Paperweights, 2nd Ed.
AI legislation, cybersecurity, and poetry
This is the second edition of Paperweights, offering links and commentary on major developments in AI governance.
This week on Machine Culture, Kevin Frazier — the AI Innovation and Law Fellow at the University of Texas School of Law — did a Model Convo, and I interviewed Matt Mittelsteadt and Lauren Wagner on the costs of settling for a nationwide AI regulatory patchwork. Here’s one excerpt:
Matt Mittelsteadt: No matter what our approach is, disharmonized regulation is going to be a huge problem moving forward if we have 50 different sets of laws… [Consumers and firms are] going to be spending unnecessary amounts of resources trying to comply with these things. That can actually have real costs if you look at other areas in which there already is a regulatory patchwork.
The risks of getting this wrong were certainly underscored later in the week (see below).
Congress is back in session, and that means the potential for AI-focused legislation — even if Polymarket (currently) only has an 8% chance the US will enact an AI safety bill by year’s end. I have some comments at Inside AI Policy detailing where the Hill might be headed in the next 2.5 months. Here’s an excerpt from my commentary:
I would expect more of the same — continued executive branch implementation of the White House’s AI Action Plan and other agency-based moves, possibly alongside narrower bills on watermarking, deepfakes, [and] adversarial AI bans within the US government. There’s also the GAIN AI Act… If the US can’t get comprehensive, pro-innovation, preemptive federal AI legislation done, my hope is that lawmakers can at least scuttle the costliest [bills] and focus on empowering agencies like CISA, NIST/CAISI, and others to execute the AI Action Plan and promote public-private cooperation to protect Americans from malign foreign influence.
And speaking of that malign influence…
Anthropic reported the first AI-orchestrated cyber espionage campaign, likely conducted by a Chinese state-sponsored group. Congress recently extended the Cybersecurity Information Sharing Act of 2015 — but only through the end of January. In the absence of more comprehensive AI legislation, this may be an area where we see some more traction on the Hill. The Institute for Progress also has a great proposal out on leveraging AI proactively to close gaps in our cyber defenses — Operation Patchlight.
An AI-generated country song is topping one of Billboard’s digital sales charts. Critics might say the song is bland, faceless, and superficial, but that strikes me as par for the course for much (certainly not all) pop music. Is the problem here the supply side or the demand side? For deeper commentary, I recommend my colleagues Rebecca Lowe and Henry Oliver.
AI tutoring has some merit, as evidenced by a randomized controlled experiment published in Scientific Reports. Carl Hendrick has more. Research in cultural evolution has much to say about maladaptive loss as well as the costs and benefits of specialization. And what will that process look like in the long run anyway? The authors of AI 2027 and “AI as Normal Technology” teamed up to outline their mutual prognostications. Also, according to Revana Sharfuddin, AI is not taking consulting jobs — for now.