Revisiting the AI Alignment (Control) Problem by creating a mind map
The AGI Control Problem was not included in my list of existential risks—such as post-modernism, nuclear war, and natural catastrophic events—until I discovered that people are willing to invest cash prizes for a theoretical approach to solving the shutdown problem in AI. The realization that there is a willingness to pay for solutions to this problem was eye-opening. Now, a year and two months later, I find myself cataloging my ideas and reflecting on the progress I've made during this period. In a related development, the Future of Life Institute issued an open call for proposed designs for global institutions to govern AI, with a specific focus on Artificial General Intelligence. My colleague, Doc Ligot, and I, representing Data Ethics Ph, decided to participate, motivated by the significant opportunity to contribute our thoughts through this mind mapping exercise. We believe we can offer a conceptual framework potentially helpful in addressing the control problem. TSame as in one of my recent post, I also used whimsical to build my RLLM visual map.
This mind map is currently incomplete. Further updates will be made to enhance its comprehensiveness. Despite this, this visual tool is strong enough to share the complexity of my thoughts around the problem.
Main Parts?
I added below some of my ideas/ thoughts on why I used these questions to organize my thinking:
"Why build a mechanism to govern or control an AGI or many AGIs?" It's crucial to understand why addressing the AGI control problem is significant.
"How am I helping to solve the alignment problem?" Here, I outline my contributions (or attempts) to solving one of the most pivotal problems in history.
"How is humanity attempting to solve the AGI control problem?" In this section, I share my personal perspective on the community addressing this issue, highlighting individuals who have insightful approaches to the problem and potential solutions. (Note: As of the time of this writing, this section is incomplete.)
"When do I think AGI will arrive?" I predict the AGI singularity will occur in three years. This question is included on this map to serve as a reminder of my personal estimate.
"Who will build AGI?" Understanding or conceptualizing which groups—such as AI labs, governments, the open-source community, or rogue entities—are likely to develop AGI is important.
Consider this map as the first version and this will definitely evolve over time.
Some Takeaways
This mental mapping, or modeling, is a continuous work in progress, much like it is for anyone else. There are numerous ideas that require further development and refinement.
Throughout the process of this project, I've encountered new revelations, such as potential research into game theory, insights into my understanding of automated harm, and even simple realizations like the capacity of my brain to process concepts (as depicted by nodes on the map) for idea extraction.
I have experimented with various tools, including Google Docs and Notion, but Whimsical has proven to be the most suitable for me. This preference is likely due to its intuitive visuals and user-friendly interface. I'm hopeful that I won't encounter any bugs that would deter me from continuing to use it or force me to switch to a different product.
I hope you find these insights beneficial despite this post being short or incomplete. If you have any questions, or would like to share your thoughts and ideas, please feel free to add them in the comments section below.
If you found this post helpful, consider clicking the share button or explore my posts on mental modeling in the age of generative AI. Additionally, if you are interested in viewing the services I offer, please feel free to browse them here. Also, consider checking out Doc Ligot’s webinars on Generative AI; the link is available here.