On the Limits of Recursively Self-Improving AGI

Abstract

Abstract. Self-improving software has been a goal of computer scientists since the founding of the field of Artificial Intelligence. In this work we analyze limits on computation which might restrict recursive self-improvement. We also introduce Convergence Theory which aims to predict general behavior of RSI systems.

The Singularity May Be Near

“…Almost exactly a year before, on 23 February 2015, Roman Yampolskiy archived his paper "From Seed AI to Technological Singularity via Recursively Self-Improving Software" [2] which was subsequently published as two peer-reviewed papers at AGI15 [3,4]. In it, Yampolskiy makes arguments similar to those made by Walsh, but also considers evidence in favor of intelligence explosion.…”

Analysis of Types of Self-Improving Software

“…Finally, they showed that the problem of deliberately choosing a limited number of deliberation or information gathering actions to disambiguate the state of the world is PSPACE Hard in general [68]. This paper is a part of a two paper set presented at AGI2015 with the complementary paper being: "On the Limits of Recursively Self-Improving AGI" [69].…”

On the origin of synthetic life: attribution of output to a particular algorithm

“…In addition to potential development of novel deadly pathogens [129], genetically modified humans [130] and other organisms, we are also facing a potential runaway evolutionary process. An outcome of such process could be appearance of dangerous and potentially superintelligent robots [131], which may cause human extinction in the same way that a large number of previously existing species went extinct because of appearance of an intellectually superior specie -Homo Sapiens.…”