The first quarter recommendations of the U.S. National Security
Commission on Artificial Intelligence are now available in StratML
format at https://stratml.us/drybridge/index.htm#NSCAI
Here are a couple of their key recommendations:
Objective 5.1-3
<https://stratml.us/carmel/iso/NSCAI20200331wStyle.xml#_8d84cea6-86a6-11ea-8400-476c1c83ea00>:
AI Collaboration Plan - Based on the assessment of allied
comparative strengths, the U.S. National Security Point of Contact
for AI should convene a multilateral working group for AI
collaboration and interoperability, beginning with the Five Eyes, to
develop a plan for deeper AI collaboration.
The plan should be published in an open, standard, machine-readable
format like StratML. New and better online services should be developed
to support more efficient and effective collaboration focusing
explicitly on the objectives set forth in the plan(s).
The allies should document their respective capabilities in performance
plans and reports in such a machine-readable format.
POCs and working groups are incapable of effectively addressing the
breadth and depth of the issues without such machine-readable data and
supporting online services. Lacking such data and services, the proper
name for their role is "scapegoat" and the outcomes will inevitably be
suboptimized.
Objective 6.4
<https://stratml.us/carmel/iso/NSCAI20200331wStyle.xml#_8d851ae6-86a6-11ea-8400-476c1c83ea00>:
Documentation - Develop Strategies for Documentation.
Both the strategies as well as documentation of the AI agents themselves
should be published in an open, standard, machine-readable format.
Otherwise there is no hope of scaling cooperative efforts to the scope
and pace of the problem.
One hopeful sign is that the word "metrics" appears 15 times in the
recommendations. On the other hand, although the word "machine" occurs
14 times, the word "readable" appears not even once, thus implying a
highly immature process.
https://en.wikipedia.org/wiki/Machine-readable_document
Owen