The Pentagon scraps JEDI, the House passes NSF for the Future, GitHub announces an AI autocomplete tool, and a drone swarm may have been used in combat for the first time

Sign up
Worth Knowing

GitHub AI-Powered Tool Generates Code — and Controversy: GitHub, the Microsoft-owned software and code-hosting platform, announced a new AI-powered programming tool that works like an auto-complete for code, suggesting code based on what a programmer has already written and even generating it based on plain text comments. The tool — called “Copilot” — is the result of a collaboration with OpenAI, the research lab behind GPT-3 (and in which Microsoft invested $1 billion in 2019). Copilot uses a new OpenAI system called Codex, a model trained on both natural language and billions of lines of code, much of it public code hosted on Github. While the tool is still available only in a technical preview, it has already generated some controversy, mostly related to its returning private information or code that potentially violates copyright law. Github says such instances are rare and solutions are in the works, but observers note that the issue may run deeper: when AI systems are trained on public code created by individual coders, the line between fair use and plagiarism becomes difficult to discern.
Israel May Have Been the First to Deploy a Drone Swarm in Combat: According to recent reports, the Israeli Defense Forces used a drone swarm — a group of drones flying in coordination without a central controller — during the conflict between Israel and Hamas this May. If accurate, it would mark the first time that a drone swarm had been used in combat. While it appears the drones, manufactured by Israel-based Elbit Systems, were used to conduct reconnaissance and communicate targeting information to human partners, there is no indication that the swarm was tasked with — or capable of — killing autonomously. Reporting from May highlighted the role artificial intelligence played behind the scenes for the IDF — helping to process and analyze data, for example — but the delayed reporting on the use of drone swarms highlights a problem that seems likely to recur: because autonomous systems can be difficult to distinguish from human-controlled systems, their use in combat may not be immediately obvious to observers. That could make it harder to monitor, track and regulate their use in the field.
The GPU Shortage Might Be Coming to an End: After nearly a year of graphics processing unit (GPU) shortages that pushed the prices of some commercially available graphics cards to more than three times their MSRP, there might finally be a light at the end of the tunnel. GPU prices in Germany have fallen to a six-month low, and prices in China fell by as much as two thirds in late June. The market changes appear to be a result of Beijing’s recent crackdown on cryptocurrency mining — a GPU-intensive enterprise that had been largely concentrated in China (an estimated 65 to 75 percent of global Bitcoin mining took place there). With the Chinese government banning cryptocurrency mining in Inner Mongolia, Xinjiang and Sichuan — the three most popular provinces for mining — Chinese second-hand markets have reportedly been flooded with GPUs from no-longer-profitable mining operations. While dropping prices would be welcome news for AI development — GPUs are one of the most common types of chip used for AI training — observers say the broader semiconductor shortages are likely to persist, even if GPU prices do fall back to earth.      
Government Updates

Pentagon strikes down JEDI — Hopes for More Powerful Replacement: The Pentagon announced last week that it had cancelled its cloud-computing contract with Microsoft and will seek proposals for a new contract better-suited to its needs. The DOD’s 2019 decision to award Microsoft the Joint Enterprise Defense Infrastructure (JEDI) Cloud contract had proved a headache for the government, with both Amazon and Oracle filing lawsuits alleging undue political meddling in the process. While observers said the decision to scrap the JEDI contract — which could have been worth up to $10 billion over 10 years — was a convenient way around years of potential litigation and delays, the DOD’s announcement highlighted the computing requirements of two relatively new initiatives — Joint All Domain Command and Control and the Artificial Intelligence and Data Acceleration Initiative — as motivating factors behind the decision to scrap the deal. The replacement — the Joint Warfighter Cloud Capability contract — will be a multi-vendor deal over five years, though a price tag has not yet been set. A Pentagon spokesman said a multicloud — multiple clouds from different providers — will provide the military with the “varying capabilities” it needs to carry out its missions. The Pentagon singled out Microsoft and Amazon as the chief candidates for the new contract, but many smaller cloud service providers are also aiming to get in on the action.  

