Thursday Thoughts on AI + LAW (9/7/23)

It was a shorter work week here in the U.S. but that doesn't mean it was short on news. Hope you all find this as interesting as I do. And, as always, feel free to share with anyone else who might find it interesting or useful!

1. The inaugural Time AI100 was released this week, with lots of obvious (and some non-obvious) choices. In the Time issue: Walter Isaacson makes Elon Musk look pretty good .?
2. If you’re familiar with the Gartner Hype Cycle, one might think we’re in or maybe entering the ‘trough of disillusionment.’
3. HeyGen is now enabling personalized deepfake generation as a service . This is going to cause far more problems than it will solve.
4. The Biden Administration is apparently restricting some chip sales to countries in the Middle East. Meanwhile, facing sanctions, China is aiming to achieve AI chip independence .
5. If the Biden Administration goes solo on trying to regulate AI, it’s unclear whether those efforts would survive a review under the Supreme Court’s newfound ‘major questions doctrine.’
6. It is worth repeating: AI is going to revolutionize healthcare. To get a better sense of how, check this: the FDA released a listing of all 500+ AI-powered tools that the agency has approved for health-related uses . And Stat+ has a larger list of players in the AI/health space . Relatedly, Sidley Austin recently published a state-of-play for regulations in the drug development process.
7. The Washington Post surveyed the global AI regulation landscape . And the IAPP built a tracker for global regulatory proposals .
8. Anguilla, which owns the .ai country code, is going to rake in revenue through the AI boom.?
9. Getting ahead of Senator Schumer’s AI summit next week, Senator Bill Cassidy released a white paper on the potential benefits and risks of AI .
10. Falcon 180B dropped and reshuffled the open model deck a bit.
11. Axios surveyed a group of AI experts about all sorts of topics , including who should regulate AI. Some experts are hoping for an international civil AI governance organization .
12. Wired Magazine did a deep dive into OpenAI , and how it’s changed as it has grown.
13. Just in time for school, OpenAI released a guide to teaching with ChatGPT .
14. I mentioned Lupe Fiasco’s work with Google last week. Now, MSNBC is digging deeper into the future intersection of hip-hop and AI .
15. It’s almost laughable, but some in the U.S. think that the EU’s approach to drafting and finalizing the AI Act is “too fast .”
16. Venture funding is down in aggregate, but AI startups account for 20% of funding in 2023 thus far . One stellar example: $110m for an AI chipmaker that aims to compete with Nvidia .
17. Even the big players are facing resource challenges, though, as Meta is evidently struggling to supply sufficient compute to its teams .
18. AI is helping map the world’s trees in an effort to help combat climate change.
19. Evidently, ChatGPT makes for a very unreliable financial advisor .
20. If WIRED is correct and it costs only $400 to build an AI-powered disinformation machine , you can imagine the potential societal implications.
21. Well, I would be surprised if the Tencent AI model didn’t outperform GPT models on tests involving the Chinese language…
22. DeepMind co-founder Mustafa Suleyman is working hard to evangelize the idea of human ‘containment’ of AI. More on the containment idea here .
23. Apple realized that Siri has to step up, so it’s spending lots of money quickly to catch up .
24. Widespread AI use costs tremendous amounts of energy, but could AI also improve our energy efficiency and infrastructure ?
25. Azeem Azhar digs into the complexity of AI governance .
26. Most major AI models are being developed by teams that speak primarily English or Chinese. What does this mean for users of those models who speak other languages ?
27. Good: Google will require disclaimers for AI use in political ads .
28. Switzerland is using AI to analyze land use within the country .
29. Fascinating to watch: ETSI, the European standards organization, is working on standards for image/audio/visual manipulation .
30. Get ready for model autophagy: it’s likely that there are already 15+ billion AI-generated images out there .
31. AI is particularly great for chat, coding, and (sleeper success) spreadsheets .
32. Another sign that OpenAI is maturing: they are now planning a developer conference . And bringing in a billion dollars .
33. The Center for AI and Digital Policy filed a complaint with the U.S. Federal Trade Commission regarding Zoom’s AI practices.
34. The U.S. Department of Energy is spending millions on AI .
35. If you’re looking for another tracker of state/federal proposals re: AI, Bryan Cave has you covered. One thing is clear, states are all over the place (and generally falling behind) on addressing the impact of AI on employment . Meanwhile, here is a good guide to all of the various proposals that the federal government is considering . If that’s concerning to you, Brownstein Hyatt Farber Schreck LLP put together a roadmap for employers considering the use of AI .
36. No-brainer: state AGs from all 50 states are urging Congress to take action to prevent AI from being used to abuse children .
37. Speaking of states, Governor Newsom of California issued an EO intended to start preparing California for generative AI’s impact on the economy .
38. Intuit is offering an AI assistant . No word from the IRS if ‘hallucinations’ are a good reason for tax return errors.
39. Gary Marcus hints at an interesting question: could AI put an end to ‘surveillance capitalism,’ or will it perpetuate it?
40. The EU AI Act regulates AI that uses “subliminal techniques” but what does that mean ?
41. Even if there is regulation regarding AI, it’s not clear that it will stop scraping by AI companies seeking training data .
42. Across large tech companies, red-teaming of AI deployments is having something of a moment.
43. Mason Hayes & Curran provides helpful advice for European companies seeking to use AI vendors.
44. Microsoft is now offering copyright assurances to commercial copilot customers.
45. The Office of Federal Contract Compliance Programs is now asking federal contractors about the AI systems they’re using for hiring and employment purposes.
46. Nathan Lambert dug into the AI research ecosystem and suggested areas for improvement.
47. People keep suing AI developers over the inclusion of [insert kind of information] in the developers’ training data sets.
48. Getting to parity with Teams, Zoom and Google are bringing bots to meetings to transcribe, take notes, and generate action items.
49. Evidently, employees are not keen on the idea of learning and development via AI .
50. AI leadership at large companies will probably continue to be a game of musical chairs .
51. Goldman Sachs is saying that AI isn’t a bubble . Same goes for executives during earnings calls .
52. As we have learned through the pandemic, the global supply chains are complicated and prone to problems. AI might help with all that .
53. Evidently, the German political parties are very split on what to do about AI .
54. As the MIT Technology Review highlights, it’s kind of wild how little we know about how LLMs actually work .
55. So, apparently, robots have been causing sporadic human deaths since the 1970s . Once AI is ubiquitous, I suspect that’ll accelerate.?
56. We keep discovering the same old problem: everyone is using generative AI at work but hardly anyone wants to admit it .
57. Ad agencies are using AI all over the place , it seems.