Hacker News Reader

Comment by mongol | Hacker News Reader

So we have a copyrighted work. A program reads it and somehow processes it, and does its AI thing . But in the very first step, the work is copied verbatim into the program, right? Even if it is not reproduced, is it allowed to process a copyrighted work for arbitrary purposes, as long as it is not further reproduced? Let's say I could mine coins from hashes of a copyrighted text. Would that infringe on the copyright? I don't know what copyright _really_ means, I realize

View on HN

But in the very first step, the work is copied verbatim into the program, right? Even if it is not reproduced, is it allowed to process a copyrighted work for arbitrary purposes, as long as it is not further reproduced?

In some cases yes, otherwise search engines couldn't exist.

Yes search engines are a good analog. Are search engines allowed to index and make copyrighted works searchable, _without_ permission from the cooyright holder? I seem to recall that Google did not get it their way with Google Books?

Google won the Google Books case. And they were actively scanning books rather than just indexing the internet like OpenAI is.

They were scanning books they had legally-obtained copies of.

OpenAI is very often downloading whole books which they do not have licenses for. Publicly-accessible does not equal legal to download, in US copyright law.

Case law in the US is that a computer copying something into RAM is a "copy" under the copyright act, and can be copyright infringement (MAI v. Peak Computer), the issue is whether making that copy is fair use, or otherwise authorized. Joseph P. Liu's "Owning Digital Copies: Copyright Law and the Incidents of Copy Ownership" is a good overview of the issues. I don't believe the law has substantial changed since he wrote the article in 2001.

https://en.wikipedia.org/wiki/MAI_Systems_Corp._v._Peak_Comp....

https://scholarship.law.wm.edu/wmlr/vol42/iss4/5/

Any time you make a copy of an expressive work without express permission from the author (in the form of a license), you are committing copyright infringement.

Fair Use is a defense to copyright infringement. Under certain circumstances you are allowed to commit copyright infringement. For instance you can copy a short scene from a movie if you are making a review/commentary video.

OpenAPI is arguing that training AI models should be a "fair use". They are not claiming they never used copyrighted materials.

Reply on news.ycombinator.com