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

WELCOME TO ANOTHER DIGITAL BRIDGE. I’m Mark Scott, POLITICO’s chief technology correspondent, and no, we’re not going to focus on the mostly hot air coming from those descending on the World Economic Forum’s annual gathering this week. I bring you actual footage of the “killer robots” that many in Davos have been warning will soon replace us.

Move over “global elite.” We’re the 99 percent:

— We need to rethink how political campaigns use data to target us online. It’s not as sophisticated as many of us once believed.

— Competition concerns over how a small group of tech giants now dominate generative AI will be hard to stand up.

— Almost everyone believes Europe’s privacy rules need an update. That doesn’t mean it will happen anytime soon.

DIGITAL DEMOCRACY: DATA-DRIVING CAMPAIGNING

KATE DOMMETT IS HERE TO BURST YOUR BUBBLE — and she has data to prove it. The British academic (and former digital adviser to the United Kingdom’s House of Lords) spent the last year, alongside Australian Glenn Kefford and German Simon Kruschinkski, interviewing more than 200 people across the United States, Canada, Australia, Germany and the U.K. who had worked on the digital side of political campaigns. The researchers’ conclusion? Much of what we think about sci-fi depictions of targeted data-driven online campaigning — akin to the 2018 Cambridge Analytica scandal — is just plain wrong.

“Political parties are doing what they have always done: to collect data to inform an understanding of what the public want,” she told me ahead of the publication of her team’s new book, “Data-driven Campaigning and Political Parties.” “There is a recurrent tendency to have this doom-mongering about every new technology that comes along. The alternative isn’t that glamorous, which is parties have been gathering data since the 1950s, and now, they’re just gathering data in slightly different ways.”

Dommett’s thesis is this: In the wake of the Cambridge Analytica scandal, in which people’s Facebook data was illegally collected and used for political purposes, the public’s imagination has been filled with misconceptions about how sophisticated such online political targeting — especially via social media — has become. In reality, almost all political parties and outside groups are hamstrung by a lack of money, few technical capabilities, domestic regulation limiting the use of people’s personal information, and differing electoral systems that make such “Minority Report“-esque targeting more fiction than reality.

The researchers discovered a wide range of know-how across the five Western countries that they surveyed. The U.S. — with a well-financed domestic industry of consultants and relatively lax data laws — was by far the most advanced. However, even in that country, few, if any, political parties had access to detailed specifics on individual voters. Germany, based on those interviewed, was the most conservative in its use of data to target would-be voters. Strict privacy rules, a historic aversion to collecting people’s personal details, and a federal system that gave much of the power to local party affiliates meant that Europe’s largest economy still used pretty rudimentary data techniques compared with the other countries.

“Some politicians are still very data-averse,” said Dommett, adding that many political leaders still refused to follow the advice of their data boffins — even after investing millions of dollars to build up in-house databases on voters’ intentions. “There’s still a lot of skepticism of policymaking by focus groups from (political) elites.” One major hurdle confronted in all five countries was keeping political data up to date. Few political parties had full-time staff dedicated to such data management, so when the next election came around (as we are seeing during the 2024 election cycle), much of this detailed information on voters’ intentions had become useless.

That’s not to say data-driven campaigning isn’t a thing. Across all five countries, the researchers found novel ways in which political groups had tapped into the data to pinpoint would-be voters. In the U.K., for instance, one of the country’s leading parties found a correlation between an area having a Catholic church with voters likely to vote for them. In Germany, political operatives relied on Facebook’s in-depth knowledge of its users to circumvent strict privacy rules about the types of political information that could be collected on voters. “Between all the countries, we saw huge differences,” Dommett added.

Yet in a year when more than two billion people will go to the polls, the multi-country analysis is a reminder that, for most of us, our perspective of sophisticated political campaigning — via social media, apps and Google ads — is based almost exclusively on what has happened in the U.S., arguably an outlier in how this data-driven politicking is done. A lack of funding, incomplete information on voters’ intentions, and different election systems mean that almost all countries aren’t enmeshed in the types of complex digital campaigning that have captured the public’s imagination.

“Claims that there was this huge, massive, incredibly insightful, valuable data that could be mined just didn’t tally with all of what we know data management, storage and integration,” said Dommett. “Particularly in the context of political parties, which are organizations that tend to be technologically backwards.”

ANTITRUST REGULATORS’ TROUBLE WITH GENERATIVE AI

EVERYONE, INCLUDING ME, HAS FALLEN VICTIM TO the hype around generative artificial intelligence. To that list, we can add the West’s competition enforcers. Over the last six months, the U.S. Federal Trade Commission, the Department of Justice, the U.K.’s Competition and Markets Authorityand the European Commission have been trying to find a way into this emerging technology. Their concern is that a small group of primarily West Coast tech giants will come to dominate, push out smaller rivals and harm consumers. Right now, there’s just one problem with that theory: There’s little evidence to back it up.

Regulators are right to be concerned. The likes of Amazon, Microsoft, Google and Meta have invested billions, either directly in their own generative AI projects or in upstarts like OpenAI (supported by Microsoft); Anthropic (funded by both Amazon and Google) and Inflection AI (backed by Nvidia and Microsoft). Throw in Microsoft, Amazon and Google’s two-thirds global market share in cloud computing, an essential part in powering so-called large language models, and it becomes more concerning. Add to that Nvidia’s roughly 85 percent slice of the global AI chips market, and you can quickly see the emergence of potentially abusive behavior.

