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

JUST LIKE A WONKY ADVENT CALENDAR, there are only two Digital Bridge newsletters to go for 2023. I’m Mark Scott, POLITICO’s chief technology correspondent, and I bring you live footage of why my family hates it when I turn my journalism spidey senses toward domestic matters. True story.

We’ve got something for everyone this week:

— Two months into the war, there’s a lot more pro-Palestinian content than pro-Israeli material. What that tells us about social media is unclear.

— The European Union’s Artificial Intelligence Act is (kinda) done. What happens next is a case study in why Brussels isn’t good at enforcing its digital rules.

— Section 702 of the U.S. Foreign Intelligence Surveillance Act is likely to get a four-month reprieve. But will that be enough to keep it alive?

THE FOG OF A DIGITAL WAR

IT HAS BEEN JUST OVER TWO MONTHS SINCE Hamas militants attacked Israel, killing about 1,200 people and taking about 240 hostages, according to Israel. In response, Israel invaded Gaza, leaving at least 15,000 Palestinians dead, based on estimates from local health authorities in Hamas-run Gaza. The suffering, for both sides, is unimaginable — and it looks unlikely to stop anytime soon. Within this cacophony of awfulness, social media has become an an active participant — used by those supportive of Israel and Palestine alike to promote their side of the conflict; raise awareness; and, often, peddle falsehoods.

In that sense, the Gaza conflict is no different from other 21st-century wars where dominating the digital media landscape is now part of combatants’ overall strategy to win hearts and minds, engage with like-minded supporters, and count claims from opponents. But what sets the Middle East war apart — alongside this being the first real test of Elon Musk’s ownership of X, formerly Twitter — is how disproportionate the online messaging has become. It’s hard to generalize because of how unique everyone’s social media experiences have become via algorithms tailored to individual interests. But, on average, there are just much more pro-Palestinian messages on the likes of Instagram, TikTok and YouTube compared with pro-Israeli posts.

Let’s look at the stats. On TikTok, I gathered roughly 50 hashtags that were either pro-Palestinian, pro-Israeli or viewed as used by both sides. This is not a perfect science. But it’s true videos with the #StandWithPalestine hashtag garnered around 4.8 billion views, based on TikTok’s own data. In contrast, the #StandWithIsrael counterpart received just under 600 million views. Similar disparities could be seen with everything from #PrayforPalestine and #PrayforIsrael to #ProPalestine and #ProIsrael. This isn’t just a TikTok phenomenon. On other platforms like Facebook, there’s an equal asymmetry in which there’s more pro-Palestinian content being created compared to pro-Israeli material, based on Digital Bridge’s analysis via CrowdTangle, a social analytics tool owned by Meta.

Yet for me, this “hashtag comparing” doesn’t really tell us much. In response to criticism, tech giants like TikTok argue such analysis doesn’t capture the nuance of how each hashtag is used. I hate to admit it, but that’s correct. But I have different reasons than TikTok for suggesting hashtags aren’t a good proxy for tracking how the Gaza conflict is playing out online. For the companies, it’s all about context. How long someone spends on a video, they argue, is arguably more important than how many overall posts were created. It doesn’t matter if there are 20 times more pro-Palestinian posts, for instance, if pro-Israeli counter-messaging is getting more eyeballs.

And that is where my frustration lies. In the end, the volume of social media content produced is meaningless. What matters is the type of material served up to people via their online feeds. And that almost exclusively comes down to a platform’s so-called recommender systems, complex algorithms that decide what tiny fraction of the billions of pieces of content produced daily will make it in front of people’s eyeballs. Just counting hashtags isn’t going to get under the hood of those systems. What you need to do is look beyond the collective total amount of pro-Palestinian and pro-Israeli content, and figure out exactly how much of that is making it into people’s feeds.

Leaked documents and corporate whistleblowers suggest those decisions are geared toward displaying posts that people will engage with the most and, therefore, stay on each respective platform longer — generating more ad revenue. It’s an agnostic process, though one that — at least for Meta’s platforms — led to more polarizing content being served up because internal research confirmed such material was more likely to engage people than less divisive posts. It’s also an ecosystem that we have almost no information about because tech giants jealously guard these algorithms because they are the secret sauce for how, in the end, they make money.

