Thursday Thoughts on AI + Law (12/14/23)

It's getting close to the end of the year, but things certainly aren't quieting down in the world of AI!

1. Last Friday, the EU negotiators agreed in principle on the provisions of the AI Act. Why it was hard to pass. The MIT Tech Review published some takeaways of the ‘historic’ deal. Not everyone is happy with it, with some questioning the overall cost, the Economist calling it a ‘mixed bag,’ and others asking how well it’ll keep up with evolving tech and how potential loopholes will be addressed. Yesterday, the EC issued a helpful guide to the upcoming law. And here’s what it might mean for IP/copyright.
2. One key point, of course, is that the AI Act doesn’t come into effect until…2025. (Think about where we were in 2021 versus today in AI development.)
3. Federal agencies keep working towards implementation of the Biden AI EO. On Monday, NIST published a set of draft guidance documents for evaluating privacy protections in AI development. Here’s where one law firm thinks the U.S. will go on AI regulation in 2024.
4. The Pope is advocating for AI regulation now.
5. A group of leaders and scholars from MIT published a set of policy briefs explaining how they think AI should be regulated (probably less useful in the EU at this point but may be helpful elsewhere). 
6. We need more efforts like this: Microsoft is reportedly working with labor unions to train employees and find ways to mitigate negative impacts to workers from the broad introduction of AI into the workforce. 
7. According to a GAO report released Tuesday, federal agencies are all thinking about how to use AI but playing catch-up on implementing various requirements.
8. Dr. De Kai published a thoughtful essay in the NYT regarding where AI developers should accelerate and where deceleration might be wise.
9. The NYT also dug into the effective accelerationist movement.
10. New York is investing heavily in in-state chip development.
11. Nvidia has emerged as one of the biggest players in AI startup investments.
12. “Her” was apparently only 10 years ahead of its time.
13. Wild: Oracle has reportedly been struggling to supply cloud compute (due to GPU scarcity) for xAI.
14. A second insurer has been sued for allegedly using AI to deny claims or restrict coverage.
15. The geopolitics of chips is getting more and more complicated.
16. Likewise, AI deployment is increasingly dictated by geography.
17. An interesting proposal: apply learning from Southeast Asian remuneration approaches to generative AI value allocation.
18. Instagram influencing will never be the same now that AI-generated backgrounds are easy to come by.
19. The U.S. is making a big push for AI for global good at the UN.
20. AI is helping advance scientific discovery but there’s a catch: we don’t always know how AI discovers things.
21. AI is helping parents navigate resources available to them.
22. Big law firms are finding ways to advise on AI – DLA apparently has 100 lawyers in their AI practice area?! – but most are probably brainstorming how to account for AI’s impact on billables. Related: Bloomberg columnists are now suggesting that AI might not displace associates, after all. And the AI Act might just mean a lot of new work for major firms.
23. Creating legal exceptions around AI-generated content might be the right approach (or it might not) but it’ll be very hard to implement effectively at the moment.
24. AI-powered astrology seems pretty wild.
25. Mistral announced some significant funding and also released a new, powerful open model via torrent. More on the Mistral release here.
26. I think HBR gets this right: strategy, not tech, is going to be most important for most market participants in an AI era.
27. Reversing course to a degree, the UK government now says they’re planning to unveil AI legislation in 2024.
28. This will be really annoying in 2024: AI robocallers.
29. Is AI the new browser?
30. Great question (I’m interested in the answer): can AI help you find your next running shoe?
31. Noema Magazine asks: how do we draw red lines in AI regulation?
32. The Biden Administration is wading into the open/closed model debate. Good luck! Relatedly, AI topics now prompt cabinet-level meetings at the White House.
33. Investing.com got caught using AI to create plagiarized versions of articles from other investment sites (and the team at the Motley Fool, where some of the original articles may have been sourced, had a great response).
34. German AI startup Aleph Alpha is doing really well, but the German press is reporting that their models have some content moderation issues.
35. Speaking of Germans, OpenAI is opening up ties with publishers (and announced a big deal with Axel Springer).
36. Lots of attention is being paid to AI model/system governance, but what can be done if the model and its applications are highly distributed?
37. The NYT hired a director for AI initiatives.
38. Geoffrey Hinton has some interesting views on LLMs in medicine.
39. Visual Capitalist put together an interesting map of search volume for generative AI topics.
40. Synthetic data without differential privacy seems to be prone to attack.
41. It’s one thing to avoid live demos; it’s another to cobble together a staged demo.
42. Oxford talked with Professor Helen Margetts on the importance of developing a social understanding of AI.
43. When companies announce testing results for LLMs, Nathan Lambert cautions that you should read the details.
44. The WSJ strangely argues that AI should be more spicy.
45. Platformer published its predictions for 2024 and (unsurprisingly) AI features heavily.
46. More and more data (this time, regarding X-rays) indicates that AI might be physicians’ best tool in the future.
47. But patients aren’t so sure about doctors’ use of AI tools.
48. And HHS is starting to think through regulations.
49. Sign of the times? Kevin McCarthy’s next role may be in the AI space.
50. Be alert: there will inevitably be holiday-themed, AI-powered scams going on over the next couple of weeks. You don’t want to be a late addition to this list.