Goal: Help agencies buy Generative AI
Artificial Intelligence (AI) is one of the most profound technological shifts in a generation or more. If we learn how to harness its power correctly, AI tools could significantly strengthen how the federal government serves the public.
Seeing AI’s potential – and its risks – the president signed Executive Order 14110 on Safe, Secure, and Trustworthy Artificial Intelligence (AI EO) on October 30, 2023.
Since it was signed, there has been a lot of activity around highlighting AI use cases and increasing the AI talent and skills in the federal workforce.
I blogged about the procurement considerations it emphasized and we explored the pivotal role of the chief AI officer.
The AI EO also sparked an ongoing effort to guide responsible artificial intelligence development and deployment across the federal government.
Section 10.1(h) of the AI EO asks GSA to create a resource guide to help the acquisition community procure generative AI solutions and related specialized computing infrastructure.
In this post, I’ll describe our new Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide and highlight some of the specific content.
A Focus on Generative Artificial Intelligence
As many of you know, some of the most popular promising tools in the broader field of artificial intelligence are in the field called generative AI.
Fundamentally, generative AI tools are software. It is starting to show up in our email and word processing programs, the search engines we use every day, and the more sophisticated software that agencies rely on. These tools can be helpful for many agencies trying to automate simple tasks or solve complex problems. We’ve seen agencies using generative AI tools to write summaries of rules, create first drafts of memos, and make more helpful chatbots. And many more uses are spooling up right now including using generative AI tools to write computer code and develop new training scenarios for agency staff.
These generative AI tools are getting better and more agencies are asking their contracting officers to help procure the right solutions.
At the most basic level, because generative AI tools are software, acquiring them must follow the same acquisition policies and rules as other IT and software purchases. Contracting officers should consider cybersecurity, supply chain risk management, data governance and other standards and guidelines just as they would with other IT procurements.
At the same time, generative AI tools are unique. We are all hearing about the risks of generative AI solutions, some of which we talk about in the guide – from bias in how the systems were trained… to “hallucinations” where a generative AI tool states wrong information that it just made up.
Contracting officers play a critical role in ensuring commercial generative AI offerings conform with federal and agency guidance, laws and regulations and have the right safeguards and protections while enabling their agencies to get the most out of generative AI projects.
We put together the Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide to help contracting officers and their teams understand how to do just that.
Practical Tips for the Acquisition Community
Because the field is emerging and the use cases are diverse, it’s impossible to provide guidance that applies to every situation. So the guide offers questions that contracting officers should ask and a process to use when scoping a generative AI acquisition.
The guide also makes a few specific recommendations of other actions the acquisition workforce should take to procure generative AI solutions effectively. Many generative AI tools may already be available to agency staff in tools they use every day or through government cloud platforms they already have accounts on. And these tools may be available through professional service and system integrator contracts the agencies already have in place. In that way, the fastest acquisition may be no acquisition, or as simple as adding more “credits” to an existing cloud platform account.
Before embarking on a large scale or complex new acquisition for generative AI tools, see if there is a simpler route. Work with your agency’s chief information officer, chief artificial intelligence officer, and chief information security officer to determine what you already have in place and whether you can just use an existing solution or contract.
Here are a few other recommendations in the guide:
- Start with Your Agency’s Needs. Rather than starting with solutions and specifications, define the problem that the agency wants generative AI tools to help solve.
- Scope and Test Solutions. Given the evolving nature of most generative AI tools, it is essential for agencies to use testbeds and sandboxes to try solutions before committing to large scale buys with too many unknowns about product performance.
- Manage and Protect Data. Generative AI relies on data “inputs” to create content “outputs” so it is critical to know where data is coming from, what are its limitations and how data will be used and protected.
- Control Costs. Generative AI is very often billed like other Software as a Service so usage costs can really grow quickly if not appropriately monitored and managed.
Acquisition staff also benefit from knowing what procurement actions their agency and others have already taken. You’ll also find a searchable data dashboard to give information about recent AI-related contract actions.
Specialized Computing Infrastructure
The guide also talks about “specialized computing infrastructure” per the AI EO. Specialized computing infrastructure can be thought of as the high-performance computers, powerful chips, software, networks and resources made specifically for building, training, fine-tuning, testing, using and maintaining artificial intelligence applications. Computing infrastructure can be on-premise, cloud based or a combination of both.
While most agencies will likely access generative AI tools through the cloud, some agencies may need to build some light specialized computing infrastructure to support their specific requirements.
This is the start.
The biggest challenge to producing any sort of guidance around a technology is anticipating and accommodating change. To do it, we organized a working group, gathered input from a wide array of acquisition specialists and technical experts, and collaborated with our IT Vendor Management Office to inform and support faster, smarter IT buying decisions across the federal community. We welcome your feedback at genai@gsa.gov.
Generative AI technology will continue to evolve. The risks and benefits will shift over time. Agencies will experiment with generative AI tools. And contracting officers will play a critical role by working closely with program and IT staff to find, source and acquire the right generative AI solutions for agencies’ needs. We hope the Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide helps the acquisition community enable their agencies to start to responsibly harness the power of this promising technology and better serve the American people.