Abstract
We argue that the high energy use by present-day semiconductor computing
technology will prevent the emergence of an artificial intelligence
system which could reasonably be described as a “superintelligence”.
This hard limit on artificial superintelligence (ASI) emerges from the
energy requirements of an intelligent system more intelligent, and
orders of magnitude less efficient in energy use than human brains.
Furthermore, an ASI would have to supersede not only a single human
brain, but a large community of humans, and hence expend multiple times
the energy needed to replicate the power of a single human brain.
A hypothetical ASI would likely consume enormous amounts of energy,
possibly orders of magnitude above what is available in industrialized
society, making it impossible on energetic grounds alone. We estimate
the energy use by ASI in excess of a human brain with an equation we
term the ”Erasi equation”, for the Energy Requirement
for Artificial SuperIntelligence.
An additional challenge is the current developmental trajectory of AI
research, the majority of which is not focused on the creation of
superintelligent systems. An extremely sophisticated technology like the
hypothesized ASI will typically not emerge by chance from scattered
efforts.
Taken together, these arguments suggest that the emergence of an ASI is
highly unlikely, if not impossible, in the foreseeable future based on
current computer architectures, primarily due to energy constraints.