3-Shot Learning #7
3-Shot Learning #7
AI hits elections, supercomputers strengthen Meta, and No AI FRAUD Act enters Congress
Welcome to 3-Shot Learning, a weekly newsletter from the Center for AI Policy. Each issue explores three important developments in AI, curated specifically for AI policy professionals.
AI Tools Enable Election Shenanigans
On Sunday, two days before the New Hampshire Primaries, numerous New Hampshire voters received a call from an AI voice imitating President Biden. The fake audio urged New Hampshirites to stay away from the primaries and instead “save” their votes for the November general election. Importantly, it's entirely untrue and misleading to suggest that voting in the primaries prevents individuals from voting in the general election.
The New Hampshire Department of Justice Election Law Unit is currently investigating the case, which it calls “an unlawful attempt [...] to suppress New Hampshire voters.”
The Biden robocall was not the only AI election story this week. Last Thursday, the Washington Post reported that a new super PAC had launched Dean.Bot, an AI voice clone of the Democratic presidential candidate, Dean Phillips. The bot could talk with anyone on the internet who visited its eponymous website. Dean.Bot notably utilized OpenAI’s ChatGPT, even though OpenAI’s forbids usage in political campaigns.
As such, it was not terribly surprising when OpenAI suspended the account behind the bot. In response, the bot’s creators said they would transition to using open source AI systems. But at the time of writing, the bot is still down.
AI’s electoral impacts are not limited to the US; they are global, with over fifty different countries holding elections in 2024. One of those countries is the United Kingdom, whose Conservative PM was recently impersonated in over 100 deepfakes that circulated on Meta as paid advertisements between December 8th and January 8th, according to a new report.
The ads may have reached over 400,000 people, and represent some of the first paid promotions of AI videos impersonating a UK political figure. When asked about the incident, a Meta spokesperson said the company has provided “industry-leading transparency for ads about social issues, elections or politics.”
Meta Stockpiles Chips for Open Source “General Intelligence”
In an announcement post on Threads and Instagram, Meta CEO Mark Zuckerberg announced that the company will have 350,000 units of NVIDIA’s industry-leading H100 AI chip by the end of the year. Further, after accounting for non-H100 chips, the full Meta compute arsenal will wield computational power equivalent to 600,000 NVIDIA H100s.
Based on the H100’s performance stats, this should allow Meta to execute almost a sextillion mathematical operations per second, which would be enough to train a massive model like GPT-4 in under a day (for reference, training sessions for the largest AI models typically span multiple months).
Zuckerberg stated that Meta plans to use these chips to “support our long-term goals of building general intelligence,” which Meta will make “as widely available as we responsibly can.”
Bipartisan House Group Seeks to Prohibit Nonconsensual Deepfakes
In the House this month, Representatives Maria Salazar (R-FL), Madeleine Dean (D-PA), Nathaniel Moran (R-TX), Joe Morelle (D-NY), and Rob Wittman (R-VA) introduced the No Artificial Intelligence Fake Replicas And Unauthorized Duplications (No AI FRAUD) Act in the House.
The bill would give Americans intellectual property rights to their own image, likeness, and voice, and establish legal liability for anyone who distributes digital impersonations of someone else without permission.
Notably, the liability would apply even when such content clearly discloses that it is fake or unauthorized. Critics of the bill argue that its definitions are so broad that it might affect video recordings, political cartoons, and other areas unrelated to AI, so the representatives may have some edits to make. Nonetheless, the intention to take action on this topic is commendable, given the risks posed by increasingly undetectable synthetic content.
News at CAIP
We just launched an account on Twitter (X). Follow @aipolicyus to stay informed daily about key AI policy stories. (Also: don’t sleep on our LinkedIn.)
CAIP on the road: Olivia Jimenez paid a visit to Boston for the launch of the MIT Shaping the Future of Work Initiative. We’re excited to see more work from their excellent team.
Quote of the Week
There’s a lot of people that want AlphaFold to do everything, and a lot of structural biologists want to find reasons to say we are still needed.
—biochemistry professor Jens Carlsson, commenting on the recent AI-powered discovery of potential new psychedelics
This edition was authored by Jakub Kraus.
If you have feedback to share, a story to suggest, or wish to share music recommendations, please drop me a note at jakub@aipolicy.us.
—Jakub