Presentation
Where does accountability lie when AI systems are deployed? Using Cognitive Work Analysis to address the Value-Alignment problem
SessionThursday Poster Session
DescriptionWe demonstrate a simple approach to prioritising risk from AI in terms of value-alignment. We use the well-known Human Factors suite of methods called Cognitive Work Analysis to describe a work system in terms of its purposes and the system's values (i.e., the objectives by which the system's performance is evaluated). The values are weighted (through Thurstone Paired Comparisons or Conjoint Analysis) by stakeholders affected by the system. These weights are multiplied by the strength of mapping between each purpose as it is perceived by agents in the system. This identifies which values exceed critical threshold and which agents are most critical in this.
Contributor
Event Type
Poster
TimeThursday, October 22nd5:30pm - 6:30pm PDT
Location
Similar Presentations