AI Resource Management Policy Template| ISO 42001 AIMS

The AI resource management policy process defines guidelines to handle AI resources which enable responsible and ethical operations of AI systems that follow AI Principles described in AI Policy. This policy exists to allow the organization complete knowledge of AI system resources including components and assets for addressing risks and impacts.

AI Resource Management Principles

1. Resource Documentation

The organization needs to identify and document every necessary resource needed for activities that happen during AI system life cycle phases in addition to activities relating to AI that are important for the organization.

A comprehensive understanding of risks and potential AI system impacts on individuals and groups of individuals along with societies depends on proper resource documentation of the AI system. Documentation of these resources utilizes data flow diagrams or system architecture diagrams to generate information for AI system impact assessments. 

The organization will document AI system resources with minimum annual reviews followed by mandatory updates. 

These resources include a combination of items which could also encompass:

AI System Components.

• Data resources include all datasets that the AI system needs for its lifecycle operations.
  
  
• The resources include AI algorithms as well as models and tools that function as tools.
  
  
• The system needs hardware development and model execution capabilities with sufficient data storage facilities for tools as computing platform resources.
  
  
• Human resources consist of people who possess the needed expertise to support AI system development from development through maintenance thus enabling organizational involvement at all stages.
  
  
• The organization can access these resources either through internal provision or customer or third-party supply.

2. Data Resources

The organization needs to document every type of data resource it uses for building its AI system through resource identification activities. The team will perform data documentation updates at least once annually and at any time when technological changes occur.

 The relevant documentation of data must cover the following details:

• The provenance of the data.
• The metadata contains information about the update or modification date of the data resources (e.g. date tag in metadata).
• When using machine learning the distinct data groups require documentation (training, validation, test and production data).
• As defined in ISO/IEC 19944-1 the categories of data must be added to the documentation.
• Process for labelling data.
• Intended use of the data.
• The quality standards of data adhere strictly to the ISO/IEC 5259 series documentation.
• Applicable data retention and disposal policies.
• The data shows known bias types along with estimated biases which exist within its records.
• Data preparation.
  

3. Tooling Resources

Resource identification at our organization demands documentation of all tooling elements which power the AI system. Regular updates of tooling resource documents are scheduled for annual periods alongside changes made to the technological environment. The list of tooling resources needed for AI systems and machine learning specifically includes multiple items beyond the specified range but does not include every possible resource.

Limited to:

• The document records machine learning models together with algorithms and other algorithm types.
  
  
• Data conditioning tools or processes.
  
  
• Optimization methods.
  
  
• Evaluation methods.
  
  
• Provisioning tools for resources.
  
  
• Tools to aid model development.
  
  
• Software and hardware for AI system design, development and deployment.

4. System and Computing Resources

• The identification process requires organization to document the system and computing resources needed to implement the AI system. Organizational management conducts reviewing and updates of system and computing resources on an annual basis or through assessments prompted by technological environment changes.
  
  
• The location for system resources along with cloud or edge computing and physical facilities determine the deployment site.
  
  
• Processing resources include network structure and storage functionality among others.
  
  
• The environmental effect of using specific hardware components for AI system processing includes both manufacturing-related environmental impact and hardware operational expense.
  
  
• The organization needs to recognize that different types of resources will help improve AI systems in a continuous manner. The system requirements during development, deployment phase and operational phase should be evaluated separately.

5. Human Resources

The organization has a responsibility to identify all human resources who will work on AI system development alongside their required competencies from deployment through operation to maintenance and verification and integration and decommissioning and transfer. The human resource documentation undergoes periodic evaluations for updates which occur annually as well as when technological or organizational alterations take effect.

Organization needs to assess role requirements for the system and include jobs from different professional domains. Organization needs to add specific demographic groups to system design if they play a required role in training machine learning models using data sets. 

The required human resources to execute such projects may consist of the following list:

• Data scientists.
  
  
• Organization needs professionals who will perform human supervision duties for AI systems.
  
  
• Experts on trustworthiness topics such as safety, security and privacy.
  
  
• AI researchers and specialists, and domain experts relevant to the AI systems. Throughout the AI system life cycle different types of resources need to be employed according to specific stages.
  

Procedure Review and Continuous Improvement

Management must review this procedure on a yearly basis to verify whether it maintains its suitability together with necessary adequacy and effectiveness.

The management review will include:

• The status of actions from previous management reviews
  
  
• The management review should analyze existing changes regarding external and internal elements relevant to the AI management system.
  
  
• The review evaluates alterations in the needs and expectations of relevant interested parties regarding the AI management system.

Information about performance outcomes of the AI management system needs evaluation including observational trends from these categories:

1) Nonconformities and corrective actions

2) Monitoring and measurement results

3) Audit results

4) Opportunities for continual improvement.

The results from the management review will decide both the necessary improvements and modifications to the AI management system. Regular AI Ethics and Governance Committee meetings ensure proper conformity of our AI management system through AI risk management practices as well as audit result follow-up processes and continual improvement of its suitability, adequacy and effectiveness.

Conclusion

The AI Resource Management Policy enables organizations to give proper attention to resource allocation that benefits functional AI management systems through skilled personnel and data usage as well as technological infrastructure. The policy enables both safe and reliable AI systems by creating an environment which combines innovation with regulatory conformity.