Many AIs train on enormous datasets of content. They draw on huge troves of images for art apps and oceans of text for chatbots. These datasets can sprawl across the internet, from Wikipedia and discussion threads to personal websites, paywalled materials, artist pages, and more.
I really enjoyed the way you wrote this out with lots of different possibilities. I personally think that AI will be viewed as needed service. If it can be freely accessed on the Internet, then it will be viewed as something that can be incorporated into the AI. The argument will be made that if a person or institution doesn't want it scraped then they shouldn't make it freely available (paywall required). I believe the courts won't view it as a direct derivative work and will instead view it as a general evolution and allowed use.
Re: the courts, not being a lawyer means I have to struggle with reading actual lawyers and their analyses. I'm not getting a good feel for a single path forward yet.
Re: not making content available - I wonder how widespread such decisions will go. Will an AI-scraping setting become something we expect from web pages, podcasts, Instagram photos, etc?
I’ll be curious how the copyright cases turn out. Everything you, Bryan, have ever written online is a vanishingly small subset of a typical LLM training set. And you’re fairly prolific with your online writing! The transformative use part of the fair use doctrine is going to be strong with the AI defendants, I think. With the caveat that our current SCOTUS does very surprising things.
Copyright could shut it all down. Yes, this is transformative - but without permission or compensation. Claiming fair use won't work, as these are clearly for-profits, unless they try to claim the business model doesn't work.
I really enjoyed the way you wrote this out with lots of different possibilities. I personally think that AI will be viewed as needed service. If it can be freely accessed on the Internet, then it will be viewed as something that can be incorporated into the AI. The argument will be made that if a person or institution doesn't want it scraped then they shouldn't make it freely available (paywall required). I believe the courts won't view it as a direct derivative work and will instead view it as a general evolution and allowed use.
Thank you for the thoughtful comments, Brent.
Re: the courts, not being a lawyer means I have to struggle with reading actual lawyers and their analyses. I'm not getting a good feel for a single path forward yet.
Re: not making content available - I wonder how widespread such decisions will go. Will an AI-scraping setting become something we expect from web pages, podcasts, Instagram photos, etc?
I’ll be curious how the copyright cases turn out. Everything you, Bryan, have ever written online is a vanishingly small subset of a typical LLM training set. And you’re fairly prolific with your online writing! The transformative use part of the fair use doctrine is going to be strong with the AI defendants, I think. With the caveat that our current SCOTUS does very surprising things.
Copyright could shut it all down. Yes, this is transformative - but without permission or compensation. Claiming fair use won't work, as these are clearly for-profits, unless they try to claim the business model doesn't work.
The largest source of content "sourced" for AI has a lot to say! https://medium.com/freely-sharing-the-sum-of-all-knowledge/wikipedias-value-in-the-age-of-generative-ai-b19fec06bbee
I like her recommendations.
Wisdom In - Wisdom Out. Priceless.