Trust-busters come for AI's building blocks

From: POLITICO's Digital Future Daily - Monday Oct 02,2023 08:21 pm
How the next wave of technology is upending the global economy and its power structures
Oct 02, 2023 View in browser
 
POLITICO's Digital Future Daily newsletter logo

By Ben Schreckinger

With help from Derek Robertson

The seal of the Federal Trade Commission(FTC) is seen sewn into the carpet in their Commissioner's Conference Room January 15, 2014 inside the FTC headquarters in Washington, DC . AFP Photo/Paul J. Richards / AFP PHOTO / Paul J. RICHARDS (Photo credit should read PAUL J. RICHARDS/AFP via Getty Images)

The Federal Trade Commission logo. | AFP via Getty Images

As regulators worldwide hash out their response to the rapid advance of artificial intelligence, it appears they have found an early target: the microchips that power the most advanced systems like OpenAI’s ChatGPT or Google’s Bard.

French regulators raided the office of a major chipmaker last week, reportedly market leader Nvidia, and EU regulators are said to be in the early stages of investigating whether chipmakers have taken steps to illegally hamper competition.

The moves in Europe are the latest sign that policy fights over the virtual future are set to increasingly focus on its physical underpinnings, in particular the vast quantities of raw computing power needed to make AI work. There are fears that the biggest chipmakers — or dominant players in other facets of computing infrastructure — could become something like the Standard Oil of AI if they are able to further entrench their positions in the technology’s supply chain.

And it’s not just the EU that’s concerned. At POLITICO’s Tech & AI Summit Wednesday, Federal Trade Commission Chair Lina Khan made clear that antitrust issues with AI are a priority, including a focus on the technology’s building blocks:

“I met with some VCs the other month, many of whom are investing in startups that are using AI tools, and several shared that one of their big concerns is that right now, some of these tools can be acquired relatively cheaply, but the fact that a handful of companies control some of the raw materials and critical inputs, gives them pause,” Khan told POLITICO’s Josh Sisco. “In three months or six months, they could wake up and face dramatically jacked-up prices, or potentially coercive terms where access to certain key base models or other AI technologies is being conditioned on reupping your cloud contract or some other part of the business. And so those are all dynamics that we're aware of and trying to stay vigilant.”

As DFD noted last spring, much of the initial antitrust concern that followed the sudden explosion in popularity of ChatGPT focused on the data needed to train new AI models, because of the small number of companies that possess enough high-quality data to do it well.

But AI policy researcher Jai Vipra told DFD there are several reasons why computing power is emerging as a top early concern for AI competition. For one, it's easier for regulators to understand than data, which is abstract and opaque, making it difficult to know how AI firms are using it.

"Right now, compute seems like a more concrete input to regulate," said Vipra, who co-authored a "Computational Power and AI" report released last week by the AI Now Institute, a New York-based think tank focused on the technology’s social implications.

And — due largely to a shortage of the advanced chips needed to train and run the most advanced AIs — computational power has become the most important bottleneck to AI adoption, lending it a sense of urgency. ”It’s the constraint right now,” Vipra said.

Then there’s the extraordinary run-up in NVIDIA’s stock. The value of the American company has more than tripled this past year, making it the first chipmaker to reach a trillion-dollar valuation. That sent a flashing red signal to pay attention to it, Vipra said.

The early focus on computing power does not mean that the data used to train AI will evade fights over market structure forever.

“I don't think there is enough evidence in the public domain yet,” Vipra said. “Once it does come out, I think it’s going to be a really big issue.”

 

GO INSIDE THE CAPITOL DOME: From the outset, POLITICO has been your eyes and ears on Capitol Hill, providing the most thorough Congress coverage — from political characters and emerging leaders to leadership squabbles and policy nuggets during committee markups and hearings. We're stepping up our game to ensure you’re fully informed on every key detail inside the Capitol Dome, all day, every day. Start your day with Playbook AM, refuel at midday with our Playbook PM halftime report and enrich your evening discussions with Huddle. Plus, stay updated with real-time buzz all day through our brand new Inside Congress Live feature. Learn more and subscribe here.

 
 
gpt as a diagnostic tool

GPT-4 can help predict cardiovascular risk, according to a group of South Korean researchers.

In a preprint published via the Lancet last week, the researchers write that in a sample size of around 47,000 patients, GPT-4 can provide “relatively accurate” predictions of risk for cardiovascular disease by analyzing factors including age, sex, diabetes, hypertension, smoking status, cholesterol, blood pressure, and body mass index.

“These findings suggest that LLMs could potentially serve as a valuable tool in healthcare settings for risk assessment and patient education,” they conclude, but warn the opacity and unpredictability of AI tools means that “without rigorous, quantitative assessment of the quality of their outputs, employing LLMs could be a double-edged sword in healthcare environments.” — Derek Robertson

swalwell's ai bootstrapping

Eric Swalwell speaks.

Rep. Eric Swalwell (D-Calif.). | Andrew Harnik/AP Photo

Add the DIY approach to the list of ways lawmakers are getting up to speed on AI.

POLITICO’s Weekly Cybersecurity reported this morning on Rep. Eric Swalwell’s (D-Calif.) efforts to educate himself on AI, as he says lawmakers need to understand the technology “not just theoretically but in practical application.”

WC reports that Swalwell is using OpenAI tools to get himself up to speed, and taking a decidedly hands-on approach as he invites computer scientists to share their screens with him and explain exactly what it is they’re doing with AI tools that could impact everything from cybersecurity to the education system.

“Don’t bring in some ivory tower think tank-er,” Swalwell said. “Start at your keyboard.” (No offense intended to our valued DFD readers at ivory tower think tanks.) — Derek Robertson

Tweet of the Day

There it is. Shares of The Sphere Entertainment surging over 10% after everyone watched all those U2 clips over the weekend. $SPHR

THE FUTURE IN 5 LINKS
  • The crypto industry is quite eager to see Sam Bankman-Fried face conviction.
  • AI is looming as a potential boon for the insurance industry.
  • Tesla misses delivery expectations just as it gears up for more production.
  • What could alien plants look like?
  • An improved robot chef is back to stoke the hype around self-driving kitchens.

Stay in touch with the whole team: Ben Schreckinger (bschreckinger@politico.com); Derek Robertson (drobertson@politico.com); Mohar Chatterjee (mchatterjee@politico.com); Steve Heuser (sheuser@politico.com); Nate Robson (nrobson@politico.com) and Daniella Cheslow (dcheslow@politico.com).

If you’ve had this newsletter forwarded to you, you can sign up and read our mission statement at the links provided.

 

Enter the “room where it happens”, where global power players shape policy and politics, with Power Play. POLITICO’s brand-new podcast will host conversations with the leaders and power players shaping the biggest ideas and driving the global conversations, moderated by award-winning journalist Anne McElvoy. Sign up today to be notified of new episodes – click here.

 
 
 

Follow us on Twitter

Ben Schreckinger @SchreckReports

Derek Robertson @afternoondelete

Steve Heuser @sfheuser

 

Follow us

Follow us on Facebook Follow us on Twitter Follow us on Instagram Listen on Apple Podcast
 

To change your alert settings, please log in at https://www.politico.com/_login?base=https%3A%2F%2Fwww.politico.com/settings

This email was sent to by: POLITICO, LLC 1000 Wilson Blvd. Arlington, VA, 22209, USA

Please click here and follow the steps to .

More emails from POLITICO's Digital Future Daily

Sep 29,2023 08:01 pm - Friday

5 Questions for James Pethokoukis

Sep 28,2023 08:03 pm - Thursday

Why AI is such a hard problem for DC

Sep 27,2023 08:02 pm - Wednesday

Metaverse policy gets closer to home

Sep 26,2023 08:34 pm - Tuesday

A call for AI to stop impersonating people

Sep 25,2023 08:02 pm - Monday

AI vs. public opinion

Sep 22,2023 08:02 pm - Friday

5 questions for Alphabet's Kent Walker