Really loved working on this! xuan (ɕɥɛn / sh-yen) Micah Carroll and Hal Ashton were a pleasure to think with. I will write a longer post tomorrow but please read our paper or at the very least skim the summary tables

Criminal responsibility for the free actions of another autonomous person (to whom maybe there exist some sort of special responsibilities). These are interesting developments in the application of criminal law brought about by repetitive tragedy. bbc.co.uk/news/articles/…

It's been so long since I've been here I can't remember who the goodies and baddies are.

xuan (ɕɥɛn / sh-yen) Micah Carroll Matija Hal Ashton I would like to highlight the work of two of my PhD students you might be interested in: - the work on moral alignment of AI agents by Elizaveta Tennant (Karmannaya) (e.g., arxiv.org/abs/2410.01639); - the work on (ir)rationality of AI agents by Olivia Macmillan-Scott (e.g.,: royalsocietypublishing.org/doi/full/10.10…).

I think one interesting thing about sharing a 26-page paper on this website is that no one really reads it but everyone already has takes

New UCL Faculty of Laws Centre for Criminal Law event with @hirsch_ph on 'Assessing mens rea: How Criminal Law Disapproves of Our Reasons for Action' with comments by Levin Güver on Nov 25, 6:00pm. Details and booking here: eventbrite.co.uk/e/assessing-me…. Will be🔥 (Is that how you do it?)

whoop whoop 🚨 - Unsavoury LLM behaviour through human manipulation. Still.

Our paper (xuan (ɕɥɛn / sh-yen) Micah Carroll Hal Ashton) is out in Philosophical Studies link.springer.com/article/10.100…

Algo Trading Book Recommendations Some of the books I think should be on Algo Traders’ shelves (with some idiosyncratic commentary). docs.google.com/spreadsheets/d…

In our next post, this Friday, Mark Dsouza (he/him) argues against recent suggestions that the criminal law should consider treating AI entities as autonomous agents, capable of being criminal defendants in their own rights. Look out for that!

This article is well worth a read. I reserve judgment on the matter though the work I did with Matija certainly suggested the public are more than happy to consider robotic actors as "criminal" when they behave badly and are imaginative on the subject of punishment.

joe914 x.com/colin_fraser/s…

It's unfortunate that the "utility maximization" formalization of having a goal (or being usefully described as having one) has displaced other conceptions of goal-directedness that are both more general & more like the folk concept, especially in AI alignment discourse.

📜New(ish) Paper📜with the great people over at Contextual AI "LMUNIT: Fine-grained Evaluation with Natural Language Unit Tests" We introduce natural language unit tests, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring

AI harm insurance is a inevitable future. But we generally only buy insurance when there is risk of a financial loss in prospect. Unfort Harms are often too distributed and imposed upon people powerless to seek redress. Even if there was a body of case law to support them.

Nice read. I would go a lot further than this, but then I'm reliably wrong about many things and I've got some chores to do.