For leaders looking to make the most of AI while also protecting themselves, these ground rules prove essential but are just a foundation from which to begin.
You can’t throw a stone without hitting a news article or op-ed on generative AI (GAI). But what’s really missing from the conversation is not the opportunities or challenges, but the solutions. As the CEO of a holding company with nearly $1 billion assets under management, including more than 75,000 copyrights, I have a vested interest in answers, not open-ended questions. GAI is here to stay; it’s the question of how we govern and remunerate that we must answer — with speed and fairness.
First, we establish some fundamental ground rules for what I believe will support the best outcome in this new world of GAI. These are based on my own professional experience and informed by the current regulatory environment.
1. Property rights should include copyright and personal digital replica rights.
2. Statutory licensing is necessary for training because tech companies have more leverage than rights owners.
3. Direct licensing is more ideal for outputs.
4. Standardized training should include an AI.txt machine-readable file on publicly accessible websites that can be used by rights-holders and online service providers to set forth training licensing terms or convey a message that the rights-holder is amenable to negotiating licensing terms.
5. Open-source blockchain-based solutions can be used to provide copyright management information to AI companies for training and licensing and from which to provide provenance information to enable downstream business models and compliance with existing and future laws and regulations, including Federal Trade Commission truth in advertising laws.
6. The Copyright Office must design standards for copyrightability in the context of prompt engineering.
7. Every GAI platform should create a standardized API between GAI companies and the Copyright Office to provide the necessary information to determine if prompt engineering gives rise to copyrightability.
8. The artist-centric approach to licensing digital service providers as pioneered by Universal Music Group in its deal with Deezer is a first step in a licensing model enabling GAI music and human-authored music to live side by side in the consumer experience.
9. Performers’ voices and likenesses must only be used in outputs with their consent and fair market compensation for specific uses.
10. Every human’s name, image, voice, likeness should be a property right, perhaps as suggested in a better-conceived version of the discussion draft of the NO FAKES Act.
11. Section 230 of the Communications Decency Act should be amended to clarify that state right of publicity and privacy-based appropriation claims are not immunized by section 230 unless and until there is new Federal pervasive legislation that preempts state law.
For leaders looking to make the most of AI while also protecting themselves, these ground rules prove essential but are just a foundation from which to begin. While regulatory and policy-making bodies are still getting their act together, it makes sense to craft rules for the use of AI internally at your company that model off the above boundaries. A general rule of thumb is not “what you can get away with” but “does what I’m doing seem fair and fairly compensate those whose work is used?” While AI is “all the rage,” the reality is that the vast majority of people are still inexperienced in using it, so extensive use case training around your internal policies will be essential for every employee before “clearing” them for use.
Once we establish these “outside boundaries” of the GAI arena and associated implementation programs, the compensation and use models begin to take clear shape. From here, the use of copyrighted works for GAI training invokes an exclusive right (except as may be covered by fair use, fair dealing, and text and data mining exceptions), and authorization may be required from rights holders in certain circumstances with regard to output.