House Passes NSF For the Future & DOE Science for the Future Acts: Late last month, the House adopted two bills — the National Science Foundation for the Future Act and the Department of Energy Science for the Future Act — that would direct billions in new funds to the NSF and DOE. Passed by votes of 345-67 and 351-68, respectively, the bills are part of the House’s response to the Senate’s recently passed U.S. Innovation and Competition Act (which we covered here). Both the NSF for the Future Act and the Senate bill would establish a new directorate inside the NSF focused on advancing research in priority technology areas, but there are key differences between the two bills. The Senate bill increases NSF’s budget to more than $21 billion by FY2026, $9.3 billion of which would go to the new directorate, while the House proposal tops out at $18 billion and proposes a significantly smaller new directorate. The House Foreign Affairs Committee is scheduled to finish its consideration of legislation that mirrors elements of the Senate’s final bill. Plans for the chambers to negotiate a final compromise have not been announced.

Biden Issues Sweeping Competition EO With AI Implications: Last week, President Biden signed an executive order aimed at expanding competition in the U.S. economy. Among its 72 provisions are a number with important implications for the tech industry and AI development. The White House encouraged the Federal Trade Commission to focus “vigorously” on enforcing antitrust laws, to review “prior bad mergers that past Administrations did not previously challenge,” and to enact rules to curb “unfair data collection and surveillance practices” and the use of non-compete clauses (which are prevalent in the tech industry). While the FTC is independent and not legally bound by the order, its new chair, Lina Khan, and the acting chief of the DOJ’s antitrust division, Richard Powers, announced a joint review of merger guidelines. Even with stricter scrutiny and enforcement, the FTC could face an uphill battle in pursuing antitrust cases. The week before the White House issued its order, a federal judge dismissed two antitrust cases against Facebook lodged by the FTC and 46 state attorneys general.

GAO Calls for More Oversight of Federal Facial Recognition Tech Use: A report from the Government Accountability Office found that the use of facial recognition technology is widespread across the federal government, but that a number of federal agencies, including the FBI, DEA, ATF and ICE, did not have procedures in place to track which systems their employees use. The GAO surveyed 42 federal agencies and found that 20 reported using facial recognition systems. Of those, 17 reported using non-proprietary systems, including commercial tools such as Clearview AI and Amazon’s Rekognition, and 13 said they did not have systems in place to track the use of non-federal tools by employees. Noting that this lack of tracking puts the 13 agencies at risk of violating privacy-related laws and regulations, the GAO recommended that each one set up a system to track its employees’ use of non-federal facial recognition tools and assess the risks associated with their use.  

CSET Job Openings

Please share with qualified candidates in your network or consider applying:
  • Research Fellow – AI TEV&V will focus on the safety and risk of deployed AI systems by researching real-world AI incidents and use these identified incidents with other analyses of AI systems to inform policy recommendations regarding AI safety, test, evaluation, verification and validation (TEV&V) processes, standards setting and management, and the appropriate employment and operation of AI systems by businesses and the US Government (including the military). Applications due by July 19
In Translation
CSET’s translations of significant foreign language documents on AI

PRC Cybersecurity Base Plan: Notice of the [Wuhan] City People’s Government on Certain Policies to Support the Development of the National Cybersecurity Talent and Innovation Base. This document, issued by the Wuhan City government in 2017, lays out the city’s various preferential policies to encourage the development of Wuhan’s National Cybersecurity Talent and Innovation Base.

PRC State Labs Regulations: Provisional Administrative Methods for Entrusting Enterprises with Building State Key Laboratories. This regulation governs the establishment, operation, and evaluation of PRC state key laboratories that are housed at corporations. These corporate labs are intended to facilitate the adoption of scientific breakthroughs into commercially viable technologies, although the companies that host them are prohibited from using the lab facilities to produce products for sale. State key labs are typically based at universities or at or state-run research institutes.

If you have a foreign-language document related to security and emerging technologies that you’d like translated into English, CSET may be able to help! Click here for details.

What’s New at CSET

REPORTSPUBLICATIONS, PODCASTS AND VISUALIZATIONSFORECASTS
CSET maintains a crowd forecasting platform. Sign up as a forecaster, and take a look at some of the predictions so far:
EVENTSIN THE NEWS
What We’re Reading

Report: Standards Bearer? A Case Study of China’s Leadership in Autonomous Vehicle Standards, Matt Sheehan, MacroPolo (June 2021)

Report: Tasks, Automation, and the Rise in US Wage Inequality, Daron Acemoglu and Pascual Restrepo, NBER (June 2021)

Article: At the Nexus of Military-Civil Fusion and Technological Innovation in China, Audrey Fritz, The Diplomat (July 2021) 

Index: AI-ness Index, AMPLYFI (July 2021)

Upcoming Events


What else is going on? Suggest stories, documents to translate & upcoming events here.