• White paper/blog posts
  • Wenjing Chu : probably can complete one in September "Digital Trust in the Age of Generative AI". Just need time to write it up.
  • Previously Sandy Aggarwal - have not heard back for a while, in "hibernation" until hearing back from him. Will remove it from tracking now.
  • Previously Phil Wolff - Phil said he had to postpone it, also in "hibernation". Will remove it from tracking now.
  • Anything else?
• Blog post ideas floated in meetings and chats in the past - Let's review and make a decision
  
  • "ChatGPT use policy recommendations - Recommendation - by AIM TF" - Eric Drury Wenjing Chu 
  • "Intersection of AI and Metaverse - automated characters. " blog post - Wenjing Chu 
  • "EU AI Act - response from ToIP AIM Task Force - blog post." - Zaida Rivai Drummond Reed sankarshan Chi Hwa Tang Wenjing Chu 
  • "The need for an AI agent for consent" - Neil Thomson
  • Any other proposals? Please express your interests in helping writing any one of them. 

All

• Topic #1: Copyright and AI
  • The US Copyright Office policy statement
  • The US Copyright office Notice of Inquiry (August 30, 2023) (Also: https://www.copyright.gov/ai/)
  • The Creative Commons paper
  • OpenAI responds to law suits
• Steven Milstein suggested that AI algorithms provide the sources used to train it. If it is Open Source and it is Creative Commons, and everyone is following rules, and you can prove the provenance of how you concluded, it should be okay. 
• Wenjing Chu - Copyright be issued to a human. Question - How much human involvement is required to ensure it can be copyrighted? Copyright guideline is that only prompts are not enough. It is still ambiguous. Open AI could buy a published book and feed it into the training software - in the end, it will contribute to some set of parameters. To call it a copy is not clear. Ideas expressed in a book are not protected. When Chat GPT gives a result to a prompt, who owns the copyright - Chat GPT or the human who wrote the prompt? Human involvement meets the criteria for copyright. Prompt engineering is a creative process - open-ended. LLM is like psychology. It is a creative process involving two languages. Chat GPT uses your data to train their models. Does Chat GPT own part of your creation that is patented?
• Neil Thomson - It depends on the underlying data. There are data brokers out there putting human effort into qualifying the data. Raising the quality of the data. The prompts should be able to identify the data sources to ensure the correct data is used. If there is no syntax, how can the AI engine interpret the prompt?
• sankarshan - Can we prove substantially that the training data that led to the output has provenance built in? Nobody is talking about the fact that prompts are created. There is a creative human element along with the machine.  Data (input) —> machine —> Data (output) : has provenance requirements for author, data, and training set. That’s a bit perhaps we can highlight in all this. https://www.blueshadow.art/midjourney-prompt-commands/ : example of Midjourney prompts and these are not even good ones
• Steven Milstein - The prompt can be the creative part. Think about the IP. Any new product is protected, but the ingredients that go into it may not be. If it shows the chain of how you got the ingredients, then the recipe is the creation. It is difficult. Why would this not apply to 3D printers? Don't have the rights to distribute the book but can have Open AI write a book summary, then use Open AI to use the summary in a training model. The courts say it is fair use. It is fair use if we cannot use the ideas in a book to expand it further and grow from it. How else will we grow? The 3D printer is similar to Chat GPT. There are no SQL-like statements for prompt engineering. Open AI is not using prompts data for training if it is used via an API. 
• Jacob Yunger - 3 legal questions: Parameters of models. Are the parameters based on copyrighted materials? What if we take the output of a model, change a couple of parameters, and use it to create something new? Can we copyright the output of the ChatGPT model? Making a commercial product of copyrighted material - is it fair?
• Anita Rao The laws are still ambiguous. It needs to evolve. This will take time to understand the creative process. Prompt engineering is a creative process. It needs deep domain knowledge and is an art to get the response from Gen AI. 

• Topic #2 (Probably next time) : Worldcoin reivew of this white paper.