6 February 2021

The Next Steps For the Pentagon's AI Hub

By CHRIS BASSLER and BRYAN DURKEE

As the two-year-old Joint Artificial Intelligence Center shifts from a projects-and-products shop to the Pentagon’s hub for AI services and support, its leaders are working on priorities for “JAIC 2.0.” We suggest the center focus on six main efforts.

First, accelerate efforts to develop and deploy AI for back-office applications in task management, automated reporting and correspondence, HR, legal, security, budgeting, finance, contracts, and logistics. Automating grunt work in these fields would free up staff and support personnel for debate, analysis, and critical thinking. It will also save time, increase accuracy, and allow deeper analysis. For instance, an entirely digital POM build would enable more rigorous and quick excursions, and streamline “what if” drills and responses to the White House and Congress. As these efforts are deployed, the JAIC should build a repository of lessons and limitations and serve as a clearing house and strategic advisor for the services branches and DoD, whose back-office leadership and staff are currently not well-versed in AI.

Second, train program managers and acquisition program offices to be AI-literate — that is, smarter and more effective procurers of AI products and services. The JAIC should create a guidebook telling what to do, what to avoid, and what AI approaches are best suited to which applications. This would help acquisition program offices establish technical and contracting best practices and standards, and reduce the time and money spent on underdeveloped, unnecessary, or duplicative AI programs.

Third, lead the understanding and monitoring of, and connection to, AI talent hubs — within the U.S. (not just Silicon Valley) and in key ally and partner countries. The JAIC should also help service labs maximize the use of the Engineer/Science Exchange Personnel program, which sends U.S. talent to foreign military labs and vice versa.

Fourth, act as a focal point for service, interagency, and international sharing of AI approaches, algorithms, and lessons. The hub of this work should be the curation of a “one-stop-shop” repository for DoD’s R&D, acquisition, and operational communities. But the JAIC should expand into helping combatant commands to understand the use and limitations of various technical and programmatic approaches and to share work rather than recreate it. It should also include efforts to link technologists and operators — for example, by helping services’ science & technology communities better understand COCOMs’ operational requirements, and by teaching COCOMs to ready their data for applications.

Fifth, lead a handful of AI project “big bets.” JAIC should leave most AI technical work to the services’ S&T organizations, national labs, and — via funding — to industry and academia, but it should tackle the hardest problems itself — ones cannot be successfully handled elsewhere in the DoD. These should be warfighting problems that span organization control, require the revision or elimination of policy barriers, and push the technological boundaries of AI. An example might be creating multi-domain kill webs for priority missions.

Sixth, lead experimentation and testing of adversarial machine learning techniques to spoof or break AI algorithms. The JAIC could replicate and anticipate adversarial machine learning approaches, so the services could practice and improve their approaches for resiliency and defense against them. It should maintain a virtual sandbox for experimentation at all security levels. And it should act as a Red team, seeking to detect and exploit weaknesses in service AI approaches. In addition, the JAIC could maintain an AI scorecard for each service, which would help push their organizations towards more rapid AI implementation and an iterative security assessment process.

These six efforts would maximize JAIC 2.0’s value to DoD, and free critical resources for further reinvestment in AI-enabled Joint Force warfighting. By doing so, the JAIC will accelerate every aspect of the AI development and fielding pipeline within the Department, scale AI implementation, and deepen positive impacts across all DoD ecosystems

No comments: