Skip to content

An AGI with Time-Inconsistent Preferences

James D. Miller and Roman Yampolskiy


An artificial general intelligence (AGI) might have time-inconsistent preferences where it knows that it will disagree with the choices its future self will want to make. Such an AGI would not necessarily be irrational. An AGI with such preferences might seek to modify the preferences or constrain the decision making of its future self. Time-inconsistency increases the challenge of building an AGI aligned with humanity’s values.


Lx-Number Search

(e.g. A | 000123 | 01)

Export Citation