As readers of this blog have doubtless noticed, the latest hot tech – and investment – area involves “agentic AI”, where AI systems are allowed to operative autonomously on allocated tasks. There’s no doubt there are some exciting possibilities here, as well as some troubling issues concerning lack of control. It’s a rapidly-evolving area of research and experimentation, which makes it hard to discern what is happening, and what is likely to arrive in the near future. One handy way to obtain at least some idea of the possibilities is to look at patent filings. These are often speculative, floating ideas that may or may not turn into real products and services. But they do at least hint at some of the things that we may see.
For example, an interesting post on Music Business Worldwide has spotted some filings by Music IP Holdings, a unit formed last year by Universal Music Group. They concern the application of AI to various aspects of music licensing and copyright enforcement. The main patent is for a “Copyright Licensing Chatbot” :
designed to interrogate potential [music] licensees about commercial vs. non-commercial use, timeframe, and geographical scope before either clearing or escalating the request.
The filings state that the chatbot “questions the potential licensee on a number of usage attributes to identify both the appropriate licensing models and to confirm that the proposed usage of the content is aligned with the artist’s principles and overall requirements that need to be met before their material shall be licensed and used in a derivative work.”
Another idea revealed by the filings is an automated system for calculating what the licensing fees should be:
the filings describe a dynamic pricing engine that adjusts licensing costs in real time “based on demand, seasonality and market conditions” using reinforcement learning techniques to continuously optimize pricing based on “conversion rates, customer lifetime value, and market penetration goals”.
As Music Business Worldwide (MBW) notes:
The filings suggest this would be paired with personalized subscription plans, bundle offerings, loyalty rewards, and tiered pricing for AI music licensees – a model that, to MBW’s eye, at least, looks a lot less like traditional music licensing and a lot more like the demand-based dynamic pricing already familiar from the live concert ticketing business, applied to copyright permissions.
More flexible music subscription plans could be a good thing for licensees and general users, since they would allow a far wider range of options that better reflect individual circumstances. However, another proposed application of AI in the copyright world that appears in the patent filings seems a terrible idea. It involves what is called an “LLM [Large Language Model] Agent Copyright Crawler”
The filings describe the crawler as sampling content streams from the open web, detecting digital watermarks embedded in images and audio using machine learning, and cross-referencing that material against IP licenses “currently in force”.
…
The filings state that the “LLM Agent” can connect directly with a source of materials and “propose licensing terms”; it can thank users for having “properly licensed usage”; and – critically – it can send “one or more cease and desist letters to user or streamer”.
In other words, the vision here is of a potentially unlimited number of AI agents hunting down alleged copyright infringement across the Internet, and automatically sending threatening letters to those supposedly involved. As Walled Culture the book (digital versions available free of charge) noted, the move from manual to automated takedown requests has already led to a huge increase in the volume of such demands – Google receives many billions of them every year. Allowing more capable and complex AI agents to carry out searches and send takedown notices will exacerbate the situation further.
For example, a move from the fairly crude identification systems employed today to ones involving AI could lead to far looser matching – so as to cast the copyright infringement detection net more widely – with inevitably more false positives. Perhaps even more troubling, autonomous AI agents could draw on the power of large language models to hunt out personal information of the people who are allegedly involved. This might include their phone numbers, email addresses and social media accounts, which would allow cease and desist notices – or less formal warnings – to be sent to them directly and repeatedly. More unscrupulous services might even seek associated information about alleged infringers – where they work, for example – and threaten to send information about their activities to their employers.
This is not the first time that Walled Culture has discussed the potential for AI agents to be used to hunt down alleged copyright infringement: a slightly different application of the technology was discussed back in February last year. These scenarios might seem far-fetched, but the pace of development in the world of AI in general, and agentic AI in particular, means that nothing can be ruled out at this stage. Moreover, there is little hope that the legal system can keep up with the constantly-evolving AI landscape, so legal remedies are likely to be few and ineffectual. The specific patent filings spotted by Music Business Worldwide may never be turned into real-world products and services. But given the financial incentives to do so, we can be sure that many others are working on similar ideas, and that some of them will indeed be realised. And when they do, they are almost certain to make the world of copyright even worse than it is today.
Featured image from Raw Pixel.
