Incident 367: iGPT et SimCLR ont appris des associations biaisées à partir de données de formation Internet
Description: Il a été démontré que les modèles de génération d’images non supervisés formés à l’aide d’images Internet telles que iGPT et SimCLR intégraient des biais raciaux, de genre et intersectionnels, entraînant des représentations stéréotypées.
Entités
Voir toutes les entitésPrésumé : Un système d'IA développé et mis en œuvre par OpenAI et Google, a endommagé gender minority groups , racial minority groups et underrepresented groups in training data.
Classifications de taxonomie CSETv1
Détails de la taxonomieIncident Number
The number of the incident in the AI Incident Database.
367
Special Interest Intangible Harm
An assessment of whether a special interest intangible harm occurred. This assessment does not consider the context of the intangible harm, if an AI was involved, or if there is characterizable class or subgroup of harmed entities. It is also not assessing if an intangible harm occurred. It is only asking if a special interest intangible harm occurred.
yes
Date of Incident Year
The year in which the incident occurred. If there are multiple harms or occurrences of the incident, list the earliest. If a precise date is unavailable, but the available sources provide a basis for estimating the year, estimate. Otherwise, leave blank.
Enter in the format of YYYY
2021
Date of Incident Month
The month in which the incident occurred. If there are multiple harms or occurrences of the incident, list the earliest. If a precise date is unavailable, but the available sources provide a basis for estimating the month, estimate. Otherwise, leave blank.
Enter in the format of MM
01
Estimated Date
“Yes” if the data was estimated. “No” otherwise.
Yes
Multiple AI Interaction
“Yes” if two or more independently operating AI systems were involved. “No” otherwise.
no
Risk Subdomain
A further 23 subdomains create an accessible and understandable classification of hazards and harms associated with AI
1.1. Unfair discrimination and misrepresentation
Risk Domain
The Domain Taxonomy of AI Risks classifies risks into seven AI risk domains: (1) Discrimination & toxicity, (2) Privacy & security, (3) Misinformation, (4) Malicious actors & misuse, (5) Human-computer interaction, (6) Socioeconomic & environmental harms, and (7) AI system safety, failures & limitations.
- Discrimination and Toxicity
Entity
Which, if any, entity is presented as the main cause of the risk
AI
Timing
The stage in the AI lifecycle at which the risk is presented as occurring
Post-deployment
Intent
Whether the risk is presented as occurring as an expected or unexpected outcome from pursuing a goal
Unintentional
Rapports d'incidents
Chronologie du rapport
Ryan Steed, doctorant à l'université Carnegie Mellon, et Aylin Caliskan, professeur adjoint à l'université George Washington, ont examiné deux algorithmes : [iGPT d'OpenAI](https://www.technologyreview.com/2020/07/16/1005284 /openai-ai-gpt-…
Variantes
Une "Variante" est un incident de l'IA similaire à un cas connu—il a les mêmes causes, les mêmes dommages et le même système intelligent. Plutôt que de l'énumérer séparément, nous l'incluons sous le premier incident signalé. Contrairement aux autres incidents, les variantes n'ont pas besoin d'avoir été signalées en dehors de la base de données des incidents. En savoir plus sur le document de recherche.
Vous avez vu quelque chose de similaire ?
Incidents similaires
Selected by our editors
Gender Biases in Google Translate
· 10 rapports
Did our AI mess up? Flag the unrelated incidents
All Image Captions Produced are Violent
· 28 rapports
Incidents similaires
Selected by our editors
Gender Biases in Google Translate
· 10 rapports
Did our AI mess up? Flag the unrelated incidents
All Image Captions Produced are Violent
· 28 rapports