But we can make educated guesses on how this is shaking out. Based on the volume of pro-Palestinian content compared with pro-Israeli material (itself, an imperfect science), we can surmise more people, globally, are engaging collectively with those posts — just given the sheer amount of content produced. If that’s true, we can also argue platforms’ algorithms will likely favor that content — though we have no evidence for that yet. That’s because these algorithms are geared toward displaying material that will likely engage users. If the volume of pro-Palestinian content is greater, and social media algorithms are focused on meeting people’s digital interests, then — hypothetically — it’s likely that companies’ recommender systems are over-indexing on pro-Palestinian messages to fill that need.

Yet, for now, that is just a working theory — one that companies often refute and that currently lacks quantifiable data to stand up. But compared to the “hashtag hunting” that currently dominates the debate, it’s a hypothesis based on how social media actually works.

AI ACT: SHOW ME, DON’T TELL ME

THAT AURA OF SMUGNESS THAT YOU CAN FEEL is politicians and officials in Brussels congratulating themselves for getting a political deal done on the bloc’s Artificial Intelligence Act. To be clear, no final text is yet complete, though I’m told the outstanding details are merely minor tweaks — nothing substantial. Here’s the expected timeline: The final document will be approved in the second quarter of 2024; all-out bans on certain AI uses (more on that below) take effect in early 2025; and by the summer of 2026, the entire piece of legislation is fully enforceable. In short: We’re still at least a year away from seeing any bite to what has been agreed.

First, let’s give the Europeans their due. This year has seen multiple voluntary commitments on AI governance, most notably what the White House published in July. But the EU has gone further than others by making its rulemaking mandatory. Here’s an overview (also here) of what was decided. But, in short, controversial AI-based systems like social scoring will be banned; high-risk use cases like those associated with health care will be highly regulated; almost all mundane AI systems will have minimal to no oversight; and fines range between 7 percent of a company’s yearly turnover (not revenue) for the worst abuses to 1.5 percent of annual turnover for more minor misdemeanors.

Like almost all of the bloc’s new digital legislation, officials and lawmakers have focused almost exclusively on transparency and accountability to ward off wrongdoing. Companies developing high-risk AI use cases, for example, will have to document their activities, conduct outside audits and demonstrate human oversight of how these systems are developed. When it comes to the latest so-called general-purpose AI models (those requiring computer power greater than 10ˆ25 so-called floating point operations, or, basically, only those currently developed by Alphabet and OpenAI), these systems will also have to be tested by outside groups dedicated to finding potential vulnerabilities.

So far, this all sounds pretty familiar to existing voluntary commitments that every large Western AI developer has agreed to. What’s different with the European approach is that it’s all mandatory — and comes with significant fines for wrongdoing. And that, for me, is where I turn from optimist to pessimist on what Brussels announced at the weekend. The EU’s strategy is based on a false promise that, via countries’ existing agencies and a new AI Office to be created within the European Commission, there’s enough technical skill, financial resources and regulatory capacity to both keep track of existing models and keep ahead of what is to come.

If recent history has taught us anything, that is wishful thinking. The ongoing problems of enforcing Europe’s privacy regime, known as the General Data Protection Regulation, has come down to a lack of money and expertise — coupled with intra-EU regulatory in-fighting — that has left the bloc’s data protection legislation less than the sum of its parts. Europe’s social media rules, known as the Digital Services Act, are equally heading toward similar issues. That’s even after the bloc included an ongoing company levy (estimated to be $37 million a year) to pay for oversight that, like the AI Act, is predicated on transparency and accountability as key drivers for enforcement.

For me, what has been missing from the AI Act’s victory lap are details of how this is all going to work in practice. I want budget numbers. I want figures on new regulatory hires. I want an explanation of how these agencies will push back against companies’ transparency reports that may, or may not, be accurate. Recent research shows firms are pretty opaque about how their AI models work. So far, only Spain has announced the creation of a dedicated AI regulator. Other EU countries could follow, though will likely rely on already-stretched existing agencies to do the heavy-lifting on enforcing the bloc’s new AI legislation.

The same questions swirl around how the Commission’s new AI Office will function. Nominally, this yet-to-be created body will coordinate pan-EU oversight; enforce the rules specified for general-purpose AI; and oversee an outside group of AI experts to help with its work. So far, no budget for this work has been set — at a time when Europe’s finances are already stretched. Rumors abound that funding for the AI Office may come from the bloc’s Digital Europe Program, or $8 billion fund to help digitize the Continent. Still, until that happens, I fall back to the old journalistic cliche: Show me (how this is going to work), don’t tell me.

