DeepMind’s “generalist” AI, Clearview settles over privacy claims, and Congress tries to reconcile its tech bills

Sign up
Work at CSET

We’re hiring! If you’re interested in joining our team, check out the positions in the “Job Openings” section below or consult our careers page.

Worth Knowing

DeepMind Introduces a Model Capable of Performing 600+ Tasks: Last week, DeepMind introduced Gato — a single-transformer “generalist agent” capable of performing more than 600 tasks, including playing Atari games, controlling a robot arm, captioning images, and engaging in dialogue. Most state-of-the-art models are designed to perform one task in one or perhaps two modalities (e.g., text and images). One of Gato’s key achievements is that it is a single, unified system that can achieve multiple tasks across multiple modalities. Though its performance in these tasks ranged from state-of-the-art to mediocre, Gato’s success has some excited about what it means for future “generalist” agents. In a paper, DeepMind researchers hypothesized that scaling up Gato — which is relatively svelte at only 1.2 billion parameters (GPT-3 has 175 billion) — could enable future models to cover “any task, behavior and embodiment of interest.” While DeepMind researchers expressed enthusiasm on Twitter, a number of observers were quick to pump the brakes on talk of genuine artificial general intelligence. AI researcher Mike Cook pointed out that Gato wouldn’t perform particularly well on any task beyond the 604 for which it was trained — something we would expect a truly “general” AI to pull off with ease. And as long-time deep learning skeptic Gary Marcus noted, it’s not clear that scaling can solve the problems that have plagued ML systems, like hallucinating.      
Clearview Agrees to National Limits in Illinois Settlement: Last week, the facial recognition company Clearview AI agreed to limit access to its database of 20 billion facial images as part of a settlement in an Illinois lawsuit. Under the settlement’s terms, the company — which has built a powerful (and controversial) facial recognition tool with a vast database of photos scraped from social media and other online sources — would not offer access to its database to private individuals or businesses in the United States, though it can still provide access to its algorithms without the database. Clearview will also cease working with Illinois state and local agencies for five years, but will still be able to do business with state and local agencies in other states, as well as federal agencies, which are among its biggest customers. The settlement, which must still be approved by a judge, stems from a 2020 lawsuit brought by the ACLU and other privacy and civil rights groups that alleged the company had violated Illinois’ Biometric Information Privacy Act. That law had already proven a powerful asset for privacy advocates in the state — last year, Facebook agreed to pay $650 million to settle a class action lawsuit related to the company’s use of facial recognition on Illinois users’ photos — but observers say that Clearview’s settlement could provide a blueprint for how to give the law teeth on a national scale.
What A Tech Downturn Could Mean for AI: Amid tanking tech stocks and news of layoffs, concerns are growing about the tech industry’s future. The downturn has been widespread, impacting tech giants and unicorns alike, but it remains unclear what that will mean for AI in particular. Early indications are not great — according to CB Insights’ State of AI Q1 2022 Report, global funding to AI startups fell by 12 percent from Q4 2021 to just over $15 billion. Major tech companies that have spent heavily on AI in recent years have also been hard-hit — DeepMind-parent Alphabet is down nearly 23 percent in 2022, while OpenAI-backer Microsoft is down 24 percent. Nvidia, whose AI chips have helped it print money in recent years, has been hit even harder, falling over 40 percent so far. The news isn’t all bad, though — even with the AI-startup-funding dip, the most recent quarter was still the fifth biggest AI funding quarter ever, and AI funding’s drop was less than that of venture funding overall, which fell by 19 percent. A prolonged tech downturn doesn’t necessarily mean another AI winter is on the horizon, but it could increase the importance of profitability for a sector that has gotten used to running a deficit.    
Government Updates

U.S. and EU Hold Second Round of Tech Talks: Senior U.S. and EU officials — including Secretary of State Antony Blinken and European Commission Executive Vice President Margrethe Vestager — gathered in France over the weekend as part of the second meeting of the U.S.-EU Trade and Technology Council. In a joint statement issued Monday, the parties recognized a number of shared intentions and agreements, and reported some key outcomes since the inaugural TTC meeting in September. Of particular note for AI policy observers, the TTC co-chairs announced:
  • The formation of an AI sub-group that will work on a joint roadmap for evaluation and measurement tools for “trustworthy AI and risk management.” An annex to the joint statement spelled out plans to create a “shared hub/repository of metrics and methodologies for measuring AI trustworthiness and AI risks,” to develop a “interoperable terminology,” and to conduct an economic study of AI’s workforce impact.  
  • The creation of a “U.S.-EU Strategic Standardisation Information mechanism” to share information about international standards development.
  • An “early warning system” to address potential semiconductor shortages and cooperation to avoid subsidy races.
  • The creation of an “early alert dialogue on shared trade concerns regarding initiatives or measures of third countries, and a mechanism to consult each other on bilateral barriers at an early stage.” The joint statement mentions Russia, but not China. Discussions about China have been a source of disagreement since the council was formed last year, with EU officials reportedly pushing back against the United States’ more pronounced China focus.
According to the joint statement, the next meeting of the TTC will be held in the United States before the end of the year.

