How government organizations can get started with generative AI today

A public defender absorbs essential details from thousands of court records as he hurries to a new client’s arraignment. A child welfare worker digests the history of a complex case in minutes, allowing her to better serve the child and their family quickly. And with a single query, a worker instantly sees what his Social Security benefits would be if he retired at 65 versus 70. 

These scenarios and countless others are not far from becoming reality thanks to the power of generative AI. The technology — which produces insights in natural language based on learning and analyzing patterns from massive amounts of data — offers a vast opportunity for public sector organizations to remove friction from their operations, capture institutional knowledge and provide a better experience for both employees and constituents. 

AI’s potential in government is significant – but implementing AI safely and effectively can be a slow process. However, governments can take steps today to lay the foundation for deploying generative AI while making progress toward their digital transformation and modernization goals. 

Roadblocks to AI adoption

Despite the promise of generative AI, government organizations face several roadblocks to rapid implementation. Key among these are concerns about trust and security. Public agencies, which often deal with sensitive information and face legal restrictions on sharing data, need to ensure that confidential information doesn’t get exposed in open large language models. Some generative AI technologies have also been known to propagate bias, toxicity and false information. Given the critical public-facing work governments do, it’s crucial that their generative AI models produce fair, accurate and transparent results. 

Another challenge is that government data is often housed in disparate, archaic systems. Generative AI is only as good as the quality of data used to train it. Public agencies have typically pieced together technology solutions as needs emerged and don’t have their data quickly accessible or harmonized. Generative AI models can’t function unless they can access the right stores of data in the right formats. 

Finally, cultural resistance can block rapid advances; for example, employee concerns of being displaced by technology or their expertise being commoditized. Compared to the private sector, governments face significant resource constraints and decision delays.

Paving the way

Despite these obstacles, government agencies can take steps today to lay the groundwork for generative AI.

Step 1: Embrace a data strategy

It’s hard to jump straight from today’s disconnected data environment to full-fledged generative AI adoption. Organizations need a data strategy to understand where data currently lives, how it feeds into workflows and how to access it. The next step is to implement trusted technology that can integrate data into a single source of truth through a real-time, automated platform. Having the right tool means agencies don’t need a team of data scientists to clean up their data. Even before embracing generative AI at scale, a better data strategy can go a long way toward enabling automation and a better user experience. 

Step 2: Adopt a change management plan

Generative AI shouldn’t be treated as just an IT project; rather, leaders should focus this initiative (or any tech initiative) on enhancing the organization’s mission.  From my own experience as a government CIO, I have observed that public servants are generally very passionately dedicated to their mission and those they serve. Leaders will do well to understand any employee concerns and be transparent and responsive to those concerns. Generative AI that augments and simplifies the work, leaving more time to focus on helping people, will be well received.

Training is critical as well. GenAI will have a hugely democratizing effect, making complex tasks like software development faster with fewer errors. But people need new skills–such as prompt development–to be effective. Leaders should ensure workers have the technical fluency needed to use generative AI models (free online platforms can help). Further, we expect that  training in soft skills will become even more important, such as relationship-building, active listening and collaboration.   

Step 3: Prioritize trust

Governments should establish foundational principles around data protection, privacy and ethical use to guide generative AI development. They need to ensure that models are purpose-built for their use cases and grounded in robust internal data. And they need to verify that models operate in an unbiased and compliant manner. Agencies don’t need to reinvent the wheel here –they can learn best practices from other sectors, collaborate with experts and work with technology vendors who build trust and security into their products.

Step 4: Start small and iterate

Government organizations don’t need to perfect their data strategy or solve every security concern before getting started with generative AI. Plenty of use cases exist that involve less sensitive data but have immediate impact. Many of these involve back-office or administrative processes, such as travel approvals and employee onboarding. Moving quickly allows governments to benefit from AI in the near term while gaining expertise they can apply to more complex and sensitive workflows. 

More effective government, today and tomorrow

Governments provide critical services, from disaster response to child nutrition to small business loans. At a time when most aspects of our lives have been digitized, it’s time for governments to modernize, too. Generative AI can be a key enabler for the public sector to provide a faster, higher-quality and more engaging experience for employees and constituents. It may take time to fully embrace generative AI in a trusted and transparent manner, but that shouldn’t stop organizations from taking action now. By making moves that pave the way toward generative AI, governments can start working in a smarter, more automated fashion today.