MBW Reacts is a series of analytical (and sometimes opinionated) articles by Music Business Worldwide written in response to recent major entertainment events and news stories.
The sudden emergence of generative AI over the last few years, especially over the last six months with the introduction of the ChatGPT chatbot, has opened a Pandora’s box of issues lawmakers, businesses and creators are struggling to address.
Aside from the (certainly important) question of what AI means for human artists and skilled workers, a number of policy issues are also rising to the forefront. Of particular relevance to the music business is the issue of copyright.
Many prominent music companies have expressed concerns about AI training itself to generate music using copyrighted musical works. At the core of that concern is the fear that AI will take human-made works and use them to generate music that competes with human creators in the marketplace.
Those concerns are perhaps best articulated by Universal Music Group (UMG) chairman and CEO Lucien Grainge, who spoke with analysts earlier this year on the company’s first-quarter earnings call. revealed the problem.
“The recent explosion of generative AI, if left unchecked, will increase the flood of undesirable content hosted on platforms, and violate existing copyright laws in the United States and other countries, as well as trademarks, names, and It raises rights issues with respect to laws governing rights: likeness, voice impersonation, publicity rights,” Grange said.
Grange continued: [i.e. earlier generations of AI]Many of the latest generative AIs are trained on copyrighted material, which clearly violates the rights of artists and labels, and the platform will not allow us, our artists’ partnerships, and their success. You will be completely at odds with the partner who promotes it. ”
While AI appropriation of artist names and voices (such as the “Fake Drake” track that went viral earlier this spring) is a fairly obvious example of copyright infringement, Grange’s second allegation, namely copyright Training an AI with protected material is copyright infringement. It’s actually far from clear.
And recently, a Japanese government representative made a statement that could have huge implications for the development of AI and the future of music and other creative industries. In the view of the Japanese government, training an AI with copyrighted material does not violate copyright law.
At a public hearing in late April, Japan’s Minister of Education, Culture, Sports, Science and Technology Keiko Nagaoka revealed the government’s view that Japan’s copyright law does not prohibit the use of learning AI for copyrighted works.
This is true even if the material is expressly for commercial purposes, so long as the material is not reproduced. AI can also be trained using material illegally hosted online, the minister revealed.
This is as broad and liberal a policy for AI technology as any country has expressed. And arguably, this poses a challenge for many in the creative industry who have argued that AI should not have free access to copyrighted material.
controversial stance
Over the past few days, rumors about the Japanese government’s position have provoked violent reactions from both supporters and opponents of the government.
“Japan has become a machine learning paradise” Yan LeCun declaredChief AI Scientist at Meta Platforms, owner of Facebook.
In response, one Twitter user sarcastically responded, “You can steal intellectual property without any repercussions = paradise.”
LeCun defended his position, arguing that the ‘driving principle’ behind intellectual property law is ‘maximizing the public interest, not the power of content owners’. bottom.
(It should be noted that LeCun is not an AI anarchist of any sort. He was one of the signatories of a recent letter warning that AI could endanger humanity if not properly controlled. was.)
LeCun’s position may be one many in the AI community would agree with, but it is likely to face serious opposition from rights holders, especially in businesses such as music, which are flooded with AI-generated content. It is also a position.
EU and China go their separate ways
However, at least for now, it is clear that Japan has diverged from other major jurisdictions in its approach to AI regulation, which many observers attribute to Japan’s desire to be a leader in the field. there is
In its draft regulation on AI development, which was revised earlier this year, the European Union said AI developers would be allowed to use legally accessible copyrighted material, but not what copyright. It came up with a kind of compromise that requires public declaration of protected content. Used for AI technology training.
The EU also prohibits the use of data mining on its IP, even if it is publicly available, so long as rights holders use appropriate mechanisms to make it clear that it is prohibited. We are proposing a carve-out that can be done. Include such clauses in your Terms of Service.
