AI, Law, and Otter Things - Issue #10
Hello, dear reader, and welcome to another issue of AI, Law, and Otter Things! This time, I write a bit about the paper on humour and AI that I mentioned in the previous issue and briefly mention what else I've been up to in the last few days.
As I mentioned last week, the previous issue of this newsletter was a bit more fragmentary than usual because I was participating in a workshop on AI, Business, and Human Rights. The event had five panels in two days, skillfully organised by Lottie Lane (Groningen) and Antoine Duval (TMC Asser), and Renata Shimbo and I presented a paper on humour and content moderation. In that paper, we explore the various factors that make automated moderation struggle with humoristic content: the contextual dependencies of humoristic discourse, the lack of legitimacy of private stakeholders in setting standards and priorities for moderating humour (a subject that is challenging even for humans), and the various biases and perspectives that are encoded during the life cycle of moderation systems. Because of that, full automation of content moderation is likely to substantially constrain online humour, with particularly nocive effects to minoritised groups.
Given that laws requiring some sort of content moderation, such as the German NetzDG, often impose tight deadlines, automation is often touted as a way to avoid human biases and reduce the costs of compliance with moderation requirements. Both expectations, however, are frustrated in practice. Bias cannot be reduced to a technical problem; in fact, framing things in terms of bias can distract us to important questions of power involved in the design and use of computer systems. Furthermore, the (fundamental?) challenges involved in computational modelling of humour means that many decisions will require some sort of human review, thus reintroducing the human labour that such practices (if taken in good faith) are meant to avoid. Finally, even adequate human-AI integration (which is far from granted) will not be sufficient to avoid direct and indirect constraints to the freedom of speech in online environments.
As a consequence, automated humour moderation is a topic that falls right into the intersection between Renata's research interests and mine: while my current research direction focuses on how the law should approach the technical aspects of AI, hers deal with the current undertheorisation of humour as an element of freedom of expression. So, our collaboration was quite an interesting experience, which we intend to further develop (any feedback is welcome!).
My last week had quite a few other events, only a few of which are good material for this newsletter. On the technological front, I had some progress with my pure-text workflow: I could not get the Zotero API to work with my emacs on WSL, so I am currently relying on manual exports of .bib files to handle my bibliography. Still, I finally managed to tinker my LaTeX headers a bit to ensure a smooth export from org-mode into a proper Oscola-style paper, as a stopgap solution while org-cite does not come into play.
Now that the academic year has started, I am working with my colleagues Francesca Palmiotto, Sarah Tas, and Natalia Menéndez to prepare a seminar called AI in Practice: Legal and Technical Perspectives. Our goal with this internal EUI seminar is to explore connections between legal and technological debates on AI, with special attention to how the AI Act introduces some software engineering concepts — such as the life cycle of a software system — into legislative text. Since I am trying to avoid overextension, my other collaborative work is advancing slowly, but I will hopefully have more to write about it soon.
I also have followed up on the topic of future-proof regulation, which I touched upon in some previous newsletters. Twitter users have been very helpful in recommending stuff about the effectiveness of predictions about technological developments, and it seems very likely that this topic will appear in future newsletters before finding its way into my work. In the meantime, I would like to thank Maaike Verbruggen and Matthijs Maas for their insights and reading suggestions on the topic.
Beyond work, I am finally done with my first run in Crusader Kings III, in which I played with Tuscany from 1066 to 1453. It was a very immersive experience, as Paradox games usually are: hence, it seems unlikely I will come back to Crusader Kings soon, despite having enjoyed it a lot. I have also decided to re-read This Is How You Lose The Time War, an epistolary novel with much romance and time travel. It is a short book, of the kind you can finish during an insomniac night, and incredibly heartwarming. If any of my sci-fi-minded readers have not checked this one out yet, I strongly recommend it.
Last but not least, I watched the entirety of Ted Lasso with my wife, who had not seen the show yet. That show is usually hilarious, and the second season managed to pick itself up after a slow start. Even though it is a football-themed show, its greatest emotional highs and lows come from the humans around the game, not from the matches themselves. And this is something I think the series gets right, unlike many other fictional representations of football: a film, or even a written narrative, is unlikely to capture what makes the game itself such a thrilling experience, which means that any fictional matches tend to be trite and underwhelming. But, if you focus on football as a human endeavour, you can tell powerful stories, and that is why I think Ted Lasso might be interesting even for those of you who are not really fond of the beautiful game.
Speaking of football, Bayern Munich is about to play FC Barcelona as I finish these words (update: Bayern won, 3-0, with a typical Thomas Müller goal and a very ugly kit). So I'll conclude here (saving you all from a rant about humour and my writing in English), and let's stay in touch!