Generative AI is everywhere, but few places are as challenging as the public sector. We’re working with SAS to explore the opportunities, challenges and risks.

As government agencies and organizations continue to leverage the potential of AI and data analytics, it is crucial to explore the different tools and technologies. During SAS Innovate on Tour in Rotterdam, we spoke to Michel Philippens, Analytics Advisor and Head of Customer Solutions, about this.

He takes a deeper look at the challenges and opportunities of implementing AI in the public sector. From limiting the risks of AI to modernizing services, everything is covered. First, he wants to start with the basics.

Opportunities

Generative AI (GenAI), which can learn from various forms of data such as text, images, audio and 3D models, can generate new outputs such as conversations, videos, images and 3D systems. According to Philippens, there are four opportunities for GenAI in the public sector:

• question and answer: Improving customer and client services through chatbots that can answer questions and customize services.
• Content creation: Generate different types of content, including emails, social media posts, contracts and offers.
• Content summary and information retrieval: Extracting relevant insights from large bodies of knowledge.
• Coding, Software and Data Science: Improve productivity in software development by automating coding and testing.

The above topics are already being fully embraced by organizations today. Philippens sees many opportunities for the public sector. “Take regulations and laws: you can save a lot of time in drafting and analyzing regulations. It is also possible to improve the development and implementation of policies and programs.”

“The list goes on: tools for better governance and administration, improving the efficiency and quality of public services, streamlining internal processes through automation.”

Rules of the game

In theory, we can easily follow Philippens’ report, but the government usually has stricter rules when it comes to personal data. His talk addresses the main challenges and barriers. “The potential is enormous, but there are some problems that hinder the implementation of GenAI.”

He lists the main challenges:

• Privacy and security: Risks of data manipulation, theft and creation of false information.
• accuracy: Challenges in ensuring the factual accuracy and contextual relevance of AI results.
• Costs and management: High computational effort, changing business models and the need for management frameworks.
• Legal and ethical concerns: Topics such as AI attacks via instant injection, legal liability and compliance with regulations such as the EU AI Law.

SAS highlights some important things to implement generative AI effectively. Below we list the most important ones:

• Determine your risk attitude: Assess the potential risks and establish a clear risk management strategy.
• Identify and prioritize use cases: Focus on high-impact areas where GenAI can add significant value.
• Develop GenAI apps with end users: Involve end users in the development process to ensure that the tools meet their needs.
• Keep people moving: Provide human oversight to maintain control and accountability.
• Communication and training plan: Develop a comprehensive plan to educate stakeholders and users about GenAI.
• Select suitable LLMs: Choose appropriate models that meet the needs and constraints of the organization.
• Improve infrastructure and data capabilities: Providing the necessary infrastructure and data management practices.
• Develop new skills and roles: Develop new competencies and roles within the organization to support GenAI initiatives.
• Start small and grow: Start with pilot projects and expand gradually as the organization builds experience and confidence.

Practical example

The theory is nice, but fortunately Philippens also included a clear practical example in his presentation. “Legal experts in government often have to search through thousands of pages of laws to find relevant paragraphs. In addition, experts also have to take previous interpretations into account.”

In a demo, we see SAS inserting the PDF document of the EU AI law into an LLM. The model shows the answer by asking a very specific question (in collaboration with the specialist Notilyze). The result is a clear answer with additional information. This allows you to search specifically for confirmations, which saves a lot of time.

“Large language models should help governments extract their own data. They can then conduct further investigations under guidance.”

Diploma

While generative AI shows promise for transforming public sector operations, successful implementation requires careful consideration of potential risks, strategic planning, and ongoing human oversight.

By following the best practices described, government agencies can use GenAI to increase efficiency, improve services, and support better decision-making.

This is an editorial contribution in collaboration with SAS. For more information about SAS AI solutions, click here Here rightly so.