China appears to have gone a step further and declared that AI developers cannot automatically use protected IP for training AI models. Since Chinese copyright does not have an exemption for data mining, this means that Chinese AI developers must obtain explicit permission from the rights holders to use copyrighted material for AI training. means
Due to the vast amount of material used to train generative AI, both EU and Chinese regulations can pose serious challenges for AI developers. All of these AI models are pretty much the same, sucking huge amounts of data and finding patterns to produce the output that is most likely what you are looking for.
(Some scholars, such as the famous linguist Noam Chomsky, argue that siphoning vast amounts of information to spit out the most probable patterns is simply not how the human mind works, so this form of AI is doomed to failure, but that’s a story for another time.)
For example, consider Baidu’s ERNIE bot. According to Baidu, the training uses a “knowledge graph consisting of trillions of web page data, tens of billions of search and image data, hundreds of billions of daily voice data, and 55 trillion facts and more.” It is explained that Scrutinizing all this data to determine exactly what is copyrighted and what is not is a daunting task.
The U.S. seems to lag behind other regions in developing legislation to regulate AI, but steps are being taken. The U.S. Copyright Office has launched an effort to investigate the copyright impact of AI technology, which will run through the first half of 2023.
Soaring lawsuits?
Many of these issues could potentially be resolved through litigation, and the US is not far behind. A number of lawsuits have recently been filed over the use of copyrighted material in AI training.
one of the most famous, Andersen vs. Stability AI Ltd.which is pending in federal court in San Francisco, involves a group of artists who have sued AI labs Stability AI and Midjourney, along with online art platform DeviantArt, for alleged mass copyright infringement in the development of generative AI models. .
The companies filed a class action suit proposed by the plaintiffs, alleging that their AI-generated work did not resemble the work of the suing artist and that the lawsuit did not specify which artwork was infringed. asked the court to dismiss the
Getty Images has filed another lawsuit, also involving Stability AI, against the company in federal court in Delaware. Getty, a news and stock photo provider, claims its Stability AI technology scraped millions of images from websites and used them in the generation process. The lawsuit alleges that the Getty logo appears in a distorted form on some of Stability AI’s creations. A similar case involving Getty and Stability is ongoing in English courts.
The outcome of these lawsuits is uncertain, as much depends on how courts interpret existing copyright law in this new context.
Rights granted to copyright holders are generally limited to the following:
- Right to make copies
- The right to create derivative works based on the original work
- Right to distribute the work
- The right to display or perform the work in public.
It is the first two of these that AI developers can be vulnerable to. If training on copyrighted material means copying that material to a server where the AI learns, it could be construed as an unauthorized copy of the work. And courts may rule that the work produced by generative AI is “derivative” of the original material it learned.
But AI developers also have solid defenses to work with. The concept of “fair use” is an exception to copyright law in some countries that permits the use of copyrighted material for the purpose of certain activities that are considered open to the public. Interests such as news reporting, education and research.
In general, students are permitted to reproduce copyrighted material to some extent for coursework. If so, why shouldn’t AI be allowed to do the same when learning?
All of this means that the right balance between the outcome of these lawsuits and AI regulation is not fully clear today. Therefore, it is necessary to pay attention to the precedent set by Japan.
In the world of digital technology, 1 is a magic number. If there are no entities participating in a particular activity, that activity will not occur. However, if one entity is engaged in an activity, that activity will occur even if there are no other entities doing the same.
In the future, if the rules on AI development are looser than those of other countries, Japan is likely to become a “wild west” for AI development, where AI companies band together to make things that are not allowed in other jurisdictions. will do. .
Nonetheless, even in Japan, some are beginning to voice concerns about the government’s indifference to AI copyright issues. One of the key speakers is Takashi Kii, a Diet member representing Fukuoka’s 10 wards who is calling for new copyright rules to address the AI revolution.
“There is a problem from the viewpoint of rights protection” [copyrighted work in training AI] Even if it goes against the intentions of copyright holders, and … new regulations are needed to protect copyright holders,” Takashi wrote on his blog.
Moreover, advocates for stricter rules say Japan’s lack of regulation around AI could mean something other than an AI boom in Japan, and rather a spate of copyright infringement lawsuits. warned that it could lead toglobal music business
