Speed vs. Accuracy
In a recent Air Force experiment called DASH-2, AI algorithms generated attack plans at a staggering rate, outpacing human staff by a factor of 400. However, the increased speed came with a caveat: not all AI-generated plans were viable.
DASH-2 Results
Maj. Gen. Robert Claude explained that AI tools produced 10 „Courses Of Action” (COAs) in approximately eight seconds, compared to the three COAs generated by human staff in about 16 minutes. While the AI demonstrated remarkable speed, the plans were not without flaws.
Subtle Errors
The errors were not glaring but rather subtle, such as failing to account for the appropriate sensor for specific weather conditions. These subtle errors highlight the importance of human expertise in identifying and correcting mistakes.
Human-in-the-Loop
Claude emphasized the need for human oversight in the foreseeable future to ensure the viability of AI-generated plans and make informed decisions. While AI can significantly accelerate the planning process, human judgment remains essential.
Algorithm Refinement
Claude expressed confidence that future iterations of AI planning aides can reduce the error rate through algorithm refinement. He acknowledged that the limited two-week sprint for DASH-2 did not allow for the inclusion of all necessary checks and balances.
Future Development
Claude indicated that a more extensive development period would be required to create a reliable AI planning system. The focus of DASH-2 was on exploring the potential of AI, not on deploying a fully functional system.
DASH-3 Underway
The third and final DASH experiment of the year is currently taking place at the Shadow Operations Center — Nellis in Las Vegas. Claude shared his observations from the beginning of DASH-3, noting the overwhelming amount of information that Air Force planners must process.
Human-Machine Teaming
Claude highlighted the potential value of a successful human-machine team arrangement, where AI assists humans in making better-informed decisions.