ICE Uses Data Brokers and AI to Track Large Share of U.S. Adults: Using private data brokers, access to state and local records, and commercially available facial recognition and data processing tools, Immigration and Customs Enforcement has built a system capable of processing personal information about more than half of U.S. adults — no warrant needed. According to a report released last week by the Georgetown Law Center on Privacy & Technology, through contracts with private data brokers such as Thomson Reuters and LexisNexis, ICE has gained access to the driver’s licenses and utility records information of approximately 75 percent of U.S. adults. Between 2008 and 2021, the agency spent nearly $570 million on data analysis and contracted services with facial recognition companies — including Clearview AI — to scan the driver’s licenses of nearly a third of U.S. adults. While a number of states have passed laws that aim to prevent ICE from accessing their residents’ data, the report’s authors say those laws have proven ineffective — ICE is frequently able to just purchase the same data through private brokers.

Congress Tries to Reconcile Competitiveness Bills: Last week, Congress held the kickoff meeting of the 107-member conference committee in charge of reconciling its two innovation and competitiveness bills, the House-passed America COMPETES Act and the Senate-passed U.S. Innovation and Competition Act. Despite pressure from President Biden, support from the semiconductor industry, and bipartisan backing (the Senate bill passed with 68 votes), the bills — not to mention the $52 billion in semiconductor funding they would unlock — have remained stuck as Congress has worked slowly to overcome the differences between them. One such difference — the bills’ treatment of skilled immigration — was in the spotlight last week. A bipartisan group of 49 former high-ranking national security officials wrote a letter to Congressional leaders and members of the conference committee urging them to preserve in the final bill a provision in the House-passed measure that would exempt foreign-born STEM advanced-degree holders from green card caps. The letter — which included former Secretaries of Defense Hagel and Cohen as signatories — argued that without improved access to the global pool of STEM talent, “it will be very hard for America to win” in competition with China.

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

PRC Science and Technology Budget: Ministry of Science and Technology 2022 Annual Budget. In addition to detailing expenditures by category, the PRC Ministry of Science and Technology (MOST) 2022 budget also describes several domestic and international S&T projects that MOST is funding.

PRC Industry and Information Technology Budget: Ministry of Industry and Information Technology 2022 Annual Budget. The Ministry of Industry and Information Technology oversees a wide variety of technology fields and institutions, including basic and applied scientific research, industrial policy, China’s telecommunications industry, internet regulation, cybersecurity, radio frequency regulation, emergency broadcasting, “military-civil fusion,” and the “Seven Sons of National Defense” universities.

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.

Job Openings

We’re hiring! Please apply or share the roles below with candidates in your network:
  • Software Engineer: Software Engineer to assist with data cleaning and normalization, data pipeline development and/or web development, and creating automated test suites for CSET’s new Emerging Technology Observatory. Apply by June 3.
  • UI/UX Designer: We are currently seeking applications for a UI/UX Designer to perform user interviews, write user stories, create user interface mockups, and conduct usability testing for public-facing Emerging Technology Observatory products. Apply by June 3.
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

PUBLICATIONSTESTIMONY
  • Senate Select Committee on Intelligence: On May 11, CSET Director Dewey Murdick testified at a Senate Select Committee on Intelligence hearing on “Countering the People’s Republic of China’s Economic and Technological Plan for Dominance.” Murdick discussed China’s strategy to move toward self-sufficiency in key technologies and steps the United States can take to respond. Read his testimony or watch it here.
  • House Science, Space and Technology Subcommittee on Investigations and Oversight and Subcommittee on Research and Technology: On May 11, CSET Senior Fellow Andrew Lohn testified before two House Science, Space and Technology subcommittees at a hearing on “Securing the Digital Commons: Open-Source Software Cybersecurity.” Lohn discussed how the United States can maximize sharing within the artificial intelligence community while reducing risks to the AI supply chain. Read his testimony or watch it here.
IN THE NEWS
What We’re Reading

Report: Exploring the Civil-Military Divide over Artificial Intelligence, James Ryseff, Eric Landree, Noah Johnson, Bonnie Ghosh-Dastidar, Max Izenberg, Sydne Newberry, Christopher Ferris and Melissa A. Bradley, RAND Corporation (May 2022) 

Blog Post: Shared Responsibility: Enacting Military AI Ethics in U.S. Coalitions, Casey Mahoney, The National Interest (April 2022) 

Commentary: Digital Transformation Is a Cultural Problem, Not a Technological One, Brandon Leshchinskiy and Andrew Bowne, War on the Rocks (May 2022)

Upcoming Events


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