4/16

12-1:30pm

Carnegie 113

On superintelligence: a critical examination of singularity arguments

Abstract:
In this presentation, I will undertake a critical analysis of the core arguments given
in support of the singularity hypothesis, highlighting a number of conceptual
confusions and empirical problems. After a brief review of how the notion
of intelligence in general developed in 20th-century psychometrics, I will
question the coherence of superintelligence as a concept, arguing that
it rests on fundamental misconceptions about problem solving and rationality.
I will formalize the "recursive self-improvement" argument common to all
defenses of the singularity hypothesis and show that the key needed premises
are very likely false. I will focus especially on the making of science and will
suggest that popular scenarios envisioning superintelligent machines that cure
all known diseases and achieve major scientific breakthroughs that advance
our understanding of the world are implausible and rely on obsolete Baconian
views of scientific progress.