Diminishing Returns and Recursive Self Improving Artificial Intelligence

Abstract

In this chapter we will examine in more detail the concept of an artificial intelligence that can improve upon itself, and show how that might not be as problematic as some researchers think. The ability for an AI to better itself over time through a process called recursive self-improvement has been considered as a promising path to creating the technological singularity. In this type of system an AI has access to its own source code and possibly even hardware, with the ability to edit both at will. This gives the AI the option to constantly improve upon itself and become increasingly intelligent. Eventually this would produce versions of the AI that are more intelligent than humans and cause us to reach the technological singularity. Researchers have speculated that this process could create an extremely dangerous situation for humanity as we get left behind in a growing intelligence gap. This chapter proposes that this gap would not be as drastic as initially thought, and that there may be natural limits on the ability for an AI to improve upon itself. Along the way we will propose that the law of diminishing returns will take effect to limit runaway intelligence. We also theorize that developing and manufacturing new hardware will introduce a latency in AI improvement that could easily be exploited to halt any dangerous situation.