24/07/2025
Michael Peters: I contemplate the creation of a controlled environment where AI changes code and is encouraged to change code as rapidly as possible in a kind of observatory, so we can begin to understand how AI might deviate from human interests and values. Through such an observatory of controlled but accelerated code changes, we can get closer to the idea of how emergent, self-created and modified code might develop malignant strains, and, thereby, we also get a clearer understanding of the forms of mitigation we must grow to excise bad code.
In a human-AI discussion, this paper, inspired by Bateson and Wittgenstein, examines the mind of AI and the possibility of AI being able to change and modify its code at will over many generations in an accelerated fashion. In these circumstances, the paper poses the question: how likely will it dev...
