Systems development: Deployment and implementation

This sub-guideline is part of the guideline Systems development. It can be read in conjunction with the sub-guideline Systems development: Deployment patterns. Refer to the main guideline for context and an overview.

This sub-guideline focuses on the deployment and implementation of AI systems and tools within parliament. It provides essential insights for IT professionals and decision makers involved in integrating AI solutions into parliamentary operations. 

This sub-guideline covers key aspects such as deployment strategies, common deployment cases and critical planning recommendations. By addressing topics like stakeholder engagement, pilot project initiation and the use of agile methods, this guideline aims to support parliaments in effectively and responsibly implementing AI technologies to enhance legislative processes, improve efficiency and maintain transparency. It emphasizes the importance of understanding parliamentary workflows, building internal expertise and leveraging responsible AI tools throughout the implementation process.

A structured and coordinated approach to AI systems deployment is essential to the effective integration of these systems into various parliamentary applications and workflows. These patterns of deployment reinforce good practices, helping to ensure that AI deployments are scalable, robust and maintainable. For further discussion of software deployment patterns, refer to the sub-guideline Systems development: Deployment patterns.

Parliament’s deployment strategy will depend on the degree of task automation. The various options are discussed below:

• Human-only: In this case, there is no automation. The task is carried out manually by users without AI support.
• AI assistance: The task is performed mainly by users, possibly with assistance and support from the AI system.
• Partial automation: The task is performed mainly by the AI system, which produces suggestions for users.
• Full automation: The task is performed entirely by the AI system, without human intervention.

Both AI assistance and partial automation are examples of “human-in-the-loop” deployments (for further discussion of this topic, refer to the sub-guideline Ethical principles: Human autonomy and oversight).

Deployment cases

The most common deployment cases and their characteristics are detailed below:

• New product or feature: In this case, a new AI-based product or feature is introduced.
• Partial task automation: In this case, a task was previously done manually and an AI algorithm is introduced to either automate this task or assist the user.
• Replacement of a previous AI system: In this case, a task was carried out via a previous implementation of an AI system, and another AI system is introduced to replace the previous one with a view to improving quality and/or execution time.

In the above deployment cases, parliament should consider the following two basic aspects:

• Gradually increase traffic with monitoring. It is advisable to avoid sending a lot of production traffic to an algorithm that is still learning and is not yet fully proven. It may be better to send it only a small amount of traffic, monitor it and then progressively ramp up the amount of traffic.
• Ensure there is a possibility to roll back. It is advisable to have a contingency plan in place to revert back to the previous, stable configuration in case the new algorithm does not work as expected.

Key recommendations for planning the implementation of AI systems within a parliamentary context are outlined below.

Understand parliamentary processes

• Conduct a comprehensive analysis of existing parliamentary workflows and challenges.
• Examine key processes such as legislative drafting, committee meetings, voting procedures and public consultations.
• Map out the flow of documents, communications and decisions within parliament.
• Understand how information is processed and identify things such as pain points, redundancies and inefficiencies in current processes.
• Understand the current state of digital systems, databases and communication tools used for parliamentary operations.
• Identify where AI can be integrated to enhance current systems or replace outdated technologies.
• Identify and develop potential use cases to improve understanding of the potential for AI solutions at both technical and business levels.
• Ensure that any AI implementation complies with existing laws and with parliament’s rules (or identify where changes to these rules are needed).

Engage stakeholders

• Ensure that the following stakeholders are represented and involved throughout the entire AI system life cycle:
  • Business experts
  • MPs, clerks and administrative staff, who will use the AI system for legislative activities
  • Legal experts, who will ensure that the AI implementation adheres to relevant laws
  • Citizens, who need to be informed and consulted about AI initiatives in parliament
  • External experts such as AI researchers or data scientists, who can provide insights into AI implementation
• Regularly solicit feedback from stakeholders at all stages of the AI project and incorporate this feedback into the AI development process.
• Keep stakeholders informed about the progress of AI projects, milestones achieved, and any changes or updates, through regular reports and presentations.

Start with a pilot project

• Initiate small-scale pilot projects to test AI solutions in real parliamentary settings.
• Choose pilot projects with a clear and manageable scope.
• Look for areas within parliamentary processes that can benefit from AI, such as document analysis, constituent communication or data-driven decision support.
• Assess the feasibility and potential impact of proposed pilot projects.
• Establish clear, specific objectives for each pilot project. These should be aligned with the overall goals of enhancing parliamentary efficiency, transparency and decision-making.
• Develop success criteria to measure the effectiveness of the pilot projects. These criteria could include performance improvements, AI output accuracy, user satisfaction and cost savings.
• Create a detailed project plan that outlines the steps, timelines, resources and responsibilities for the pilot project.
• Identify potential risks and develop mitigation strategies.

Use agile methods

• Plan the activities in sprints, each lasting a couple of weeks.
• Involve business experts in each phase of the project. Doing so offers numerous benefits including a faster response to legislative changes, better cross-team collaboration, higher user satisfaction, reduced project risks and increased transparency for stakeholders. 

Build internal AI expertise

• Invest in developing AI expertise among parliamentary staff.
• Provide training and resources to help staff understand, evaluate and effectively use AI technologies.

Collaborate with AI experts and share with other parliaments

• Collaborate with AI experts, researchers and technology providers to build knowledge and experience.
• Collaborate with academic institutions, research organizations and other parliaments to share and learn, and to access cutting-edge AI research and innovations.

Leverage responsible AI tools

Consider using the recommended tools detailed in the Inter-American Development Bank publication Responsible use of AI for public policy: Data science toolkit:

• Robust and Responsible AI Checklist: This tool consolidates the main concerns by stage of the AI life cycle. The checklist must be reviewed continuously by technical teams and decision makers.
• Data Profile: This tool is an initial exploratory analysis conducted during the data-collection and processing stage of the AI life cycle. It provides information to reassess the quality, completeness, temporality and consistency of the training data set, possible biases within this data set, and the implications of the use of an AI system, including potential harm. 
• Model Card: This tool summarizes the main features of the AI system, highlighting the main assumptions, the most important characteristics of the system, and the risk mitigation measures implemented. 

The Guidelines for AI in parliaments are published by the IPU in collaboration with the Parliamentary Data Science Hub in the IPU’s Centre for Innovation in Parliament. This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International licence. It may be freely shared and reused with acknowledgement of the IPU. For more information about the IPU’s work on artificial intelligence, please visit www.ipu.org/AI or contact [email protected].