The week when AI and geopolitics collided

POLITICO’s weekly transatlantic tech newsletter for global technology elites and political influencers.

By MARK SCOTT

Send tips here | Subscribe for free | View in your browser

HELLO FROM A RAINY BLETCHLEY PARK. I’m Mark Scott, POLITICO’s chief technology correspondent, and bring you this week’s newsletter from the United Kingdom’s AI Safety Summit. Lucky me. I bring you a statement about 21st-century society that, once you’ve read it, you can never un-read it. Be warned.

A LOT TO UNPACK THIS WEEK. LET’S GET GOING:

— It’s been a busy week for artificial intelligence and geopolitics. Time to unpack what you need to know.

— Some of the biggest digital platforms pulled back the veil on how they operate — it’s a mixed bag for transparency.

— A complex web of international data transfer deals is creating an ad hoc version of global data protection standards.

AI AND POLITICS: EVERYONE’S AN EXPERT NOW

THE WHITE HOUSE EXECUTIVE ORDER ON ARTIFICIAL INTELLIGENCE. The G7 code of conduct on generative AI. The United Kingdom’s AI Safety Summit. If one week summed up the geopolitical scramble to respond to artificial intelligence, it was this one. The technology didn’t appear overnight like a ChatGPT-style deepfake. But it’s certainly true that generative AI jumpstarted the public’s — and therefore politicians’ — interest in something that has been around for decades. “Leadership for the United States and AI is not just about inventing the technology,” said Ben Buchanan, the White House’s AI adviser. “But (it’s) also about inventing and co-developing the governance mechanisms, the safety protocols, the norms and the institutions, internationally.”

Before we get into the differences, I should make clear Western democracies agree on a lot more than they don’t when it comes to artificial intelligence. That’s mostly because everyone’s response, in some form, is based on the 2019 AI Principles from the Organization for Economic Cooperation and Development. Those are evergreen tenets like the transparency, accountability, security and high levels of data protection that must be built into these complex systems. This is inevitably the “Puppy Paradox” of AI in that — like puppies — everyone can agree that those are good things. But what they actually mean differs markedly depending on whom you talk to.

But when push comes to shove, this week has all been about power. Power to determine how these AI models work in the wild. Power to demonstrate to voters that governments are in control of a technology that, at first view, can appear ungodly. And power to tell other global leaders that it’s a particular country that is leading the way — and that others should fall in line. What they are doing — and how they are doing it — is less central to the perception that political leaders have taken the bull by the horns and, for once, are getting ahead of a new technological shift before it’s too late.

There’s a lot to like about what has been announced this week. The White House’s executive order is a sprawling document. But its mandate for U.S. federal agencies to use existing powers to oversee AI’s roll-out is a good one. Its requirement that companies test future AI systems vigorously, and then report back to Washington before the tech is deployed, is a good one. The focus on short-term risks — around data bias, work displacement and civil rights — is a good one. Dare I say it: The U.S. may have finally done something useful when it comes to digital policy. But a key missing factor: any real way to enforce those requirements.

Ditto for the Brits, albeit from a lower bar. I remain immensely skeptical that one country — even one that I call home — can play alongside the likes of the U.S., China and the European Union. But the announcement that London would work in conjunction with Washington on a joint AI Safety Institute (an organization to track potential harms and risks of the technology) is a laudable effort. If that subsequently gets expanded to other like-minded countries, even if limited to the so-called 5 Eyes countries of Australia, New Zealand, Canada, the U.S. and the U.K., it would be a step forward in terms of international information sharing.

Into this world, the EU now looks a little out of place. It’s fair to say Brussels was ahead of the game when it announced, back in 2020, its push for legislation via its Artificial Intelligence Act. Those rules are still likely to be finished by December. But amid internal wrangling on what AI uses should be banned; what limits, if any, should be placed on law enforcement; and fundamental differences over whether to regulate the large language models underpinning generative AI, the 27-country bloc is showing its weakness. That, to be clear, is that the EU is not a single country, and getting consensus across such an unwieldy group of nations is always difficult, if not impossible.

One thought I’ve taken from this week of AI geopolitics is how fragile politicians’ perceptions are when it comes to their place in the world. I get that AI is a sexy topic and they want to be seen as responsive. I also get that everyone wants to be a global leader. But a year ago, no one really knew what AI was — and the prism of OpenAI’s ChatGPT is very small if you think of all the use cases that can encompass the technology. But mostly because of the generative AI craze, world leaders are bending over backward to police a technology that few understand; even fewer know what to do with; and almost no one wants to admit that, just maybe, isn’t the massive threat that the consensus believes it to be.

WHEN TRANSPARENCY IS THE LESS OF THE SUM OF ITS PARTS

THE EU WOULD LIKE YOU TO KNOW IT’S A GLOBAL DIGITAL RULE-SETTER. And that its latest iteration — the bloc’s content-moderation rulebook, known as the Digital Services Act — is the best thing since sliced bread. It requires tech giants like Meta, Alphabet and X (formerly Twitter) to be a lot more open about how they go about their business. If they are not and don’t clamp down on illegal and hateful content, EU regulators, we’re told, will come down like a ton of bricks to fine them up to 6 percent of their global revenue. We could even see the first formal investigations under Europe’s content rules — linked to graphic content related to the ongoing Middle East conflict — be announced as soon as this month.

As part of this new regulatory oversight, Brussels also requires the world’s largest social media platforms to publish regular reports detailing everything from how much content they have removed (and how they’ve done that) to the number of content moderators each company employs, split by individual European countries. The goal is to pull back the veil on what can often appear to be walled-off gardens where policy decisions on what we all see in our social media feeds can feel disconnected from everyday life. If only more was known about how these platforms worked, goes the theory, then we’d make more educated choices about where to spend our time.

That’s a wonderful idea. But I’ve spent the last week trolling through the first tranche of transparency reports required under the Digital Services Act. They are a mixed blessing. It’s great to know how many content moderators TikTok has targeted at Europe (6,125, including people who speak Hungarian and Sloevene, but not Maltese!). And it’s interesting to know that Germany is by far the European country that asks for the most amount of material to be taken down across all social media platforms, based on the companies’ disclosures. But do these figures add up to anything more than digital voyeurism when tackling the dangers of social media?

Let’s take AI and automated content moderation. I understand that the emerging technology has become the digital policy topic du jour (see above). But for years, social media companies have been increasingly using complex algorithms and automated systems to rid their platforms of awfulness. The newly published transparency reports shed light on how far those systems have come. On Instagram, for instance, 98.4 percent of the 76.3 million pieces of content that were removed from April 25 to September 30 were done so via automation. Globally, Google also removed 90 percent of violent content that would have appeared in its search engine’s “discover” feature automatically. Are these good numbers? Do they encompass everything that’s illegal or hateful on these networks? Honestly, we don’t know.

It’s not hard to see that, in a few years, those levels of AI-aided content moderation will have become even more sophisticated — and that the tens of people who currently review such material, manually, will be out of a job. That’s probably a good thing to stop these individuals from having to look at gruesome content online. But it also gives even more power to blackbox AI systems that aren’t really described in depth in any of the reports published by the companies.

Another big downside to this transparency push is that it’s almost impossible to compare one company’s response to another. All of the companies have provided data across different reporting periods. They published the information in often contradictory formats. And there’s a level of over-complexity in how the data is described that leaves the reader less, not more aware, of what’s going on. Case in point: Google provides figures on the number of content moderators, per EU language. But it also blurs what that means by referring to moderators that have looked at at least one piece of content, per language. Does that mean someone who’s reviewed one post in, say, Greek, all of a sudden gets counted as a Greek content expert? It’s unclear and needlessly complex.

After spending the week banging my head on the table while reading these reports, I’m left wondering: What’s the point? I understand this is a compliance requirement, and the publications are not supposed to be bestsellers. But if the aim is to provide average citizens with a greater understanding of how social media work, I, for one, was left underwhelmed. There’s a lot of data available. But what does it actually say? And, to the wider point, how does it improve people’s understanding of how these platforms operate? Europe has legitimately led the way in social media regulation. But saying that you’ve made these giants more transparent doesn’t mean that you’ve actually done so.

THE MESSY WEB OF INTERNATIONAL DATA

WE’RE STILL WAITING FOR MAX SCHREMS, the Austrian privacy campaigner, to file his appeal against the new transatlantic data agreement. That lawsuit, I’m still told, is weeks (not days) away. But while people’s attention has been focused on what happens between data shifted between the U.S. and EU, other parts of the complex web of global data protection arrangements are coming together. These legal agreements allow companies to move people’s personal information freely between jurisdictions that often have varying levels of privacy standards. It’s the closest thing we have to a global data protection regime.

This week, Brussels finalized a data transfer pact with Tokyo. It’s the latest effort by the EU to expand its own data protection regime globally by requiring other countries to alter their privacy standards to align with those of the bloc. Not to be outdone, the U.S.-led Global Cross-Border Privacy Rules — based on existing rules via the Asia-Pacific Economic Cooperation, a free-trade body — also continue to gain pace. Officials are busy at work rewriting the group’s outdated standards, as well as scooping up new signatories like the U.K., to offer a potential alternative to what the EU is pitching to countries.

**Do you want to know more about the Digital Decade 2030 targets and the future of Europe’s connectivity? Join us on November 30 at POLITICO Live’s event “Connectivity for all: EU’s vision for a competitive sector” to hear from our experts. Don’t miss this opportunity and reserve your seat!**

WONK OF THE WEEK

WE’RE GOING TO HEAD TO WASHINGTON THIS WEEK, where Bruce Reed, the White House’s deputy chief of staff for policy, was tapped to lead the Biden administration’s newly created AI Council. That group will coordinate the federal government’s response to the AI-focused executive order that was published Monday.

A longtime confidante to the U.S. president, the Idaho native has spent much of his career as a power player in the Democratic Party — primarily associated with his work for Joe Biden. He was chief executive of the Democratic Leadership Council, and got his start in politics as a speech writer for then-Tennessee Senator Al Gore.

“If he hadn’t already been concerned about what could go wrong with AI before that movie, he saw plenty more to worry about,” Reed told the Associated Press when discussing how the U.S. president became worried about the technology after watching, um, “Mission: Impossible — Dead Reckoning Part One.”

THEY SAID WHAT, NOW?

“It’s very rare that governments across the world get together to talk about technology, and how to respond to some of the challenges,” Ed Husic, Australia’s minister for industry and science, told me on the sidelines of the U.K.’s AI Safety Summit this week. “The big thing I’ve taken out of this summit is the fact that there is a preparedness now, particularly from some corners of the planet that have probably been a bit resistant to this type of notion that governments can involve themselves in the way that technology evolves.“

**Rishi Sunak aims for the U.K. to lead in AI governance. How will it turn out? POLITICO Pro Technology UK sheds light on these critical developments. Unpack the policies and power plays shaping London’s digital future. Sign up for a demo now.**

WHAT I’M READING

— The British regulator in charge of that country’s new online safety laws outlines what the legislation means for how the agency will go about its work in the years to come. Read more here.

— The European Commission needs to be very careful about how it interprets the bloc’s Digital Services Act when it comes to overseeing potentially harmful online content related to the Israeli-Hamas conflict, argues Access Now, a nonprofit.

— Replacing front-line workers, especially those involved in health care, with AI-generated chatbots is problematic and may lead to poor outcomes for often the most vulnerable people in society, claims Mark Tsagas for The Conversation.

— A coalition of governments, including Germany and the U.S., and tech firms like TikTok and SnapChat signed a nonbinding pledge to combat the use of generative AI in child sexual abuse material.

— The European Data Protection Supervisor, the privacy agency that oversees EU institutions, published its data protection recommendations for the bloc’s Artificial Intelligence Act. take a look here.

— New York City Mayor Eric Adams released a blueprint for AI regulation, though the plan does not include any specific new legislation to police the emerging technology.