What We Can Do Now About Superintelligence
Keywords
Loading...
Authors
Issue Date
2021-01-11
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
We can observe a clash between people that prioritize near-term issues and people that
prioritize long-term issues in AI. We show that both factions are reconcilable. For this
purpose, we use the long-term issue of superintelligence and the near-term issue of
openness in AI development. We demonstrate how openness can be charted as a multidimensional
space and traversing through this space can be modelled as a dynamic
process. This process turns out to be equilibrium dependent with a unique lock-in in
the most open corner of the openness space. Avoiding this lock-in is desirable for both
factions. Concretely, we recommend to go for maximal openness in safety techniques,
organizational goals and composition of ethics boards. Minimal openness should be
maintained in science, source code, data and development tools. Further near-term AI
issues might also exhibit undesirable lock-ins.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen