Skip to content

IEEE standards on AI

Ethically aligned design (Version II)

Ethically Aligned Design paperwork, published under the auspices of the IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems, highlights the ethical considerations in the design of autonomous and intelligent systems (A/IS). The document contains eleven chapters, and each chapter accents the following aspects.

  • Ch. 1 - From Principles to Practice: In this chapter, first, they introduce the three main areas which their framework expands into, namely, universal human values, political self-determination and data agency, and technical dependability. Then they connect these with the high-level general principles introduced in the second chapter.

  • Ch. 2 - General Principles: Chapter two details the general principles: human rights, well-being, data agency, effectiveness, transparency, accountability, awareness of misuse, and competence. These general principles guide all manner of ethical A/IS design.

  • Ch. 3 - Classical Ethics in A/IS: In the current digital age, human control of AI tools in the deployment phase has been reduced. Hence, the creator of such tools should ask themselves how culturally and ethically sound the design of these tools are before they are implemented. This is the emphasis of chapter three where they bring two thousand years’ worth of classical ethics traditions into consideration.

  • Ch. 4 - Well-being: The current measures of success of AI tools such as profit, gross domestic product (GDP), consumption levels, and occupational safety do not encompass the well-being of humans and society. This raises the issue that AI developers overlook innovation toward well-being and societal value. Chapter four makes AI developers aware of these issues and makes them consider the well-being aspect while designing AI tools.

  • Ch. 5 - Affective Computing: During the interaction between humans and AI, AI systems could simulate emotions in humans. This human emotional experience due to AI systems has the potential to change human life, and eventually society, both positively and negatively. Chapter five addresses issues related to emotions and emotion-like control in interactions between humans and the design of A/IS.

  • Ch. 6 - Personal Data and Individual Agency: Chapter six discusses the unpopular aspect of data protection, i.e., algorithms learned from personal behavior influence the individuals’ choices and shape their life trajectory. Hence, governments and organizations should institute regulations as to who can process what personal data for what purposes.

  • Ch. 7 - Methods to Guide Ethical Research and Design: Chapter seven contains three sections: Interdisciplinary Education and Research, Practices on A/IS, and Responsibility and Assessment. The aim of the chapter is to avail the researchers, product developers, and technologists to research and develop methods that evaluate their processes, products, values, and design practices in light of ethics.

  • Ch. 8 - A/IS for Sustainable Development: If operated aptly, AI is a great opportunity for high-income and low-and middle-income countries to attain positive socioeconomic outcomes and sustainable development goals. Chapter eight provides an insight into how AI should be harnessed to achieve sustainable development.

  • Ch. 9 - Embedding Values into Autonomous and Intelligent Systems: Although it is essential that AI systems adapt, learn, and follow the norms and values of the community they serve, current AI standards and principles do not review embedding human values and norms into AI. This chapter helps identify, implement, and evaluate values that should be embedded into AI.

  • Ch. 10 – Policy: AI policies and regulations should be developed to protect and promote safety, privacy, human rights, and cybersecurity, as well as enhance the public’s understanding of the potential impacts of A/IS on society. Otherwise, there may be critical technology failures, loss of life, and high-profile social controversies. Chapter ten presents a rights-based approach for the policymakers to build AI policies and regulations.

  • Ch. 11 – Law: Technological innovation caused by AI should be guided by laws. On the other hand, the law should respond to technological innovation due to AI. This complex interactive process should eventually ensure that A/IS, in both design and operation, is aligned with principles of ethics and human well-being. This is the focus of the last chapter of this document.

The aim of the document is to provide recommended practices for AI developers to increase and help safeguard human well-being at the individual, population, and societal levels. The document specifies specific and contextual well-being metrics that facilitate the use of the Well-Being Impact Assessment (WIA) process. WIA is an iterative procedure that should be repeated throughout the AI lifecycle. The WIA is composed of five activities, as follows:

  • Activity 1 (Internal, User, and Stakeholder Impact Assessment) involves first identifying and later refining well-being domains and indicators.

  • Activity 2 (Well-Being Indicators Dashboard) helps to form the well-being indicator dashboard based on the domains and indicators selected in Activity 1.

  • Activity 3 (Data Collection Plan and Data Collection) populates the well-being indicator dashboard with data. This involves both planning and collecting data from users and stakeholders.

  • In Activity 4 (Well-Being Data Analysis and Use of Well-Being Data), the well-being data is analyzed and used for the design, development, deployment, monitoring, and iterative improvement of the AI tool.

  • Activity 5 (Iterate) involves assessing the WIA process and well-being indicator dashboard. Changes necessary to be made for the next round are marked and continue.

If one follows successfully,

  • One will develop AI with well-being concepts and indicators in mind.

  • One will be able to monitor, evaluate, and address AI impact on human well-being.

  • One will get the ability to evaluate the ongoing well-being impacts of AI.

  • One will be able to use acquired knowledge to help the continuation of safeguarding and improving human well-being.

  • One will gain the ability to avoid unintentionally harming the well-being of users.