Deceit and Power
Year: 2025
Type: book-chapter
Abstract: Recent decades have seen rapid advances in the capabilities of AI systems, and the coming decades could plausibly see this rate of progress continue or accelerate. Eventually, we might develop an AI system which is better than humans at most, and perhaps all, tasks. In this chapter, we argue that if much more sophisticated AI systems are developed in similar ways as current cutting-edge systems, two forms of undesirable behavior could plausibly emerge: those systems may attempt to deceive us and they might seek power. Insofar as this is a serious possibility, it calls for further, careful reflection and work.
Source: Oxford University Press eBooks
Authors: Richard Ngo, Adam Bales
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: University of Oxford
OpenAlex ID: W4414367696