But let’s look at the facts. First, the number of foundation models — the most advanced versions of AI systems — currently in the wild. Most people’s attention has been captured by OpenAI’s GPT-4 and, possibly, Google’s PaLM2. But Stanford University academics reviewed 10 competing models, and that’s only the tip of the iceberg. Sure, not all of these models are truly comparable — those from Google and OpenAI use a lot more computing power than many of the others. But to suggest there are no alternatives to the main players, currently, is not correct. May that change? Sure. But for now, it’s still wide open.

Then there’s cloud computing — arguably the mainstay for how these latest AI models will grow in complexity. Right now, Amazon, Microsoft and Google have a combined 65 percent stake in the global market, followed by the likes of China’s Alibaba, IBM and Oracle. That, too, is potentially concerning. But these three Big Tech companies also offer their cloud services to rival AI providers, mostly because it’s a lucrative new revenue opportunity. There are questions about whether such access could be turned off — or slowed down — in the future. But, for now, a cut-throat market of three rivals fighting it out for customers feels pretty healthy to me.

So where should regulators be focusing? In terms of short-term competition threats, Nvidia’s dominance of the AI chips market is pretty acute. The FTC even went as far as to block the semiconductor giant’s takeover of ARM Holdings, a chip designer, over antitrust fears. That certainly is a bottleneck that gives one firm a pretty dominant position. But Intel and AMD are quickly investing billions of dollars to scoop up some of this burgeoning AI chip market — and Western governments are plowing in subsidies in these high-end chips to fuel the red hot market. Again, there’s smoke here, but no clear signs of a fire.

For now, regulators have two clear goals. First: to gauge whether Big Tech’s high-profile investments in the likes of OpenAI, Anthropic and Inflection AI are unfairly crowding out potential competitors. Yet the fact many of these upstart AI companies have taken money, in different funding rounds, from rival Big Tech firms — with the notable exception of OpenAI — makes such a worry harder to stand up. Just because a dominant player becomes an investor does not mean that competition is under threat.

Second: to keep tabs on whether previously anticompetitive behavior becomes ingrained in generative AI. Regulators worry that so-called self-preferencing, or unfairly promoting a firm’s service over that of a rival, may be on the rise. That could happen, say, if Google gave its generative AI tools an advantage in certain search products. Dominant cloud providers could also potentially give better/cheaper access to their computing capabilities to those who also bought these firms’ other AI-based tools or services. All these are potential risks. But the key word here, for now, is potential — and regulators have yet to find concrete cases of wrongdoing.

BY THE NUMBERS

EUROPE’S PRIVACY PROBLEM

DESPITE THE EUROPEAN UNION setting the de facto privacy rules for much of the world (outside China), many within the bloc are still not happy. They view the current system — in which Ireland and Luxembourg have almost exclusive enforcement powers over Big Tech because of where these companies are headquartered — as failing. It’d be better, campaigners and some data protection regulators insist, to give such cross-border oversight to Brussels. That’s what happened with the EU’s new social media laws, known as the Digital Services Act, where the European Commission now has new powers. So why not do the same with the EU’s General Data Protection Regulation?

Over the summer, the European Commission proposedsomething along those lines, where Ireland and Luxembourg would have to share more information on how they were investigating Big Tech cases. Yet it’s still not enough for those who perceive, incorrectly, that mostly Dublin has gone soft on Big Tech enforcement. What they want is a revamp of the so-called one-stop-shop principle, in which the regulator in the country where a company is based handles all oversight. Because these firms have a pan-EU impact, the regulatory powers should also be shared among the many. The aim is to get something on the books in the coming months.

But with the European Parliament elections in June — and the European Commission’s tenure over in November — time is almost gone for any such changes to happen this year, if at all. Those calling for a privacy enforcement revamp must also question if it’s worth it. Since 2018, Europe’s data protection regime has squeezed out more than $4 billion in fines and imposed greater restrictions on aggressive data-gathering practices. In all honesty, it still is not enough. But would changing enforcement to allow all regulators to have an equal say lead to a better outcome? The short answer is no.

WONK OF THE WEEK

ITALY HOLD THIS YEAR’S G7 PRESIDENCY, so it felt only right to focus this week on Gianluigi Greco, head of the country’s coordination committee overseeing the update to Italy’s national strategy on AI, which will be published later this year.

The professor at the University of Calabria, in southern Italy, doubles as the president of the Italian Association for Artificial Intelligence, a nonprofit research body focused on emerging technology.

“The limit between what machines can do and what they cannot do is linked to the type of thinking we ask them to emulate,” he told a local newspaper. “They allow us to optimize processes and also enable new ones. But they cannot help us, and therefore cannot replace us, in all the mental processes.”

THEY SAID WHAT, NOW?

“You could argue that every AI-generated image actually victimizes thousands of real children,” said Marc Berman, a Democratic lawmaker in the California State Assembly, after publishing a draft law that would crack down on AI-generated child sexual abuse material. “Because they are a part of the formula that goes into creating that AI-generated image.”

WHAT I’M READING

— The German Marshall Fund of the United States outlines what it believes will be the major transatlantic technology policy issues for this year. Take a look here.

— About 100 so-called deepfake videos of British Prime Minister Rishi Sunak are circulating on Meta’s platforms, mostly in aid of financial scams, according to research from a former U.K. political operative.

— OpenAI outlines how it is approaching this year’s election cycle. That includes flagging when content is generated via AI and clamping down on abuse. More here.

— The European Data Protection Supervisor outlines its reservations on the data agreement attached to the EU-Japan trading relationship. More here.

— Researchers from Carnegie Mellon and the University of Oxford outline existing and new privacy risks associated with artificial intelligence. Read more here.

— The ability of Brussels’ new digital competition overhaul to be effective relies on implementation based on effectiveness, proportionality and legal predictability, based on analysis from the Center on Regulation in Europe.

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