Q&A: GSA’s Chief AI Officer on the Future of Generative AI

FEDTECH: You are pursuing generative AI pilots. Can you give us examples of how you are testing and experimenting with the technology?

WHITMAN: We’re taking several approaches to assessing generative AI, ranging from what it takes to set up an environment designed to support it — from an agency perspective — to what it would look like to implement GenAI tools across the workforce. Some of these pilots are examining the infrastructure needs, others the data needs and ultimately the specific AI that supports mission delivery. The underlying goal of our pilots is to gather data that we can use in our next steps.

For example, the robotic process automation initiative looks at how AI can automate repetitive tasks within the agency. This project assesses the infrastructure requirements for deploying RPA solutions and evaluates their impact on workforce efficiency and productivity.

Likewise, GSA’s pilot of an AI chatbot for customer service investigates how AI can enhance customer interactions by providing real-time support and information.

These pilots are largely designed to provide insights into the practical aspects of implementing GenAI, from setting up the necessary governance and technical infrastructure to understanding the data and AI models that best support our mission delivery.

FEDTECH: What have you learned so far from these efforts? What are the benefits you see and the potential for this technology?

WHITMAN: While it’s still early days, we’ve learned quite a bit from our GenAI pilot projects; specifically, the importance of having the right infrastructure and high-quality data in place for AI to perform accurately and reliably.

As for benefits, one of the most notable findings we’ve seen is GenAI’s potential to increase efficiency for GSA staff by streamlining processes and reducing manual effort. We were one of the first agencies to establish a security policy that allowed staff to test public large language models. We then surveyed those users and learned a lot about what the perceived benefits and drawbacks of those tools were. Research like that plays a big role in shaping our initiatives.

We’ve also found issues with our GenAI deployment tests that really stress the importance of our team’s ability to monitor and improve upon the outputs of these systems. Ultimately, these pilots underscore a crucial point: The expertise and dedication of GSA staff are indispensable to any successful GenAI integration.

FEDTECH: You talked about using public large language models. Do you plan to build your own generative AI tools in-house, or will it be a mix?

WHITMAN: In our GenAI projects, we prioritize leveraging existing public generative AI tools to avoid reinventing the wheel. At the same time, we are committed to developing some in-house models tailored to our unique requirements.

This balance enables us to maintain control over our AI solutions while benefiting from the advancements made by the broader AI community. Moreover, we want to share whatever we can — be that our insights or customized tools and models — with other agencies so that the whole of government can work collaboratively, find efficiencies together and ultimately benefit from our collective efforts.

FEDTECH: In terms of implementation, do you have to train users or develop guidelines for usage?

WHITMAN: It’s crucial to ensure our user base fully understands the capabilities and risks associated with GenAI. To this end, we are coordinating with programs like the AI Community of Practice to provide training for our federal staff, which helps them use these tools while being aware of potential pitfalls, such as data privacy issues and the risk of biased outputs. We’re finding that practical experience with GenAI tools helps ground these abstract risks into more tangible and specific examples.

Additionally, we have developed policy and guidance for GenAI usage to ensure that we approach its use in an open, transparent and trustworthy manner. We grounded our policy and guidance in the tenets established in the president’s executive order on safe, secure, and trustworthy artificial intelligence, along with the Office of Management and Budget’s guidance memo, including ethical considerations, data management protocols and security measures. By combining training with the policy guidelines, we aim to empower staff to leverage GenAI ’s potential while mitigating the associated risks.

DISCOVER: Is the government’s ‘blueprint’ for AI enough for agencies to deploy it ethically?

FEDTECH: What plans do you have for generative AI — and AI as a whole — in the future?

WHITMAN: Looking ahead, our plans for AI center on responsibly expanding its integration across various agency functions to enhance efficiency, decision-making and service delivery.

We aim to scale successful pilots, tailor existing models to our specific mission needs and continue leveraging public AI tools.

Additionally, we plan to foster greater collaboration and knowledge-sharing across other agencies to maximize the benefits of AI innovations, including things like our GenAI procurement guidance, the Technology Transformation Services’ work on the AI talent surge, supporting the AI Center of Excellence and continuing AI-specific cohorts of the Presidential Innovation Fellows and U.S. Digital Corps.