How generative AI can unlock value in federal data

Data is intrinsic to every federal agency’s mission and can be one of an agency’s most valuable assets. To realize its full potential, data must be made accessible in a way that allows its value to be unlocked. Yet, recent Accenture research found that fewer than one in five companies are adept at unlocking the full value of their data.

Recent improvements in generative AI have made data utilization more accessible. With the rise of these new models, federal agencies can access and discover decades of embedded value in their data, helping them more seamlessly scale their data programs to drive actionable insight and support more agile, advanced decision-making.

Agencies will need to examine their readiness to embrace this technology – culturally, technologically, and for enterprise or mission use. By developing use cases and exploring prototypes, agencies can drive the internal data-led transformation required to harness this technology at scale.

Generative AI’s benefits

1. Generative AI allows for a holistic approach to data utilization

Recent advances in AI have developed a new tier of foundation models – generative neural networks trained on massive datasets that can perform a diverse set of tasks. The generality of these models allows more effective data utilization to occur regardless of type, structure, or scale of the data.

Previous methods for using diverse data required a universal interoperability layer, or a standard format for data in order to be utilized with a specific approach.

Large datasets can be explored and used with less preprocessing, which is time-consuming and resource-intensive when performed at scale.

2. Foundation models will be the basis for diverse data illumination approaches

The large datasets used in the training of foundation models allow them to be effective few-shot learners. This means that with relatively small amounts of data, these models can proficiently perform new, custom tasks.

For federal agencies with specialized use cases, low-effort learning will be an important feature to fully extract value from data.

Without the use of foundation models, agencies requiring custom models would need to provide relatively large, labeled datasets to effectively train models for specific use cases. This is a time-intensive exercise which needs to be repeated for each additional task.

Foundation models will lower the level of effort for agencies that need to perform a range of tasks applied across diverse datasets.

Federal agencies will need to balance innovation and affordability

While foundation models are a disruptive technology in terms of their potential compared with predecessors, running them in secure environments suitable for federal use cases can be cost- and resource-prohibitive. Agencies can balance innovation and affordability by exploring open-source models whose performance is effective on specific, focused use cases.

Agencies can also consider taking the following steps for generative AI-based data utilization:

• Evaluate security considerations to identify appropriate environments for models (ex: utilizing hosted APIs vs. local models).
• Identify candidate models and determine whether models need to be fine-tuned.
• Weigh the performance of these models against other factors such as cost and compute requirements.

Foundation models will allow federal agencies to extract value from their data like never before. Agencies can start exploring how these models will allow them to unite diverse data holdings and perform innovative tasks while balancing security, affordability, and performance.

Discover other areas generative AI will impact for federal agencies in our perspective, “Generative AI for federal agencies: Five focus areas.”