is expected to play a role in assisting the formulation of regulations to ensure that AI develops in accordance with ethical and moral principles, ultimately benefiting human society.
Corresponding Author Introduction:
Li Xuelong, Vice Chairman of the Academic Committee of Northwestern Polytechnical University, Professor at the Institute of Optoelectronics and Intelligence (iOPEN), with main research directions in situational security, image processing, and imaging.
E-mail: [email protected]
References:
Li Xuelong, Multimodal Cognitive Computing, Chinese Science: Information Science, 53 (1), 1-32, 2023, doi: 10.1360/SSI-2022-0226.
Full text download:
https://www.sciengine.com/SSI/doi/10.1360/SSI-2022-0226
Li Xuelong, Vicinagearth Security, Communications of the Chinese Computer Society, 18 (11), 44-52, 2022.
Full text download:
https://dl.ccf.org.cn/article/articleDetail.html?type=xhtx_thesis&_ack=1&id=6219452051015680
© THE END
For reprints, please contact this public account for authorization
For submissions or seeking reports: [email protected]