ship between funding and implemen- tation. There is much that we do not yet know about AI’s capabilities or how to implement them optimally. Despite these challenges to gaug- ing the contribution of AI to produc- tivity, the DoW is making significant AI investments. These investments are not aimed simply at cutting costs and reducing the denominator of the ROI ratio. They are made to increase capabilities that will allow us to over- match our adversaries, who are also making extremely large investments in AI. So, we are involved in an AI arms race. These investments in improving our capabilities are a value play that requires a useful numerator estimate. Capabilities find their way into core processes where the AI is actually used for targeting, signal intelligence interpretation and detection, cy- bersecurity, and other core military processes. For this reason, we need a value metric that can help make a robust and convincing argument in forecasting AI’s future value. Unless value can be measured in common units—e.g., common units of value in terms of capability or readiness—we will have a very difficult challenge in convincing skeptical lawmakers, executives, and generals that the AI Various methods are available for measuring the impact of processes as they transform a set of inputs into a set of outputs, from raw materials to a finished product. A technique called knowledge value-added (KVA) uses Kolmogorov complexity, or K- complexity, originally based on a measure of the length of a computer program to produce a particular out- put. For example, outputting a string of eight zeroes (00000000) could be simplified by telling the program to print “0” eight times, so it has a low K-complexity. However, an output of random alphanumeric characters (12CGTj52) has a higher K-complexity because the program used to define investments are a good bet. Complexity and Value
a series of random characters would be longer. KVA assumes that value is added when inputs are changed into out- puts. In lieu of measuring the length of computer programs needed to pro- duce a particular output, KVA mea- sures complexity by the time it takes a notional average employee, given the necessary inputs, to learn how to pro- duce the specified outputs. The gen- eral idea is that the more complex the process, the longer it takes to learn. Therefore, learning time provides an estimate of the complexity—i.e., inputs being transformed into out- puts. This enables the comparison of various processes, such as gathering signal intelligence or planning a ship alteration. The DoW has applied KVA to at least 45 areas within the depart- ment, including flight scheduling and ship maintenance. The application of KVA to ques- tions of AI implementation is straight- forward. It is simply a matter of de- termining the difference between the time it would take a civilian or military employee to learn how to produce
Unless value can be measured in common units— e.g., common units of value in terms of capability or readiness— we will have a very difficult challenge in convincing skeptical lawmakers, executives, and generals that the AI investments are a good bet.
November-December 2025 | DEFENSE ACQUISITION | 15
Made with FlippingBook - Online Brochure Maker