Download

Download eBook for Free

Full Document

FormatFile SizeNotes
PDF file 1.5 MB

Use Adobe Acrobat Reader version 10 or higher for the best experience.

Research Summary

FormatFile SizeNotes
PDF file 0.1 MB

Use Adobe Acrobat Reader version 10 or higher for the best experience.

Purchase

Purchase Print Copy

 Format Price
Add to Cart Paperback34 pages $21.00

Research Questions

  1. What warfighting applications could be used as potential use cases?
  2. What type of data is needed to train and test AI systems?
  3. What are the limitations of AI algorithms?

The U.S. Air Force is increasingly interested in the potential for artificial intelligence (AI) to enhance various aspects of warfighting. For this project, the Air Force asked RAND Corporation researchers to consider instead what AI cannot do in order to understand the limits of AI for warfighting applications.

Rather than attempting to determine the limits of AI in general, the researchers selected and investigated four specific warfighting applications as potential use cases: cybersecurity, predictive maintenance, wargames, and mission planning. These applications were chosen to represent a variety of possible uses while highlighting different constraints. AI experiments were performed for the three cases for which sufficient data could be obtained; the remaining case, wargames, explored broadly how AI could or could not be applied.

This report is the first in a five-volume series and summarizes the findings and recommendations from all use cases. It is aimed at policymakers, acquisition professionals, and those with a general interest in the application of AI to warfighting.

Key Findings

  • To recognize adaptive threats, data must be recent. Distributional shift degrades model performance, and it cannot be avoided, especially for high-dimensional data.
  • AI classification algorithms cannot be relied on to learn what they are not taught. AI did not anticipate or recognize new kinds of cyberattacks.
  • Data must be accessible and well-conditioned. Relevant logistics data are maintained in multiple databases and are often ill-conditioned. Without an automated data pipeline, sufficient data cannot be captured to enable AI.
  • Peacetime data cannot be substituted for wartime data. AI cannot make up for a scarcity of appropriate data.
  • Digitization must precede AI development. Most wargames are not conducted in a digital environment and do not generate electronic data. Digitization is a precursor to an AI data pipeline.
  • New kinds of data are needed. To enable AI, human-computer interaction (HCI) technology is needed to capture aspects of wargaming that are not captured today.
  • AI is far from achieving human-level intelligence. Therefore, it cannot stand in for humans, nor can it apply human judgments.
  • To counter adaptive threats, data must be recent. Models must be refreshed with updated conditions to survive against dynamic threats.
  • AI is tactically brilliant but strategically naive. It tends to win by getting within the opponent's observe, orient, decide, act loop rather than by coming up with a clever grand strategy.
  • AI is less accurate than traditional optimization methods. But its solutions can be more robust, and it can reach them faster.

Recommendations

  • The Department of the Air Force (DAF) should perform dataset segmentation tests to determine the significance of distributional shift for AI systems and to determine an approximate decay rate and AI shelf life.
  • The DAF should experiment with AI to improve demand forecasting for readiness spares packages (RSPs) and extend the proof-of-concept models to all aircraft. This will likely have to be done on a part-by-part, platform-by-platform basis.
  • The DAF should consider using AI to solve the larger operations research problem of selecting which parts to send where.
  • The DAF should build a data operations pipeline to conduct a retrospective analysis of aircraft maintenance and RSP efficiently for multiple parts and platforms.
  • The DAF should concentrate resources for developing AI applications for wargames on the most promising areas: those that investigate alternative conditions or that are used for evaluation with well-defined criteria; those that already incorporate digital infrastructure, including HCI technologies; and those that are regularly repeated.
  • The DAF should increase the use of digital gaming infrastructure and HCI technologies, especially in games designed for systems exploration and innovation, to gather data to support AI development.
  • The DAF should employ AI capabilities to support future wargaming efforts more generally.
  • The DAF should consider how AI could power a fast-reaction policy for drones facing unexpected conditions.
  • The DAF should invest in developing tools to apply reinforcement learning to existing mission planning models and in simulations, such as the Advanced Framework for Simulation, Integration, and Modeling (AFSIM).

Research conducted by

This research was prepared for the Department of the Air Force and conducted within the Force Modernization and Employment Program of RAND Project AIR FORCE.

This report is part of the RAND research report series. RAND reports present research findings and objective analysis that address the challenges facing the public and private sectors. All RAND reports undergo rigorous peer review to ensure high standards for research quality and objectivity.

This document and trademark(s) contained herein are protected by law. This representation of RAND intellectual property is provided for noncommercial use only. Unauthorized posting of this publication online is prohibited; linking directly to this product page is encouraged. Permission is required from RAND to reproduce, or reuse in another form, any of its research documents for commercial purposes. For information on reprint and reuse permissions, please visit www.rand.org/pubs/permissions.

RAND is a nonprofit institution that helps improve policy and decisionmaking through research and analysis. RAND's publications do not necessarily reflect the opinions of its research clients and sponsors.