Getting started with AI: Tips from parliamentary experts

In early December 2024, the Centre for Innovation Parliament launched the new Guidelines for AI in Parliaments and a comprehensive set of use cases. Developed by parliaments for parliaments, the Guidelines mark an important milestone in the adoption of artificial intelligence (AI), supporting parliaments as they embark on their own journey towards using AI. While we expect parliaments to map out their own journey – deciding how much AI to use and how quickly – the Guidelines provide support for that process and can be adapted to suit individual needs.

In this article, we share some of the key lessons and advice from the parliamentary experts who spoke at the launch event for the Guidelines, all of whom made it clear that implementing AI successfully was a journey that required careful planning and execution. The collective experience of parliamentary early adopters points to some clear principles: 

• Start small and experiment.
• Focus on solving real parliamentary problems.
• Ensure good data practices.
• Maintain a clear strategic vision.

Most importantly, the speakers emphasized that AI should serve parliament’s fundamental mission of effective democratic representation and oversight.

Start small, build incrementally

The Chamber of Deputies of Chile has adopted an agile approach to developing parliamentary applications, which has stood it in good stead when it comes to AI. Mr. Esteban Sanchez described how this approach allowed the chamber to prototype ideas and develop them quickly, with the close involvement of users. Lessons are learned and systems modified as they evolve. 

The Chamber of Deputies of Italy is working in a similar way. Mr. Giovanni Ciccone stressed that it was important to follow an iterative approach, starting with small projects as proofs of concept. He suggested taking a “lab-oriented approach”, emphasizing the importance of regular check-ins with users right across parliament and of building cross-functional teams to experiment with AI solutions.

Use the right tools

While much of the talk is about generative AI, Mr. Ciccone pointed out that many parliamentary problems can be solved with discriminatory models such as regression models and classification models. In cases like this, the knowledge already exists and parliaments do not need to reinvent the wheel. They can adopt existing frameworks and technology solutions to create pilots. These projects can then be iteratively refined and the quality improved, leading to better end products. Again, this approach reflects the principle of starting small and experimenting, as well as the importance of understanding where and how AI can help.

AI needs good business processes

The parliaments we have spoken to stress that strong collaboration between IT and business experts is vital. Well-defined and clear business processes are an important starting point for AI adoption – parliaments must know what business problem they are trying to solve and understand how this process can be improved by AI.

AI feeds on data

At its heart, AI needs data to power such things as large language models and to make sense of the world. Just as a good understanding of business processes will help parliaments to design effective AI-driven solutions, so strong data management practices will enable AI systems to function better. Mr. Ricardo Vilarins, from the Chamber of Deputies of Brazil, said:

“Parliaments need to know how to prepare users to provide good-quality data.”

Parliaments need strong policies for data governance. All stakeholders need to be aware of the data they manage and how they can continuously improve data quality. This is where having good practices and policies helps, as they can guide users towards a culture of strong data management. 

Data literacy is important right across parliament

In Brazil, the Chamber of Deputies is working on projects to increase data literacy and knowledge of data-related issues. As well as informing and educating people, this approach also helps them feel a part of the data governance process. As this process develops, parliamentary staff come to better understand how good data quality leads to more effective AI usage.

Strategy is important

One benefit of good data literacy is that it increases awareness of the inter-connectedness of data, processes, people and culture. This, in turn, helps to ensure that there is a stronger strategic focus on AI adoption. 

This growing organizational awareness reinforces the point made by Mr. Fotis Fitsilis of the Hellenic Parliament of Greece about the need for a broader strategic framework. While starting small and experimenting are crucial tactics, Mr. Fitsilis proposed that these efforts should be guided by clear strategic direction:

“Strategy comes up front. Without strategy you go nowhere.”

Having a strategy for AI and – as the Guidelines themselves make clear – good governance practices around its use means that there are clear parameters for using AI, rules and regulations to manage the risks, and a strong sense of what success looks like. Mr. Fitsilis explained his view that knowledge-sharing was a key part of this process: 

“You can have your own experts. But you should consult with the specialists at the Centre for Innovation in Parliament (CIP) to understand how to build your own strategy. You can start with your own expertise but you must definitely rely on expertise from outside parliament.”

The goal is to make parliament more effective

According to Mr. Michael Evraire, the number-one lesson for the House of Commons of Canada, an early adopter of AI systems, was as follows:

“AI is not the goal in itself. At the end of the day, AI should be a means to enable strategic alignment with what you’re already trying to accomplish in your parliament.”

Just like any technology, AI is there to support the institution of parliament in working more effectively and efficiently. It is only ever a tool and should never be the starting point for a project. Mr. Evraire reinforced the point made by others that data quality mattered and that building a high-quality data set over time would make AI more effective:

“You need the data first before AI can realize any benefits. So don’t think of AI in terms of looking ahead to the next project. That’s not the best approach. Think, instead, about how AI can help you realize your business objectives.”