Friday, December 1, 2023

I went to my congressperson’s office to talk about AI. Here’s what I learned.

After writing some comments for the US Copyright Office’s comment period on artificial intelligence, I decided this topic was important enough that I should talk to someone at my local congressperson’s office about it. After a bit of phone tag and filling out forms, I was able to get a meeting with one of the staffers during lunchtime earlier this week.

The office was located in downtown Madison, WI, not far from the state capitol building. I wanted to do a live demo, so this involved carrying my personal laptop and a full external GPU setup around for the day (which was hard on my legs, as you can imagine). I arrived at the office just before noon (after picking up lunch) and waited to be let into the office. After a short time, I was let in and was able to set up in their conference room.

As I was setting up and starting Stable Diffusion, I asked the two staffers in attendance what their background in artificial intelligence was. It seemed that neither was familiar with much beyond ChatGPT, so I gave them a chance to suggest images to be rendered, and showed them around the Automatic1111 interface, including the upscaling and img2img features.

The staffers had a number of technical questions about the technology, which was to be expected. I did have to caveat that even as someone with a computer science background, some of the aspects were difficult for me to understand as well. That aside, I explained how diffusion models work, with a focus on how data is represented in the model, and how an image model is more a loose understanding of concepts and connections between them, not a database of image data. I also did my best to explain the importance of GPU technology, cost of training/operating models, etc.

A few other things I tried to get across:

-Any regulation that gets enacted needs to have some level of public compliance to be viable, especially since the things individuals and small groups can do will continue to expand as compute power continues to grow.
-Nearly every application of AI has both legitimate and illicit uses, so it’s going to be difficult (if not impossible) to make clear lines in regulation.
-I don’t expect to get exactly what I want out of court rulings and legislation. That’s life.
-Video/audio enhancement AI probably has as much or more potential in the near term as generative AI, but isn’t getting as much attention because it isn’t controversial enough.
-Media outlets will start making use of generative AI in speculative content (what if they made a movie about X starring actor Y?), so it’s not just a “freedom of expression” issue, but also a “freedom of the press” issue.
-In addition to a wide array of first amendment issues, there may also be fifth amendment considerations-I see a good chance of a case arising from the use of an AI voice in court.

Here's what I learned/noticed from them:

-Although some legislation is being worked on (I mentioned I was aware of the “No Fakes” act), they didn’t seem to think much would get passed before cases got to trial or during this session of congress (i.e. through the end of next year), so if you want to contact your elected officials, now’s a good time.
-They seemed genuinely interested in learning more, and I might get asked again in the future.
-The older staffer took notes (handwritten!) on some of the things I said and showed.

Overall, I thought the meeting was productive and I was listened to with respect and consideration.  If you’re similarly interested/concerned about artificial intelligence, I’d suggest reaching out to your elected officials as well.

No comments: