Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Prime_Evil" data-source="post: 9375471" data-attributes="member: 11984"><p>I suspect the truth lies somewhere in the middle. AI models are trained on material from the Internet, much of which is under copyright. However, it does not COPY these materials but rather draws statistical inferences from them. The models do not contain the original data. But...at the same time, they can be used to create works that echo the original source material. That's the way the modelling works. Are these derivative works in the copyright sense? That's a question for the legal system to sort out. There are some interesting precedents for this kind of argument. When online phone directories appeared, the companies that make Yellow Pages sued various online services for copyright violation. The courts (pretty much worldwide) argued that phone numbers were numbers. And that numbers and other mathematical properties could not be copyrighted. A more pertinent example might be the various legal actions against Google, arguing that the process of scanning the Internet to build its search index violated copyright. The most important case involved the scanning of books so Google could index the contents to help searchers discover relevant works. All the major publishers sued Google over this practice and lost. The argument was that this was a transformative use of the source material permitted under copyright law. Plus Google was not exposing the source material to users of its search engine - it was merely using them to build its search engines. Are these precedents relevant to AI? Who knows. I suspect there will be several rounds of court cases before we land on a clear position one way or another. In the meantime, most companies will continue to avoid the use of AI due to the current legal risks. Having said all of this, the concerns of authors and artists about this technology do have some validity too. Unfortunately the new technology is being pushed by tech giants who see it primarily as a means to reduce wages and cut jobs. Most are wedded to an obnoxious libertarian philosophy that's rapacious and hostile to government regulation. So authors and artists have good reasons to fear. And the law does need to offer them some protection against abuse. What form that intervention should take? I don't think we'll know for several years. Personally, I would argue in favour of a system similar to that introduced for older recording media. A few cents from every sale of the recording media is paid into a central fund used to reimburse artists for potential copyright violations involving the media. Such a scheme could work for AI, especially if the reimbursement was based upon how many of the artist's works were used to train the model. But there is still a problem - with older technologies the money rarely reached the artists themselves because the music labels inserted themselves as intermediaries, snapping up most of the money earmarked for artists. A new scheme would need careful design to ensure the same doesn't happen here too.</p></blockquote><p></p>

Verification