When people and algorithms meet: user-reported problems in intelligent everyday applications
summary
Analysis of 35,448 reviews from Facebook, Netflix and Google Maps uses topic modelling and Google Cloud Natural Language API sentiment scoring, yielding 34/34/22 topics then 27 actionable problems; a follow-up N=286 survey links each problem to desired support. Key findings: concrete errors (Maps location off by >1 km, Maps overriding chosen route, Netflix dropping synopses, ubiquitous “98% match” scores) map to four issue classes — knowledge base, algorithm, user choice, feedback — and users want fine-grained control, interactive explanations of outputs, an ability to turn intelligence off, and richer, human-centered feedback mechanisms.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:53Research Questions
- 5:04User-reported Problems
- 8:23Our Approach
- 10:01Online Survey
- 10:25System Support
- 14:40Conclusion
related talks
26mArtificial Intelligence? - More like Artificial Stupidity!
26mConsequences of an Insightful Algorithm
20mEmploying AI to extract actionable insights from qualitative user...
32mConsequences of an Insightful Algorithm
23mUser Experience Now Includes Machine Experience
26mDesigning Anticipated User Experiences
28mUser Research GenAI Better Together
24mDesigning for Transparency in Machine Learning