Alignment Traits

HMIA 2025

Class Title

HMIA 2025

"Readings"

Video: x [3m21s]

Activity: TBD

PRE-CLASS

CLASS

Outline

  1. TBD
  2. TBD

HMIA 2025

HMIA 2025

PRE-CLASS

HMIA 2025

PRE-CLASS

HMIA 2025

PRE-CLASS

Designer Friends, Experts, and Robots

HMIA 2025

CLASS

HMIA 2025

CLASS

HMIA 2025

CLASS

Our alignment cards have 10 alignment traits. In this class we will design three agents: a household robot; an expert (doctor, lawyer or other high stakes advisor); a friend. For each one you can choose N traits that it will have and 10-N traits that it will lack. First response is to indicate the traits you choose and justify what you included and what you left out. For each agent you have to characterize the failure modes associated with the traits they lack. E.g., "my robot lacks shared intentionality: it carries out commands but never quite groks what we are trying to do." "my professional lacks accountability: it makes mistakes and then denies it." "my friend lacks trustworthiness: it says it will do things and then doesn't."  Student to explain who is harmed, how detectable the failure would be (oversight question), how fixable it would be interactively (two questions: might some other trait if strong help cover for this? could we add external mechanisms (e.g., an audit trail)?.

HMIA 2025

Resources

Author. YYYY. "Linked Title" (info)