AI and the IC: The Tangled Web of Budget and Acquisition

(Editor’s Note: This article was first published by our friends at Just Security and is the third in a series that is diving into the foundational barriers to the broad integration of AI in the IC – culture, budget, acquisition, risk, and oversight.)

OPINION — As I have written earlier, there is widespread bipartisan support for radically improving the nation’s ability to take advantage of artificial intelligence (AI). For the Intelligence Community (IC), that means using AI to more quickly, easily, and accurately analyze increasing volumes of data to produce critical foreign intelligence that can warn of and help defuse national security threats, among other things. To do that, the IC will have to partner closely with the private sector, where significant AI development occurs. But despite the billions of dollars that may ultimately flow toward this goal, there are basic hurdles the IC still must overcome to successfully transition and integrate AI into the community at speed and scale.

Among the top hurdles are the U.S. government’s slow, inflexible, and complex budget and acquisition processes. The IC’s rigid budget process follows the standard three-year cycle for the government, which means it takes years to incorporate a new program and requires confident forecasting of the future. Once a program overcomes the necessary hurdles to be included in a budget, it must follow a complex sequence of regulations to issue and manage a contract for the actual goods or services needed. These budget and acquisition processes are often considered separately as they are distinct, but I treat them together because they are closely related and inextricably intertwined in terms of the government’s purchasing of technology.

Importantly, these processes were not intended to obstruct progress; they were designed to ensure cautious and responsible spending, and for good reason. Congress, with its power of the purse, and the Office of Management and Budget (OMB), as the executive branch’s chief budget authority, have the solemn duty to ensure wise and careful use of taxpayer dollars. And their roles in this regard are vital to the U.S. government’s ability to function.

Unfortunately, despite the best of intentions, as noted by some in Congress itself, the budget process has become so “cumbersome, frustrating, and ineffective” that it has weakened the power of the purse and Congress’ capacity to govern. And when complicated acquisition processes are layered on top of the budget process, the result is a spider web of confusion and difficulty for anyone trying to navigate them.

The Need for Speed … and Flexibility and Simplicity

As currently constructed, government budget and acquisition processes cause numerous inefficiencies for the purchase of AI capabilities, negatively impacting three critical areas in particular: speed, flexibility, and simplicity. When it comes to speed and flexibility, the following difficulties jump out:

• The executive branch has a methodical and deliberate three-year budget cycle that calls for defined and steady requirements at the beginning of the cycle. Changing the requirements at any point along the way is difficult and time-consuming.
• The IC’s budgeting processes require that IC spending fit into a series of discrete sequential steps, represented by budget categories like research, development, procurement, or sustainment. Funds are not quickly or easily spent across these categories.
• Most appropriations expire at the end of each fiscal year, which means programs must develop early on, and precisely execute, detailed spending plans or lose the unspent funds at the end of one year.
• Government agencies expend significant time creating detailed Statements of Work (SOWs) that describe contract requirements. Standard contract vehicles do not support evolving requirements, and companies are evaluated over the life of the contract based on strict compliance with the original SOW created years earlier.

These rules make sense in the abstract and result from well-intentioned attempts to buy down the risk of loss or failure and promote accountability and transparency. They require the customer to know with clarity and certainty the solution it seeks in advance of investment and they narrowly limit the customer’s ability to change the plan or hastily implement it. These rules are not unreasonably problematic for the purchase of items like satellites or airplanes, the requirements for which probably should not and will not significantly change over the course of many years.

However, because AI technology is still maturing and the capabilities themselves are always adapting, developing, and adding new functionality, the rules above have become major obstacles to the quick integration of AI across the IC. First, AI requirements defined with specificity years in advance of acquisition – whether in the budget or in a statement of work – are obsolete by the time the technology is delivered. Second, as AI evolves there is often not a clear delineation between research, development, procurement, and sustainment of the technology – it continuously flows back and forth across these categories in very compressed timelines. Third, it is difficult to predict the timing of AI breakthroughs, related new requirements, and funding impacts, so money might not be spent as quickly as expected and could be lost at the end of the fiscal year. Taken together, these processes are inefficient and disruptive, cause confusion and delay, and discourage engagement from small businesses, which have neither the time nor the resources to wait years to complete a contract or to navigate laborious, uncertain processes.