Thursday Thoughts on AI + Law (8/10/2023)

Another week, another newsletter. A great deal happened this week, more on the AI-powered discoveries and business side of the house, rather than political and legal developments. But regardless as to your interest, there is ample material to sort through again.

1. The EU AI Act isn’t in effect yet, but AI developers in the EU are far behind their peers elsewhere in funding. Does the AI Act calcify the disadvantage here, or is it an opportunity for AI developers on the continent to lock in gains from ‘compliant AI’? Also, in these conversations, who is looking out for the smaller developers?
2. If you’re starting out as a lawyer or risk mitigation team advising AI developers, the IAPP has a good list of questions you may want to ask regarding any proposed products/services.
3. EPIC is cataloging all of the different U.S. state law proposals regarding AI. And let’s just say state legislators are asking lots of questions. At the federal level, people are asking whether anything will get done. (More on that.) In part, it’s hard because any regulation that passes in the U.S. can have serious consequences abroad.
4. AI is generating a lot of hype and market value, but it isn’t generating a ton of revenue (yet). But A16Z, fresh off their crypto high, sees the potential value that will come down the line.
5. AI growth is just beginning and it’s going to be exponential. Dario Amodei has some views on that point.
6. Explosive demand for AI is prompting serious growth for data centers. And Nvidia chips, which are now so in demand that they can be used as currency-equivalents for purposes of collateral. When people say that it’s hard to get ahold of the Nvidia H100 GPUs, what does that actually mean?
7. Speaking of Nvidia, they’re now making it easier for developers to experiment with AI before scaling.
8. France is apparently betting big on open AI model development. Meanwhile, the CNIL (France’s data protection authority) is engaging in a public consultation regarding the intersection of AI and the GDPR.
9. If you’re using an AI-powered tool to help make hiring decisions in the U.S. and that tool is configured to auto-reject candidates over a certain age, you’re probably going to run into legal issues.
10. If not everyone is using AI tools, those who are using them might have an advantage and they may be somewhat quiet about that use.
11. The Pope, in his 57th World Day of Peace message, focused on AI.
12. AI is helping airlines reduce their contribution to climate change.
13. Generative AI can create art. Can it create architecture?
14. Generative AI often creates incorrect information (sometimes referred to as ‘hallucinations’). Are these errors inherent in neural net transformer-based models?
15. Zoom updated its Terms of Service and, among other updates, included fairly ordinary provisions relating to the use of data for AI training. As expected, lots of people freaked out. Zoom felt it had to issue a response to the matter, and then eventually walked back the changes.
16. Speaking of TOS updates, the NYT updated theirs to combat scraping intended to benefit AI development.
17. AI has biases…including political ones, apparently.
18. Google and Universal Music are reportedly working together to protect artists from deepfakes.
19. This’ll be interesting: consumer products companies are investigating generative AI uses.
20. Mosaic, a VC firm, published on their investment framework relating to AI.
21. Here’s an interesting post on how to evaluate a model’s fitness for various purposes.
22. If you like AI and you like podcasts, Pete Davies created a useful list for you!
23. Bloomberg argues that questions over the use of AI have become existential in Hollywood.
24. Fun with open models: Stability removed an artist from Stable Diffusion datasets but other artists created a tool to reintroduce his style.
25. There are always going to be cheat codes, it seems. The latest? It allows for circumventing the safety rules for different AI models/services.
26. Responsible AI advocates argue that current generative AI uses are harmful to consumers and workers.
27. The London Interdisciplinary School created a phenomenal video highlighting the issues of representational and other biases in AI image generation.
28. DefCon has an “AI Village” this year.
29. The sooner Meta moves on from the metaverse and goes full bore on AI, the better it’ll be for their shareholders.
30. Is literally “every” team at Amazon working on generative AI? Seems a bit suspect.
31. I think the Neuron is right: prompt engineering is overrated (I think it’s a problem that AI developers will solve through better UX).
32. Regulators in Massachusetts are looking into how financial firms are using AI.
33. The American Bar Association was reportedly planning to (but did not) publish an OpEd calling for regulatory reform re: the use of AI in law. 
34. It’s no surprise that Americans are divided over the use of AI. Many priors are probably being applied. One general consensus is that the government is more trusted than businesses with respect to AI use at the moment.
35. wAIt Disney?
36. Good: Karya and others are trying to make AI development succeed as a means of economic development.
37. Microsoft released a blog post describing how they’re approaching red-teaming for AI deployment.
38. AI is expensive and large accounting/tax firms are using large investments in the space to potentially distance themselves from smaller rivals.
39. Toronto is aiming to become a big AI hub.
40. Alation’s CEO published some thoughts regarding AI biases and barriers to AI adoption.
41. Alibaba open-sourced two LLMs last week.
42. An essay in Scientific American suggests that IP law should evolve to account for nuances in (and challenges posed by) AI.
43. If foundation AI models present systemic risks, then what can be done to increase confidence in the deployment of those models? 
44. Why does AI struggle with generating realistic images of certain body parts, like hands?
45. Yikes! AI can fuel anorexia and other eating disorders. 
46. AI is being used to create mind-blowing antibodies. And recreate extinct antibiotics.
47. Hospitals are using AI tools to combat caregiver and physician burnout.
48. Korea’s privacy commissioner sanctioned OpenAI for a data protection issue.
49. Facial recognition AI should not be used as a predicate for arrests. But perhaps it’s good to use it to detect guns?
50. The British Army is using AI for recruitment. Meanwhile, the U.S. DoD is saying that AI will not make lethal determinations in combat (oh?), but they will use AI to defend online infrastructure. More here. I’ll reiterate: autonomous weapons should be banned.
51. If we can’t detect AI-generated content, and AI can’t either, then what will AI train on?
52. AI might just help us save the planet from killer asteroids.
53. And more signs that AI may save SF’s downtown.
54. But AI needs to clean up its environmental impact.
55. Last week I mentioned financial firms going gangbusters over AI. Of course, this also applies to the fintech sector.
56. The Center for American Progress issued a priorities statement for AI regulation and strategy.
57. OpenAI filed for a trademark in the U.S. for “GPT-5.” They have reportedly also deployed a new crawler, which respects robots.txt, to gather data, and are funding a journalism ethics initiative.