Hello, dear reader! I was positively surprised by how many people read the previous issue and glad that at least a few of you decided to stay. Welcome aboard, and I hope this newsletter continues to deliver interesting content. As always, I am more than happy to engage with old and new subscribers about particular topics, feedback on the newsletter, and other stuff that you might be interested in.
Today’s newsletter, unlike the previous one, is more of a collection of sparse notes on topics related to my research and personal interests. This does not mean I gave up on the monographic newsletter format; in fact, there are a few of them in my draft folder. But the last few weeks have been a bit intense in terms of deadlines; as a result, these texts are not ready for discussion yet. Instead, I will share updates about my current work, some suggestions for reading related to law & AI, some sci-fi recommendations, and the usual cute otter by the end of the newsletter.
Before that, however, Winnie wants to say “hello”:
Works in progress
As the previous issue suggests, I’ve been thinking a lot about regulation by design lately. In fact, that issue draws heavily from content I’ve cut from two texts I’m working on. The first one (with Juliano Maranhão and Giovanni Sartor) is a commentary on Article 25 GDPR, which is already at the post-proofing stage and should be out soon. The other is a single-authored paper, in which I argue that regulation by design can be an obstacle to long-run governance, to the extent that it entrenches legal rules into our digital infrastructures in ways that do not reflect the dynamism of text-driven law. These effects of entrenchment, however, can be avoided—or at least mitigated—if regulation-by-design provisions tackle them head-on.
My second line of work has been related to the role of AI systems in administrative decision-making. A few days ago, I submitted a paper on how AI systems, often presented as an antidote to the uncertainties surrounding decision-making in complex scenarios, can be themselves a source of uncertainty, and how that uncertainty can be addressed at the level of individual decision-makers using AI. I am now working on the follow-up to that manuscript: a doctrinal unpacking of the implications for administrative law in the EU of a recent batch of empirical work showing that automation does not eliminate administrative discretion. In light of these forms of use of AI law, I believe we need to complement high-level debates on “AI as delegation” with proposals for decision-level safeguards, which are not as common as they should be in current doctrinal scholarship.
I am also working on broader issues of AI regulation. With Nicolas Petit, we are putting the finishing touches on a paper that draws from law & technology scholarship to analyse the mismatch between the fundamental rights goals the AI Act seeks to pursue and the product safety tools it uses to do so. I am also putting the finishing touches on a Portuguese-language book chapter on the role that the Brazilian data protection authority can play in addressing certain gaps in the regulation of automated decision-making.
To develop these works, and their connection with my thesis, I have benefitted from participation in some events. Since the last newsletter, I have presented the work-in-progress on regulation by design on two occasions: one (remotely) at the Workshop on Long Term Risks and Future Generations in Brussels and one (in-person) as part of my presentation at the conference Living Together in the Cities of the Future and the Metaverse in Lisbon. The Lisbon Digital Rights and Freedoms group kindly gave me the opportunity to present and discuss my draft paper with Nicolas Petit. Last but not least, the administrative law work was discussed at the Doctoral Workshop in European Law organized at the EUI. I am grateful for the feedback I got from these lively debates.
Most of the writing mentioned above is not yet available to the public. But, if you are interested in something, feel free to get in touch, and I can share a draft.
A few recommendations
Much of the AI-related buzz since the last issue of this newsletter has been directed at generative models. Google announced a text-to-music system they claim to be too dangerous for immediate release. If one takes the press releases at face value, this posture might be as much a response to people overriding any safeguards as it is a savvy marketing move.
Microsoft and Google have both presented the integration of large language models into their search tools. The idea of writing search queries in something resembling natural language is, of course, tempting, and would make internet use easier now that search engines are crappy to use. But I wonder how monetization will work here, even if one can address the factual errors and tendencies to spout bullshit that plague language models in general and Google’s search demo in particular.
(On a related note, some people have suggested Elicit as a tool that can help with literature reviews. I tend to be something of a latecomer to these technologies, so I haven’t tested it yet, but look forward to doing so).
Concerns with AI-generated bullshit can impact legal practice in various ways. Last week, a Colombian judge claimed to use ChatGPT to draft answers to legal questions that appeared in a case about whether an insurance company should cover all the costs for the treatment of a child. This use does not amount to automated decision-making, as the judge evaluated the system’s output and added relevant legal references.1 Nonetheless, such use overlooks the unreliability of the outputs produced by such systems, as Juan David Gutiérrez points out in an interview about the case.
Academic writing
Here are some pieces by scholars that I read recently and might be of interest to readers:
Joe Collenette, Katie Atkinson and Trevor Bench-Capon, ‘Explainable AI Tools for Legal Reasoning about Cases: A Study on the European Court of Human Rights’ (2023) 317 Artificial Intelligence 103861.
Philipp Hacker, Andreas Engel and Theresa List, ‘Understanding and Regulating ChatGPT, and Other Large Generative AI Models: With input from ChatGPT’ (Verfassungsblog, 20 January 2023).
Albert Sanchez-Graells, ‘Regulating Public and Private Interactions in Public Sector Digitalisation through Procurement: The Clash between Agency and Gatekeeping Logics’ [2023].
Federica Russo, Eric Schliesser and Jean Wagemans, ‘Connecting Ethics and Epistemology of AI’ [2023] AI & SOCIETY early access.
Friso Selten, Marcel Robeer and Stephan Grimmelikhuijsen, ‘“Just like I Thought”. Street-Level Bureaucrats Trust AI Recommendations If They Confirm Their Professional Judgement’ Public Administration Review forthcoming.
A few interesting newsletters
Despite the news section above, I don’t write very often about recent developments in law and technology. Since it is important to keep tabs on those, I would like to suggest a few newsletters that do a better job of it than I could:
Digital Rights from Below: on digital rights, activism, and the Global South
Last Week in AI: quick, weekly overview of technical developments
Tech Brief: EURACTIV’s newsletter on EU technology policy
The Ethical Reckoner: focus on the tech industry
Beyond work
In terms of audiovisual content, I don’t have much to recommend. I am currently following the second season of The Bad Batch to get my Star Wars animation fix, but if I am being honest, it is not my cup of tea. Aside from that, I’ve speedran Cunk on Earth (funny and good to annoy your historian friends), and I am now a latecomer to Orphan Black, just starting the final season. But soon I will need to find new stuff to watch.
When it comes to books, I’ve read Life Beyond Us, a collection of short stories about astrobiology. Each story focuses on a lifeform alien to us, mostly dealing with first contact with sentient species. Each story is also paired with a short essay by a scientist discussing the plausibility of the proposed lifeforms and the scientific implications of the story. As usual, some of the stories are way more interesting than others—I particularly enjoyed Lumenfabulator, The Last Cathedral of Earth, In Flight, and Forever the Forest—but the collection was very good overall.
I am also halfway through The Spare Man, which might be interesting for people who enjoy murder mysteries. As somebody who doesn’t really care for the genre, I picked up the book because Mary Robinette Kowal wrote it. It is good enough for me not to stop reading, but it did not click with me in the same way as Shades of Milk and Honey (to stick with MRK’s incursions into genres I don’t usually read). Still, it is good to try something different from time to time.
And now, the otters
See you around!
As you probably know already, ChatGPT is at its worse when it comes to citing external content, often making up citations altogether.
Thanks for the shout-out! Have got your papers on my reading list, looking forward to checking them out.