OSPOlogy Follow Up: AI Compliance Programs & OSPOs

[anajsana]

Discuss with the community topics related to the upcoming OSPOlogy meeting:

📅 September 6th ➡️ AI Compliance Programs & OSPOs

• 🕐 Today at 15:00pm UTC | 17:00pm CEST | 8:00am PT
• 👩‍💻 Videoconference Link
• 💚Registration page (not mandatory)
• 🙋‍♀️ Panelists: Joanna, Van, Ashley, Georg, Christine
• 📝 OSPONews Presentation and Topic of the month presentation

In this open space, people can:

• Ask questions prior to the meeting starts
• Share notes from the conversation
• Add useful links and resources related to the topic
• Connect with the panelists

[ashleywolf]

Please feel free to post any questions or comments for us to address during the call!

[anajsana]

from webinar platform chat: "Is there a link available for more information around that RFC?"

[anajsana]

https://www.federalregister.gov/documents/2023/08/30/2023-18624/artificial-intelligence-and-copyright

[gioalmeida-ml]

is there any reference training for developers about a compliant use of generative AIs?

[VanL]

I don't think that anyone has a good grasp on what "compliant use" means yet, except to say that it is use that is within the guidelines that your own organization has created.

[anajsana]

from webinar platform chat: "The ZF OSPO formed an alliance with the legal/IP team for generative AI topics. We collected developer, cyber security and business development feedbacks and adressed them in a positioning paper, which was published globally"

[anajsana]

from webinar platform chat: "In our organization the OSPO reports up through a Tech Innovation function, so there’s a close integration between OSPO and the GenAI work that the innovation team is leading. In fact, the OSPO is bringing open source projects, models, applications to those investigations"

[anajsana]

from webinar platform chat "We are especially interested in the challenges of traceability and embedding review guidance for generated items. It should not end up in a blind "copy/paste" process"

[anajsana]

"it relates also to commenting guidelines for our developers/labelling generated code"

[anajsana]

answer via webinar platform chat: "One thing we are recommending is that any files that include AI-generated content be given just a comment:"

[anajsana]

from webinar platform: "What are your experiences @ Van in regards to patent inventor payments, if developers use generated code/docs to file a patent? Is there any information available?"

[anajsana]

answer: "The most recent law is that the GAI cannot create the/a point of novelty for a patent, the inventor must be human. That said, there are no restrictions on the use of AI as a part of an invention or even as part of the write-up or spec - you just need to make sure that everything is reviewed because a human needs to affirm to the truth of everything filed."

[anajsana]

from the webinar platform chat: "Having worked on SBOM and licensing provenance for years the traceability is a big issue. Forking or copying code is common in open source projects so even if an AI maintains provenance on where it got something it’s maybe wrong. Basically same issue as snippet detection"

[anajsana]

from the webinar platform chat: "I think Llama has had the same non-OSI issues - that it has restrictions on what you can use it for"

[VanL]

It does. LLaMA has a very liberal license, but it is not open source. More on that: https://ospo.co/blog/llama-2-and-open-source/

[anajsana]

from the webinar platform chat: "How are you dealing with dataset use? Do you have policies around that?"

[anajsana]

from webinar platform: For anyone interested in a deep dive into some of the legal issues, I recently had this published in the Rutgers Business Law Review: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4464001