Shulou(Shulou.com)11/24 Report--

OpenAI is mired in class action + individual writer lawsuits, what is the solution to the copyright problem?

Recently, OpenAI has been in trouble again.

Before that, Sam Altman fought senators on Capitol Hill, helping OpenAI take care of regulators.

But as the saying goes, Rip is easy to see and imps are difficult to deal with.

Unexpectedly, OpenAI has recently been targeted by American law firms and writers.

Recently, they have jumped out to sue OpenAI, asking for a large amount of compensation.

The reason for the lawsuit is that the training data is infringed.

The law firm filed a class action lawsuit. On June 28, Clarkson, a law firm based in Northern California, filed a class action against OpenAI.

He accused OpenAI of grabbing data from the Internet while training a large model, seriously violating the copyright and privacy of 16 plaintiffs, and seeking compensation of US $3 billion.

The core claim of the Clarkson lawsuit is that OpenAI's entire business model is based on theft.

Ryan Clarkson, a partner in the law firm, said:

Law firms claim their legal rights on behalf of ordinary people whose information has been stolen and used to create the real existence of this powerful technology. "

The lawsuit specifically alleges that the OpenAI:

Use private information, including personally identifiable information, stolen from hundreds of millions of Internet users, including children of all ages, to create their products without informed consent or knowledge. "

The lawyer-led class action system must have a little understanding of the class action system in the United States in order to understand the context of this matter.

To put it simply, in most lawsuits, the number of plaintiffs is limited and clear, while class action is brought by a large group of people.

The class action system is because there are too many victims of an event, and often the damage suffered by a single victim is too small to be suitable for individual prosecution, so the plaintiff can form a collective to sue the defendant collectively.

More typical is the case of environmental protection, such as the discharge of pollution by a company affects the surrounding residents, and the residents collectively sue the company.

Or the system of a large company infringes upon the rights and interests of many employees.

But a big difference between a class action and an ordinary action is that the initiator is often not the plaintiff, but the attorney.

Because in class action, the attorney will ask for a high agency fee, which may exceed 50% of the compensation, and some exaggerated cases may even reach 80%.

As for the plaintiff, most of the time, the lawyer who initiated the lawsuit went to the house on his own initiative and signed an agreement to be represented by the lawyer, and if he won, he could get some compensation.

If the case is lost, in most cases there will be no need for legal fees.

After obtaining the authorization of a large number of plaintiffs, the lawyer filed a lawsuit in the court in the name of "plaintiff collective" to demand compensation from the defendant.

Because the number of plaintiffs is relatively large, if the case is won, the overall amount of compensation will often be more.

Because lawyers play a leading role in such cases, they will have a high proportion of legal fees and tend to gain more economic benefits than the victims.

Not long ago, for example, Apple reached a settlement with employees in a class action lawsuit because of the system of bag search and inspection of employees after work, paying employees $30.5 million, with an average compensation of $1328 per employee.

But they have to pay lawyers about $9 million in legal fees.

Lawyers or law firms themselves will gain a higher reputation in the industry because of the extensive influence of class action cases.

To some extent, the real "plaintiffs" in many American class action cases are lawyers who can earn more money, rather than the real "plaintiffs" who suffer.

The class action against OpenAI, represented by law firm Clarkson LLP, is a law firm founded in 2014 and almost exclusively engaged in class action.

Compared with many law firms in the United States with a history of several decades or more, Clarkson is a baby law firm.

But in recent years, it has won a number of class action lawsuits against food and drug manufacturers.

The lawsuit against OpenAI was also handled by the firm's named partner Ryan Clarkson himself.

Now their law firm's website has put its own information about suing OpenAI at the top.

Visitors to the website can see this message anytime, anywhere.

It can be seen that this time they want to "All in OpenAI", firing the "first shot in the big model infringement case".

There is no way to follow, and it is hard to predict the outcome that the legislation on large model training data in the United States has not yet been passed.

Before the congressional legislative process gives a clear provision, the judge is likely to refer to the existing privacy and copyright cases and make a finding in favor of the plaintiff in this case.

In the absence of clear legal rules, the case is likely to develop in a direction that is not conducive to OpenAI.

As long as the firm is not vexatious, and OpenAI is eager to get rid of similar litigation.

The case is likely to end in a settlement, like most class action cases.

The plaintiff and the lawyer are likely to receive a large settlement.

