Artificial intelligence can make people's lives more efficient, More entertaining and more productive, But there are potential downsides.

From a copyright point of view, Artificial intelligence raises some interesting questions. Such as, Whether content created by artificial intelligence can be copyrighted? Whether AI can be trained on copyrighted works without restriction?

Author action OpenAI

Some authors argue that, Large language model training sets should not be allowed to use every paragraph of text they encounter online. In this year 6 In a lawsuit filed in October, Book author Paul．Tremblay (Paul Tremblay) And Mona．Awad (Mona Awad) Against American artificial intelligence companies OpenAI Direct and indirect copyright infringement.

Soon after, WRITER／Sarah the comedian．Silverman (Sarah Silverman) With writer Christopher．Gordon (Christopher Golden) And Richard．Cadre (Richard Kadrey) File a lawsuit together, allege OpenAI Use books as training data. allege, OpenAI Unauthorized use of data sets from pirate websites.

The complaint mentions the disputed Books2 and Books3 Data set, The data sets are believed to come from LibGen, Z-Library, Sci-Hub and Bibliotik Equivalent shadow library.

These sites collect books that are also available through BT System batch acquisition. The authors write: "For a long time, The AI training community has long been interested in these blatantly illegal shadow libraries... "

OpenAI Ask the court to dismiss the charges

recently, OpenAI Responded to the allegations, To dismiss most of the claims. The allegations include indirect copyright infringement, violate "Digital Millennium Copyright Act" , Unfair competition, Negligence and unjust enrichment.

OpenAI Address the court: "None of these causes of action raises a viable claim for relief, Because none of the legal theories in question here actually condemn ChatGPT, support ChatGPT Or the processes used to create these models. "

"It is very important to remove these claims from the lawsuit from the beginning, In this way, these cases will not enter the discovery stage and beyond when the legal liability theory is not established. "

Batch rejection

At present, The only charge that can win is direct copyright infringement, but OpenAI It is hoped that this claim will be dismissed at a later stage.

Fair use

The authors' claim of copyright infringement is based on copyright law. OpenAI It does not deny the role of copyright, But points out that the petitions take a hard line, Ignoring exemptions such as fair use.

OpenAI According to: "however, These claims misunderstand the scope of copyright, Limitations and exceptions are not taken into account (Including fair use) , These limitations and exceptions allow room for innovation, For example, large-scale language models are now at the forefront of artificial intelligence. "

OpenAI Point out, Be drafting "United States Constitution" when, Its creators see copyright law as a tool to promote the advancement of science and the applied arts. In this case, Artificial intelligence is seen as a useful advance, Its use of a large number of copyrighted texts can be seen as "rational" .

OpenAI write: "Many courts have applied the fair use doctrine to achieve this balance, It recognizes that innovators' use of copyrighted materials in transformative ways does not violate copyright law. "

derivative?

The author clearly sees things differently. They think, OpenAI Every output of a language model is a derivative work of copyright infringement. These derivatives were created without the permission of the rights holder.

OpenAI I think this conclusion goes too far. The organization noted, According to the author's theory, All outputs of large language models are inherently copyright infringing. Though that may be the conclusion the authors were hoping for, But it would seriously hinder AI innovation.

The artificial intelligence company noted, Courts have previously rejected overly broad interpretations of the term derivative, So it should be in this case.

According to the indictment, ChatGPT For every output——From simple answers to questions to the name of the President of the United States, To describing Homer's epic "Iliad" plot, Topic and meaning of the paragraph——Must be infringement of the plaintiff's book "Derivative work" .

OpenAI add: "What's worse, Each of these outputs is an infringing derivative of millions of other individual works contained in the training corpus——Regardless of whether there are any similarities between the output and the training work. That's not how copyright law works. "

Based on the above reasons and various arguments, OpenAI Ask the court to dismiss all claims except for direct copyright infringement. (Be compiled from torrentfreak. com)

TRANSLATORS: Rason group　proofread: Wang Dan



　　disclaimer: This network reprint or compile the original articles are from the network, Does not represent the views of this website or confirm the authenticity of its content. If the source is mislabeled or the copyright of the article is involved, Please contact us, This website will be corrected in due course, delete, thank you.