AI Weekly: AI supercomputers and facial recognition to verify taxpayers’ identities


Missed a session from the Future of Work Summit? Visit our Future of Work Summit on-demand library to stream.

Supercomputers and facial recognition dominated the headlines in AI this week — but not necessarily to an equal extent. Meta, the company formerly known as Facebook, announced it is building a server cluster for AI research that it says will be one of the fastest of its kind. Meanwhile, the IRS quietly implemented a new program with a vendor, ID.me, that controversially uses facial recognition technology to verify taxpayers’ identities.

Meta’s new “AI supercomputer” — dubbed AI Research SuperCluster (RSC) — is certainly impressive. Work on this started a year and a half ago, with phase one reaching the operational stage in recent weeks. Currently, RSC has 760 Nvidia GGX A100 systems with 6,080 connected GPUs, as well as custom cooling, power, network and cabling systems. Phase two will be completed in 2022, giving RSC a total of 16,000 GPUs and the capacity to train AI systems “on data sets as large as an exabyte.”

Meta says RSC will be applied to train a range of systems across Meta’s businesses, including content moderation algorithms, augmented reality features, and metaverse experiences. But the company has not announced plans to make RSC’s capabilities public, which many experts say highlights the resource inequality in the AI ​​industry.

“I think it’s important to remember that Meta spends money on big expensive glasses because money is their strength – they can spend more than people and get the big results, the big headlines that they want,” Mike Cook , an AI researcher at Queen Mary University in London, told VentureBeat via email. “I definitely hope that Meta can do something interesting with this and that we all benefit from it, but it’s really important that we put this into context – private labs like [Meta’s] redefine progress along these narrow lines where they excel, so they can position themselves as leaders.”

Large companies dominate the list of “AI supercomputers,” unsurprisingly, given the costs involved in building such systems. Microsoft announced two years ago that it has created a 10,000-GPU AI supercomputer that runs on its Azure platform with research lab OpenAI. Nvidia has its own in-house supercomputer, Selene, which it uses for AI research, including training natural language and computer vision models.

Os Keyes, an AI ethicist at the University of Washington, described the trend as “worrying.” Keyes says moving towards bigger and more expensive AI computing infrastructure erroneously rewards “scale and hegemony” while holding onto “monolithic organizational forms” as the logical or efficient way of doing things.

“It says some interesting things about Meta — about where it chooses to focus its efforts,” Keyes said. “That Meta’s investment direction is in algorithmic systems shows exactly how hard they’ve pinned themselves on ‘technosolutionism’… It’s change driven by what impresses shareholders and what impresses the ‘California ideology’, and that’s totally no change.”

Aiden Gomez, the CEO of Cohere, a startup developing large language models for a range of use cases, called RSC a “great achievement.” But he stressed that it is “another proof that only the largest organizations can develop and benefit from this technology.” While language models in particular have become more accessible in recent years, thanks to efforts like Hugging Face’s BigScience and EleutherAI, advanced AI systems remain expensive to train and deploy. Training language models like Nvidia’s and Microsoft’s Megatron 530B, for example, can cost up to millions of dollars — not counting storage costs. Inference – actually executing the trained model – is another barrier. A estimation puts the cost of running GPT-3 on a single Amazon Web Services instance at a minimum of $87,000 per year.

“The big push for us at Cohere is to change this and broaden access to the output of powerful supercomputing advances — large language models — through an affordable platform,” Gomez said. “Ultimately, we want to avoid the extremely resource-intensive situation where everyone has to build their own supercomputer to access high-performance AI.”

Facial Recognition for Taxes

In other news, the IRS announced this year that it is contracting ID.me, a Virgnia-based facial recognition company, to verify the identities of taxpayers online. As reported by Gizmodo, starting this summer, users with an IRS.gov account will be required to provide a government ID, a selfie, and copies of their bills to perform certain tasks, such as getting a transcript online (but not e-mailing taxes). .

The IRS is listing the new measures as a way to “protect taxpayers’ safety.” But ID.me has a troubled history, as evidenced by complaints from residents of the 30 or so states that have contracted with the unemployment benefit verification company.

In New York, News10NBC detailed accounts from residents who struggled to navigate ID.me’s system, including a woman who claimed she had been waiting 19 weeks for her benefits. Some have suggested that people of color are more likely to be misidentified by the system — which wouldn’t be surprising or unprecedented. Gender and racial bias are a well-documented phenomenon in facial analysis algorithms, which can be attributed to imbalances in the datasets used to train the algorithms. In a 2020 study, researchers showed that algorithms can even be biased toward facial expressions, such as smiles, or different outfits — which could reduce their recognition accuracy.

Worryingly, ID.me hasn’t been completely honest about the capabilities of its technology. Contrary to some of ID.me’s public statements, the company compares faces to a large database — a practice that privacy proponents fear poses a security risk and could lead to “mission creep” from government agencies.

“This dramatically increases the risk of racial and gender bias on the platform,” Surveillance Technology Oversight Project director Albert Fox Cahn told Gizmodo. “More fundamentally, we need to ask why Americans should trust this company with our data if they are not honest about how our data is being used. The IRS should not give any company that much power to decide how our biometric data is stored.”

For AI reporting, send news tips to Kyle Wiggers – and make sure to subscribe to the AI ​​Weekly newsletter and bookmark our AI channel, The Machine.

Thank you for reading,

Kyle Wiggers

AI Staff Writer

VentureBeat’s mission is to be a digital city square for tech decision makers to learn about transformative business technology and transactions. Learn more

Leave a Reply

Your email address will not be published.