Practical Guide: AI Governance & Ethics
Aiming to promote the responsible and ethical use of AI
Disclaimer: This Practical Guide: AI Governance and Ethics (“Guideline”) acts as a general guide to enhance the ethical use of artificial intelligence (“AI”). It serves to encourage a safe adoption approach for public and private sector organisations. It should be noted that this guideline should be treated as a ‘living document’ that will be continuously reviewed to ensure it assists and creates a standard for safe adoption. The definitions and illustrations of the 7 AI Principles are meant to be broad and should be carefully considered by AI actors when applying this Guideline for their own operations.
About The Practical Guide: AI Governance & Ethics
Objective
AI has great potential to bring new ideas, introduce innovation and improvements in many areas. Thus, making it an exciting tool for productivity improvement. AI has emerged as a powerful tool driving innovation across numerous industries.
While AI is powerful, organisations and society must use it wisely. Before an organisation adopts AI, it must make assessment of its risks-rewards trade-off. To mitigate the risks, and to ensure AI is used for good, organisations need standards and guidelines.
Malaysia aspires to be a leading AI hub, encouraging both local and international organizations to use AI effectively. This goal must be balanced with the need to protect our core values and safe use within society.
As part of an initiative under the Malaysia National Artificial Intelligence Roadmap 2021-2025 (AI-RMP), Ministry of Science, Technology and Innovation (MOSTI) had released the National Guidelines on AI Governance and Ethics (AIGE) on 20 September 2024. It outlines seven (7) principles for a responsible AI ecosystem (“the 7 AI Principles”).
This Guideline provides recommendations for the public and private sector to consider in adopting responsible AI practices.
Terms in AI Taxonomy
Artificial Intelligence
AI Actor
AI System
AI Lifecycle
Developer
Deployer
End User
The AI Lifecycle
The AI lifecycle encompasses the essential stages involved in developing, deploying and managing AI systems which may be broadly categorized into six stages.
While this categorisation provides a broad understanding of the AI lifecycles, it is important to note that the lifecycle can vary significantly depending on the perspective from which it is viewed. For example, technical perspectives may emphasize the iterative nature of model development, while ethical and regulatory perspectives might highlight governance and compliance checkpoint at each stage.
Among all stages in the AI lifecycle, Verifying & Validation is the most extensive yet often overlooked. It ensures models are reliable, fair, and accurate before deployment. Without proper validation, AI risks bias and inaccuracies, impacting trust and decision-making.
Furthermore the lifecycle can be made more granular or adapted based on specific use cases or sectoral needs. For instance, in healthcare, there may be additional stages for clinical validation or patient safety checks, while in the financial sector, heightened emphasis may be placed on fraud detection and regulatory compliance.
Ultimately, the AI lifecycle is not a one-size-fits-all model but rather a flexible framework that evolves to address the unique challenges and opportunities of each application.
Non-Linear Nature of the AI Lifecycle
It's important to recognize that the AI lifecycle is not strictly linear. Stages often overlap, and iterations are common. For instance, insight gained during the Operating and Monitoring phase may reveal the need for revisiting the Planing and Design stage to address unforeseen issues or enhance functionality. This iterative process ensures continuous improvement and adaptability of the AI system.
Application of Responsible AI Principles
Across the Lifecycle Implementing responsible AI principle requires contextual application at each stage of the lifecycle. Each principle may manifest differently depending on the phase. For example, during the Data Collection and Processing stage,fairness involves ensuring that the data set is representative and free from biases that could lead to discriminatory outcomes. In contrast, during the Building and Using the Model phase, fairness focuses on selecting algorithms that do not perpetuate existing biases and are equitable across different user groups.
