I am beginning the 2nd edition of AI Security, which will be a near total rewrite of the original, with a new focus on secrecy and research.
An analysis of the subjectivity of experience and sentience under an information theoretic approach. This essay contains significant updates to my core beliefs about consciousness.
I would like to propose a challenge for all AI researchers everywhere: joy.
Generating source code from natural language is not formal artificial intelligence, even though it sounds similar to some of the things I have described about it before.
A brief look into the history of the stored-program computer and its parallels to formal artificial intelligence.
Transparent machine learning is introduced as an alternative form of machine learning, where both the model and the learning system are represented in source code form. The goal of this project is to enable direct human understanding of machine learning models, giving us the ability to learn, verify, and refine them as programs. If solved, this technology could represent a best-case scenario for the safety and security of AI systems going forward.
This is a review of “Human Compatible” by computer scientist, Stewart Russell. The thesis of this book is that we need to change the way we develop AI if we want it to remain beneficial to us in the future. Russell discusses a different kind of machine learning approach to help solve the problem.
EXINT is a byte-aligned universal code with complete support for the integers. It is byte-order agnostic and has O(1) time performance when bounded by the system datapath, integer, or memory width.
The control problem is a question posed by Nick Bostrom on how to limit advanced artificial intelligence while still benefiting from its use. I propose an extension to the original control problem that separates it into a local and global version. I then provide proofs that the global version has no solution.