Do You Know the Risks of Generative AI for Your Business?
Brian O'Donnell ??
Partner & Chief Strategy Officer @ Massive Insights | #DataforGood | Enabling Better Outcomes Through Better Decisions
This post is quite timely with the recent news that Italy is one of the first jurisdictions to go to battle with OpenAI over legal and privacy concerns. It might be the first, but it's certainly not the last.
To help shed light on the broader conversation about the potential risks associated with generative AI, I'm pleased to feature a Q&A session with author Karen Kilroy.
Her 2021 book, "AI and the Law, Developing and Deploying AI in the Enterprise" is a great starting point to begin to understand what's at stake from a business perspective.
Let's dive in:
Q: What are the legal considerations businesses should be aware of when using Generative AI??
With generative AI, I would think the main consideration is where did the content come from that is used to create the derivative works? Was the content obtained legally and licensed or purchased from the source, or is it relying on being fair game because it is freely available data?
What will happen when freely available training data runs out? Will AI models then resort to less-than-ethical techniques for obtaining additional training data? Since premium training data will provide a competitive edge for generative AI models, this could be a big point of consideration.
Another consideration is whether or not the derivative work is able to be copyrighted. Right now, the U.S. Copyright Office is saying "no" to copyrights on AI-generated works. Will this decision hold up once most people are using generative AI for entertainment?
Instead of listening to a static story, song, or artwork, people are already enjoying compositions that they create specifically for themselves, making the consumer the center of the creative work. It could be that we will find that the derivative work might not be copyrightable, but the prompt used to generate the work is, much like other computer code.
For example, let's say you prompt DALL-E to paint "social media as seen by Sigmund Freud in the style of Jean-Michael Basquait." You likely couldn't copyright work that was derived from ingesting multitudes of examples of Basquait's painting style as training data, but someone could perhaps copyright the prompt used to generate the derivative work. I have not heard of this coming up yet, although it is becoming an established practice to catalog and curate prompts.
Q: How can businesses ensure that AI-generated assets do?not infringe on any intellectual property rights or raise ethical considerations?
Know where the model's training data came from. AI is a lot like a cupcake from the convenience store; you aren't quite sure what is in there, or from where it originated. However, just like you can ask for a better cupcake, you can ask for better AI models.
AI can easily be built to be trackable and traceable, in the same manner that the ingredients in a food product can be tracked and traced. I touch on this concept in AI and the Law, and it is fully explained in my book, Blockchain Tethered AI (O'Reilly, 2023).
Q: How can businesses protect themselves from any potential legal consequences when using Generative AI??
This field is changing so rapidly, I'd suggest businesses engage legal counsel that is fast-moving and willing to constantly learn. Slow-moving firms will likely be left behind.
领英推荐
Q: Are there any licensing or legal agreements they should consider?
I suggest that attorneys can assume there is no way to enforce any kind of licensing or legal agreements on behalf of content creators who have their content devoured by AI models without granting permission.
This has been proven with the lack of payment to any kind of content creators in art, music, writing, and other creative content that is shared on the world wide web without permission or payment.
Another consideration is that content creators might be fooled by fine print on new tools or storage systems, and they may be tricked into granting permission for their content to be used by the company providing the tool, with no compensation whatsoever to the creator.
Given these concerns, the best thing to do would be to collect payment for the content creators in advance of their content being used as AI training data. I am working with the Fayetteville, Arkansas Public Library Center for Innovation on a plan for libraries to curate content for creators, and collect payment in advance for the use of the training data and pay the content creators.
Under this plan, once the creators pass away, the library would continue to curate and benefit from the content. Libraries are long-lasting institutions and are very good at collecting money. With this plan, creators get to enjoy being paid for their work while they are alive, and the AI organizations get to enjoy ethical operation, knowing they are delivering a low-risk product to their consumers.
Disclaimer: None of the above should be construed as legal advice and it is always recommended to consult your own counsel for legal advice specific to your own circumstances.
See you tomorrow
for day 16/30 ??
______
See all "30 Days of AI -- Beyond the Hype" posts here:
(updated daily from March 2023 through April 2023)
Special thank you to Karen Kilroy for taking the time to answer these timely and important questions. You can see her impressive LinkedIn profile here. Make sure to give her a follow.