The Question
AI-driven UX must be evaluated for control, explanation, error recovery, fairness and the consequences of being wrong.
At the centre of Auditing Intelligent User Experiences is a governance problem: how design intent survives contact with budgets, operations, maintenance and unequal power. The discussion uses Spotify AI DJ to keep that problem concrete.
Case File: Spotify AI DJ
The feature packages recommendation as a voiced, contextual sequence, changing both the interface and the perceived agency of the system.
The feature packages recommendation as a voiced, contextual sequence, changing both the interface and the perceived agency of the system. This makes the case valuable for professional study, although the surrounding conditions must remain part of the reading. The primary references for this account are NIST — Artificial Intelligence Risk Management Framework 1.0; NIST AI Resource Center — AI RMF Core; Spotify — Behind the Scenes of AI DJ; Spotify — AI DJ expands to Spanish, July 2024.
Critical Distance
Personalised engagement metrics cannot establish accuracy or fairness for different languages, cultures and low-data users.
Personalised engagement metrics cannot establish accuracy or fairness for different languages, cultures and low-data users. Without that qualification, a visible outcome can too easily be detached from the labour, regulation and ongoing support that make it possible.
Working Method
Let users skip and correct, explain meaningful factors, monitor group-level errors, provide human support, and classify high-impact decisions separately.
Let users skip and correct, explain meaningful factors, monitor group-level errors, provide human support, and classify high-impact decisions separately. The method distributes responsibility across the lifecycle instead of placing it solely on the moment of concept approval.
AURELIS Position
Intelligence should increase a user's agency, not make the system's influence harder to see.
Intelligence should increase a user's agency, not make the system's influence harder to see. The larger implication is that design leadership must remain answerable for what happens after the visual system, building, product or service enters use.
Keywords
Intelligent UX / Explainability / Control / Error Recovery
