Meta’s AI succeeds at the negotiation-heavy “Diplomacy,” Intel’s new deepfake detector looks at blood flow to spot frauds, and the FCC’s ruling on Chinese surveillance tech

Sign up
Worth Knowing

Meta’s New AI Cracks “Diplomacy”: Last week, Meta introduced an AI system — dubbed “Cicero” — it says is capable of playing the popular strategy game “Diplomacy” at a relatively high level. AI systems have proved adept at many single and multiplayer games, but Diplomacy has been a harder nut to crack. The turn-based strategy game — in which multiple players attempt to win control of the game’s map  — relies heavily on interpersonal negotiation, alliance-forming and deceit. To play the game at a high level, not only would an AI player need to to anticipate the movements of the other players and plot a winning strategy, it would also need to be capable of convincingly engaging with its human competitors. As they describe in a paper for the journal Science, Meta researchers built Cicero by combining an algorithm-based strategic reasoning engine with a 2.7 billion parameter language model that was fine-tuned on messages from more than 40,000 online games of Diplomacy. The researchers then entered Cicero anonymously in 40 online “Blitz” games of Diplomacy (limited to five minutes per turn) against human players over several months, during which it sent more than 5,000 messages. By the end of its run, Cicero ranked in the top 10 percent of the players in its 82-person league who had played more than one game. The result is not at the world-class level of an AlphaGo, nor is it clear if it could sustain its high level in a full (as opposed to Blitz) version of the game, but Cicero’s performance nevertheless seems to be an impressive step toward AI systems capable of better interacting and cooperating with humans. But there are some concerns worth noting too — while technically designed not to lie, Cicero has raised questions about the advisability of training AI systems to outfox human strategists.
Key Chip Workers Are On the Move — Where Will They Go Next?: As governments across the world try to ramp up their semiconductor production, many have discovered that talent is among the industry’s scarcer inputs, and companies in the United States, Europe, China and South Korea continue to struggle to find the engineers they need. Now, recent reports indicate that some of those highly prized workers could be on the move, exacerbating the talent shortage in China and potentially alleviating them elsewhere. Soon after the U.S. Commerce Department announced its new export controls, multinational semiconductor and SME firms began to pull their U.S. employees off of China-related work or, in some cases, cut their operations in China entirely. Those talent outflows add to another significant loss for the PRC’s chip industry — as The New York Times reports, Taiwanese semiconductor engineers (who, in 2019, accounted for approximately 10 percent of the industry’s workforce in China) have been steadily returning home in the face of increasing China-Taiwan tensions and the PRC’s harsh pandemic restrictions. While many of them will likely find posts with Taiwan-based companies, they could also be in high demand among U.S., European, South Korean and Japanese chipmakers. If these nations can adapt their immigration processes accordingly (and do so quickly), this could be a unique opportunity to acquire needed talent.
Intel Rolls Out A Deepfake Detector: Intel introduced a commercially available real-time deepfake detection tool that it says can accurately identify 96 percent of deepfake videos. FakeCatcher, developed by researchers at Intel and SUNY Binghamton, uses a convolutional neural network to identify irregularities in the PPG signals — a way of measuring how blood moves through tissue — of a subject’s face. Deepfakes have been growing in both sophistication and prominence of late — a deepfake of Ukrainian President Volodymyr Zelenskyy appearing to surrender to Russia caused a brief stir earlier this year. Finding ways to quickly identify and debunk deepfakes could help blunt their impact, but a panacea has been elusive, with many earlier deepfake-identifying systems proving vulnerable to adversarial attacks. For now, the researchers behind FakeCatcher say that because the tool is looking for biological signals in the image, it isn’t susceptible to the same types of adversarial attacks and is, at least for now, less easily tricked. If the cat-and-mouse history of recent deepfake development is anything to go by, however, it likely won’t be long before PPG signals can be convincingly impersonated too.
Government Updates

State AGs Urge the FTC to Enact New Data Privacy Rules: With the Federal Trade Commission considering new data privacy rules, attorneys general from 32 states and the District of Columbia encouraged the agency to pursue stronger protections. In August, the FTC announced it was exploring new rules to “crack down on harmful commercial surveillance and lax data security” and called for public comment on the potential changes. In a letter to the agency, the attorneys general urged the FTC to consider new rules that would strengthen privacy protections for individuals’ location, biometric and medical data. The letter also encouraged the agency to consider “data minimization” rules that would limit the amount of personal data firms can collect to what is “reasonably necessary,” a strategy that California, Colorado, Connecticut, Utah and Virginia have already pursued. The robust AG support could strengthen the FTC’s resolve, but backing for an FTC-led effort hasn’t been universal. In a letter of their own, Senators Cramer, Lummis and Rubio expressed concern that any new FTC rules would “only add to the existing ‘patchwork’ of state privacy laws” and pushed the agency to “leave the task of creating data privacy and security rules to the elected officials in Congress.” As we’ve covered, there has been considerable bipartisan support in Congress for new data privacy legislation — such as the proposed American Data Privacy and Protection Act — but most observers are skeptical that a data privacy bill will pass this year.

FCC Votes to Restrict U.S. Sales of Chinese Telecom and Surveillance Tech: Last week, the Federal Communications Commission voted to ban U.S. sales of new telecommunications and surveillance equipment made by several Chinese companies — including Huawei, ZTE, Hytera, Hikvision and Dahua — over concerns that the products pose a risk to U.S. national security. The unanimous FCC ruling prohibits imports of any new products and opens up the possibility that the FCC could revoke existing authorizations for the products that still can be sold. The 2019 NDAA had already prohibited federal agencies from purchasing tech from many of the covered companies, but that hadn’t cut off the firms’ access to important parts of the U.S. market — as a recent CSET brief found, at least 1,681 state and local government entities bought equipment from the companies covered by the 2019 NDAA.    

In Translation
CSET’s translations of significant foreign language documents on AI

PRC Innovation Plan: Special Plan for the Technology Factor of Production Market in the 14th Five-Year Plan Period. This Chinese government plan aims to improve China’s tech ecosystem, chiefly through specific measures to make it easier to convert technological breakthroughs into commercial products or other practical applications. Most of the plan focuses on domestic innovation, but it also proposes new ways for China to exploit foreign technologies through tech transfer, talent recruitment and so forth.

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.

CSET Job Openings

We’re hiring! Please apply or share the roles below with candidates in your network:
  • People Operations Specialist: We are currently seeking a People Operations Specialist to play a key role in helping to build and develop the CSET team, with a particular focus on furthering our diversity, equity and inclusion initiatives. This Specialist will provide administrative, organizational and project management support to ensure that our people-focused operations run smoothly. Applications due by January 30.
  • Fellow – Emerging Technology Supply Chains: We are currently seeking candidates to lead and coordinate our Emerging Technology Supply Chains Line of Research, either as a Research Fellow or Senior Fellow (depending on experience). This fellow will shape priorities, lay out an overall research strategy, oversee execution of the research and production of reports, and help hire and manage supporting researchers. Applications due by January 2.
Please bookmark our careers page to stay up to date on all active job postings. You can also subscribe to receive job announcements by updating your subscription preferences in the footer of this email.

What’s New at CSET

PUBLICATIONSIN THE NEWS
What We’re Reading

Report: The State Of AI: Q3 2022 Report, CB Insights (November 2022)

Blog Post: 137 Emergent Abilities of Large Language Models, Jason Wei (November 2022)

Report: Military and Security Developments Involving the People’s Republic of China, U.S. Department of Defense (November 2022)

Upcoming Events


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