The plaintiff's law firm will also continue to consolidate its position in the industry by representing this potentially landmark case.

In the future, OpenAI and other large model providers may face more similar lawsuits.

Until there is a relevant law, clear the specific rules of training data.

There are more lawsuits and in addition to the law firm class action mentioned above, there are also many cases in which individuals sue OpenAI directly for infringement.

Such cases are mainly brought by writers who enjoy copyright in books.

Recently, two award-winning authors sued OpenAI, accusing them of violating copyright law by using books they published to train ChatGPT without consent.

The lawsuit was filed at the end of June, and the two writers said that ChatGPT's underlying LLM was directly used by the plaintiff's authors Mona Awad and Paul Tremblay's copyrighted works.

They believe that ChatGPT can generate detailed summaries of his work, which means that their books must have been included in the dataset for training.

Daniel Gervais, a law professor at an American university, told Insider that the writer's lawsuit is one of the few copyright cases against generative AI in the country.

"this will never be the last time. "

Gervais predicts that as these AI's ability to replicate the styles of writers and artists continues to develop and improve, more authors will join in suing developers LLM and AI.

He believes that nationwide, a wave of legal challenges against generative AI tools such as ChatGPT is coming.

It may be challenging to prove that the author of the case suffered financial losses as a result of OpenAI's data collection, the complaint said.

Gervais told Insider,ChatGPT that he may have collected works by Awad and Tremblay from sources other than the author.

AI of Sussex University and copyright expert Andres Guadamuz also expressed this concern.

He told Insider that even if the books were in OpenAI's training dataset, OpenAI could have obtained them legally.

The Writers Association of America, an advocacy group that supports writers' rights to work, published an open letter last week calling on CEO of large technology and AI companies to obtain "writer's permission" and use their copyrighted works to train LLM.

At the same time, writers should be compensated fairly.

So far, the open letter has received more than 2000 signatures.

Currently, Awad and Tremblay are seeking damages.

The lawsuit also includes summaries of Awad's novel "13 ways to look at Fat Girls" (13 Ways of Looking at a Fat Girl) and "Bunny" by ChatGPT, as well as Tremblay's "House at the end of the World" (The Cabin at the End of the World).

OpenAI, Awad and Tremblay did not respond to Insider's request for comment.

Similarly, comedian and writer Sarah Silverman, and writers Christopher Golden and Richard Kadrey also appealed in U.S. District Court, suing OpenAI and Meta, respectively, for copyright infringement.

The lawsuits allege that both OpenAI's ChatGPT and Meta's LLaMA are trained with illegally acquired datasets that contain their work.

They said the data sets were obtained from websites such as Bibliotik, Library Genesis and Z-Library, and pointed out that the books were available in bulk.

In the lawsuit against OpenAI, the trio provided evidence that when the relevant prompt appeared, ChatGPT would summarize their books, which they believed was an infringement of the copyright of their works.

Silverman's "Bedwetter" is the first book displayed by ChatGPT in the exhibit, while Golden's "Ararat" and Kadrey's "Sandman Slim" are also used as examples.

As for the separate lawsuit against Meta, the situation is similar. The dataset used by Meta to train the LLaMA model can also access the author's book.

In this complaint, it explains step by step why the plaintiffs believe that the sources of these data sets are illegal-in a paper on LLaMA, Meta points to the source of its training data sets, one of which is called "ThePile".

The complaint points out that an EleutherAI paper describes "ThePile" as a patchwork of copies of Bibliotik content, while Bibliotik itself is illegal.

In both statements, the participating authors said they did not agree with AI to use its copyrighted books as training content for LLM.

Their lawsuits include six different types of copyright infringement, negligence, unjust enrichment and unfair competition charges.

At present, these authors are also seeking legal compensation for losses, return of profits, and so on.

In addition to the book circle, Getty Images filed a lawsuit accusing Stability AI, who created Stable Diffusion, of training its models on "millions of copyrighted images".

No response was made by all parties concerned.

Reference:

Https://www.washingtonpost.com/technology/2023/06/28/openai-chatgpt-lawsuit-class-action/

Https://www.businessinsider.com/openai-copyright-lawsuit-authors-chatgpt-trained-on-books-2023-7

Https://www.theverge.com/2023/7/9/23788741/sarah-silverman-openai-meta-chatgpt-llama-copyright-infringement-chatbots-artificial-intelligence-a

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.