#26 Is AI as smart as Skynet or as dumb as C-3PO? It depends on who you read/ask...
It's very much the same with lawyers.
AI is either extremely smart (and we’re two days away from artificial general intelligence), or it’s very dumb, hasn’t really proven anything, and it’s all hype… depending on who you read/ask.
On one side of the ring, there’s a highly shared LinkedIn post by Dagmar Monett saying that AI is nowhere near as intelligent as it’s made out to be. And she provides a ton of papers to support her motion.
In the same vein, another even more viral paper “The illusion of thinking” was published with Apple’s support. Its bottom line is that today’s AIs are pretty good with low-complexity problems, but as soon as you give them something truly difficult, they go off track and fail miserably.
In an opposing but not extreme position (my cup of tea), Ethan Mollick responds quite wittily that the problem with Apple’s paper is that it assumes the mentioned limitations are inherent to LLMs, when in fact what’s being analyzed are the consumer versions of advanced models, which have been deliberately limited in design so they don’t consume too many tokens.
Mollick repeats tirelessly—and we agree here—that even if the mythical artificial general intelligence doesn’t arrive before 2525, the innovations we have on the table are already disruptive enough to change all our lives in no time at all.
On the other hand, my humble opinion is that Apple—Apple of all companies—should think twice before criticizing other people’s AIs: theirs are nowhere to be seen.
You’re reading ZERO PARTY DATA. The newsletter about current events, technopolies, and law by Jorge García Herrero and Darío López Rincón.
In the spare time this newsletter leaves us, we like to tackle tricky problems in personal data protection. If you’ve got one of those, give us a little wave. Or contact us by email at jgh(at)jorgegarciaherrero.com
Thank you for reading Zero Party Data! Sign up!
🕮 News from the DataWorld 🌍
– In Poland, the same thing has happened as once did in Almendralejo (Spain): sexual deepfakes where minors are both perpetrators and victims. The Polish data protection authority is cracking down: according to Mateusz Kupiec, Poland’s administrative data protection regulations foresee fines and up to two years in prison, and criminal law allows up to fifteen. We’ll see. The Danes are considering raising the minimum age for social media to 15; we’re going with 16 as the age of consent—and everyone is doing something about minors.
– As we anticipated last week, Radboud University and IMDEA staff led by Narseo Vallina have uncovered Meta’s latest major heist: a big one that could cost it billions in fines. We explained it to you (because it’s a tough topic) this very week in what has been, by far, the most-read edition in Zero Party Data’s short history.
In case you haven’t clicked the link to Tuesday’s special newsletter about Meta’s localhost tracking yet, here’s a screenshot to tempt you. The process is explained in the newsletter, but it always helps to visualize the process flow in an image. Courtesy of the researchers themselves, on the site they’ve built to explain it.
– Not all smart glasses are made by shady companies from a privacy standpoint (a warm hug to Zuck from us!). We’ve also seen the Finnish startup IXI launch a pair of… autofocus glasses. Yes, really. They use an eye-tracking sensor and “liquid lenses” 😶 that automatically adjust according to focal distance, thereby improving visual experience compared to bifocal and progressive lenses, which they aim to replace.
Eye tracking itself is another frontier that’s right there. There’s much talk about future hypotheses regarding neuroprivacy, mental privacy, and neurorights, but we still haven’t nailed the obvious way out. Is no one noticing Meta experimenting with eye-tracking and biosensors in its wide range of glasses? The Ray-Bans you’re thinking of, but also the virtual reality headsets.
Not for nothing, they have the best-selling ones in the sector: Meta Quest aka Oculus. Involuntary eye movements reveal a lot of information: mood, sleep quality, eye conditions, knowing for sure whether you like something… It’s in the patents where the really interesting stuff is—stuff that never makes it to the privacy policy.
– The series "Adolescence" falls waaay short compared to the real case of Rhianan Rudd. A terrible story from beginning to end. I recommend the long read from the Financial Times. For those without access, one way or another, here’s the link to The Guardian.
– The Oregon Attorney General + 28 states have filed a consumer lawsuit to prevent the auction of genetic data stored in 23andMe’s drawers. They were already handling things badly, so it’s hard to believe they’ll seek prior consent in case of bankruptcy.
– “For whatever reason,” Chinese AI providers suspend their systems during college entrance exams (“selectividad” for old-timers, and “Gaokao” in Chinese).
A controversial move, very similar to what LaLiga does in Spain to prevent piracy of matches, but here—with a solid public interest argument. Still, it punishes the innocent along with the guilty.
I’ll bet my bangs we’ll see this happen in old Europe once we reach Chinese-level penetration.
📄High density docs for data junkies☕️
– Do we need a new legal basis for processing personal data in AI training? If you think so, this discussion paper by Christiane Wendehorst is for you. Via Luis Montezuma. I’ll add the constructive (but sharp) critique from Heiko Roth.
– The AEPD has published a guide on federated learning.
– And the EDPB, besides its guidelines on Article 48 GDPR (international personal data transfers at the request of a foreign authority), has also launched what amounts to a full-on training course with lessons, quizzes, and all the trimmings!! Fundamentals of Secure AI Systems with Personal Data by Enrico GLEREAN.
– A 61-page (!!) presentation “summarizing” high-risk AI system regulation under the AIA. By namesake Georg Philip Krog.
– Some recommendations from the CNIL for something I’ve mentioned often and which can be tricky: you WANT to know the diversity of your workforce (just to include the info in your CSR report or to assess promotion or protection measures where appropriate) but you CANNOT ask your employees about their religion, sexual orientation, or other special category data. Recommendations for workplace diversity surveys. Translated into Spanish with a little help from Google. You’re welcome.
– Unsurprisingly, Meta is rejecting its users’ opposition requests regarding AI training on their content, invoking “overriding legitimate interests.” That, in addition to the five-step opposition process they implemented. I only have this to say:
💀Death by Meme🤣
The Elon vs Trump beef special. (We’re not including the direct tweets, because they’ve been overused by now), here are the ones that made us laugh:
And a final one, unrelated, but 120% Musk…
🤖NoRobots.txt o the AI stuff
– Digging deeper into one aspect of the same issue: What is a subliminal technique? via Josuan Eguiluz.
– In this paper Arnoud Engelfriet discusses the difference between “bias” and “discrimination,” comparing the pair to the famous Big Data-era contrast between correlation and causation. Interesting.
– Don’t know how to start using AI? The folks at Anthropic give you ideas based on what they do with Claude.
– The latest paper by Philippe Hacker and Matthias Howeg: The Regulation of Fine-Tuning: Federated Compliance for Modified General-Purpose AI Models.
📄 Paper of the Week
– Borgesius is on sabbatical in Berlin, but still hard at work. And we thank him for it. He just released a paper with Pieter Wolters on the DSA and online advertising. The DSA introduces specific requirements: transparency, a public repository of ads for VLOPs, and a ban on behavior-based ads using sensitive data or targeting children. Also, the general rules on illegal content apply here too.
🧠 Useful Tools 🔧
… not so useful:
– This one is actually useful: how to create permanent links to articles and exposés in eur-lex.
🙄 Da-Ta-Dum Bass!
If you think someone might like—or even find this newsletter useful—feel free to forward it.
If you miss any document, comment, or bit of nonsense that clearly should have been included in this week’s Zero Party Data, write to us or leave a comment and we’ll consider it for the next edition.