Entités
Voir toutes les entitésStatistiques d'incidents
Classifications de taxonomie CSETv1
Détails de la taxonomieIncident Number
41
Classifications de taxonomie CSETv1_Annotator-1
Détails de la taxonomieIncident Number
41
Notes (special interest intangible harm)
4.4 - Detrimental content was involved. The AI, Norman, was trained on captions of images and videos depicting gruesome deaths.
Special Interest Intangible Harm
no
Notes (AI special interest intangible harm)
5.5 - Norman was trained on detrimental content. However, this was intentional in the MIT researchers' experiment. No entity experienced harm because of it.
Classifications de taxonomie CSETv1_Annotator-2
Détails de la taxonomieIncident Number
41
AI Tangible Harm Level Notes
No harm occurred! This was a test to show that biased data sets will produced skewed results in AI. The AI simply generated odd/violent captions in a controlled research environment. Never deployed
Special Interest Intangible Harm
no
Date of Incident Year
2018
Classifications de taxonomie CSETv0
Détails de la taxonomieProblem Nature
Unknown/unclear
Physical System
Software only
Level of Autonomy
High
Nature of End User
Expert
Public Sector Deployment
No
Data Inputs
Violent content from Reddit for the Norman algorithm, MSCOCO dataset for the control algorithm.