BY THE NUMBERS

SECTION 702 EXTENSION DOESN’T SOLVE THE PROBLEM

SECTION 702 OF THE US FOREIGN INTELLIGENCE ACT SPLITS OPINION. The rules allow American security agencies to surveil foreign nationals (and, controversially, some U.S. citizens) anywhere in the world. The provisions, though, will expire on December 31 — something that both Congress and the White House don’t want to happen. Still, there are currently two competing bills to overhaul Section 702, which has become a lightning rod for privacy campaigners over claims it is used to unfairly collect reams of data on primarily non-Americans. One, via the House of Representative’s Judiciary Committee, would make it harder for data to be collected on U.S. citizens. The other, via the House Permanent Select Committee on Intelligence, would expand some of Section 702’s overall remit.

In a last-minute move, Section 702 is expected to be given a four-month extension under the U.S. National Defense Authorization Act, or must-pass legislation before the end of 2023. That would give U.S. lawmakers until April 19 to find a compromise. A lot of the controversy around Section 702 comes from some in the Republican party who believe the provisions allow for surveillance of U.S. citizens with little oversight. They cite the invalidation of two of four FISA-based warrants against Carter Page, a former Donald Trump aide, as evidence of such overreach. Importantly, no one in Washington really mentions how non-U.S. citizens — some, arguably, innocent — have similarly been caught up in such bulk surveillance.

WONK OF THE WEEK

WE’RE GOING SLIGHTLY OUT OF MY TYPICAL COMFORT ZONE this week to focus on Tim Sweeney, founder of Epic Games, the studio behind the blockbuster Fortnite. His company won a landmark competition case Monday against Google after a judge ruled the search giant had used anti-competitive practices via its Android operating system and app store to harm rivals and consumers. Google said it would appeal the decision.

Sweeney is an unlikely antitrust champion. A graduate of the University of Maryland, he stumbled into gaming while working as a software consultant in the early 1990s, before rebranding his company to Epic Games. Fortnite was created in 2017, but Sweeney has been fighting battles with both Apple (in a case that he lost on nine out of 10 charges) and Google over how these tech giants pocketed a share of revenue when users accessed the game via the firms’ dominant app stores.

“This is an example of the greatness of the American justice system,” Sweeney wrote on X, formerly Twitter. “A [billion-dollar] company challenges a [trillion-dollar] company over complex antitrust practices, and a jury of 9 citizens hears the testimony and renders a verdict.”

THEY SAID WHAT, NOW?

“If the United Kingdom is to avoid being held hostage to fortune, it is vital that ransomware becomes a more pressing political priority, and that more resources are devoted to tackling this pernicious threat to the U.K.’s national security,” Margaret Becket, chairwoman of the U.K. parliament’s joint committee on national security strategy, in response to a report highlighting the country’s vulnerabilities to such cyberattacks.

WHAT I’M READING

— The U.K. government accused groups with ties to the Kremlin of trying to interfere in the country’s democratic institutions by targeting local politicians, journalists and officials with cyberattacks. More here.

— Stanford University published its inaugural Emerging Technology Review, a deep-dive into 10 emerging technologies (and not just artificial intelligence) and their policy implications. Take a look here.

— Hausfeld, a law firm known for filing lawsuits against Big Tech firms, filed a class-action case against Google’s alleged misuse of copyrighted material from publishers, including for the company’s generative AI products. More here.

— To combat widespread misinformation, society must move away from trying to have better facts about contentious issues and focus on how these problems can be framed, often online, to portray legitimate topics in false ways, argues Kate Starbird from the University of Washington.

— A group of liberal U.S. politicians called on Joe Biden to support the EU’s digital competition overhaul and reject claims it represented an illegal barrier to trade. More here.

— More than half of Brits, 44 percent of French people, and 33 percent of Germans said they viewed social media as hurting democracy, according to a poll commissioned by Luminate, an advocacy group.

SUBSCRIBE to the POLITICO newsletter family: Brussels Playbook | London Playbook | London Playbook PM | Playbook Paris | Global Playbook | POLITICO Confidential | Sunday Crunch | EU Influence | London Influence | Digital Bridge | China Watcher | Berlin Bulletin | D.C. Playbook | D.C. Influence | All our POLITICO Pro policy morning newsletters

https://www.politico.eu/newsletter/digital-bridge/does-social-media-favor-palestine-over-israel/