Which steps to take if you want to regard fairness in your AI system development

Technical management processes

Introduction

The current process model was developed by the research consortium fAIr by design. It is intended for anybody who is interested in improving fairness of AI systems. It offers support in developing, deploying and maintaining fair AI systems and can be used in different contexts and for different technologies and uses. Adhering to the process model should offer help in being prepared for future standards and regulations, as well as facilitating eventual third-party audits; but it does not give any guarantee of compliance.

The process model includes steps for all stages of AI system development and deployment, for an interdisciplinary teamThe Technical Management Processes are used to manage and control the system development and the allocated resources.  The technical management processes can be thought of as “horizontal”, spanning the entire system development life cycle, and ensuring the system satisfies objectives at all stages of its life cycle. They form the basis for the technical processes and enable an efficient workflow and beneficial framework conditions (such as terms of costs, planning of the project and the overall management of the project and team). For easier orientation the Technical Management Processes are identical to the technical management processes identified in ISO 15288 and ISO 5338.

Please note, that the process model is regularly updated.

  • During the project planning process, the project is coordinated and managed. This step enables fulfilling the tasks, deliverables and outputs and includes the definition of achievement criteria and required resources. This process also includes adoption of a data governance plan to control data collection, data management, and data disposal processes for the various types of data that can arise during the AI system life cycle: training, validation and test data, production data, and logged data, as necessary. As projects change over time and therefore plans need revision, this is an ongoing process throughout the complete project.

  • Since the project plan forms the basis for all project-related activities, it is vital that fairness measures are built into the project plan, and that fairness objectives are integrated into all considerations of the project planning process.

Project assessment and control process

  • In the project assessment and control process plans, project status and process performance are assessed regularly and if required regarding the requirements and objectives. This includes information to the relevant project stakeholders for potential redirection, which may require re-planning.

  • Ensuring fairness throughout the AI system life cycle involves constant monitoring of implemented measures and resources, to ensure that they are effective in achieving the project’s fairness objectives.

Decision management process

  • In the decision management process a structured framework is developed to identify and evaluate alternatives for decisions and select the most beneficial actions. It is used to resolve any issues during the complete system lifecycle.

  • The decision management process helps make fairness objectives actionable. This involves identifying when fairness aspects require decisions to be made, obtaining support from relevant stakeholders in establishing appropriate fairness criteria, setting thresholds for fairness relevant measurements, and deciding upon the course of action when fairness thresholds or criteria are not met.

Risk management process

  • In the risk management process risks are identified, analyzed, treated and continuously monitored along the complete lifecycle of the AI system.

  • This process is responsible for identifying, monitoring and treating risks of unwanted bias, unfairness and discrimination. Regarding fairness there is a special emphasis on the continuous and iterative execution of this process, as fairness perceptions and circumstances change over time. This also includes opportunities for improving fairness.

Configuration management process

  • In the configuration management system elements and configurations get managed and controlled regarding consistency, requirements and criteria.

  • This process is essential to the overall quality, reproducibility and transparency of the complete life cycle of the system. From a fairness perspective, it is important that fairness requirements and criteria, as well as test data, test methods and results, are included as configuration elements / items and can be specifically reviewed and audited for fairness.

Information management process

  • In the information management process information is generated, retrieved, processed and disseminated to designated stakeholders.

  • This process needs to include information relating to fairness objectives – such as stakeholder needs, fairness criteria, selected metrics and thresholds, as well as risk incidents and changes, and make this information available to the appropriate stakeholders, as needed.

Measurement process

  • In the measurement process data and information to support management and demonstrate quality of the AI system and process is collected and analyzed.

  • The correct measurement and analysis of the AI system is integral to the risk management and quality of the finished product – including of course the fairness aspects, and the ability to detect and mitigate risks of bias or discrimination.

Quality assurance process

  • In the quality assurance process the project life cycle processes and outputs are analysed, and it is ensured that quality requirements and project policies are met.

  • The quality assurance process builds on objectives and requirements established in the project plan and risk management processes. In this sense, there is nothing “extra” to do for fairness, as these should already be included in the quality criteria. However, the importance of the quality assurance activities is so relevant to the achievement of fairness objectives, that they bear highlighting.

Project planning process