Abstract:
|
The horizon of AI-enabled systems is near, but current test and evaluation practices within the Department of Defense are likely to mischaracterize performance, risk, and uncertainty. One of several key test and evaluation certification challenges will be interpolating between and extrapolating beyond our test points, especially in black-box systems. In this briefing, we present a conceptual framework to overcome these challenges, paying particular attention the role of experimental design and statistical analysis through the test and evaluation lifecycle.
|