AI Incident Database System Updates and Recent Developments
Sean McGregor
Greetings Followers of the AI Incident Database,
It has been a minute since we reached out to you, our community of Google group subscribers. We want to catch you up on some significant developments over the past year with the AI Incident Database and here at the Responsible AI Collaborative (“The Collab” for short).
Below is a skimmable list of updates from most significant to least significant.
Features
Subscribe to Incidents: Follow individual incidents, entities associated with incidents, or the firehouse of all new reports added to the database. https://incidentdatabase.ai/signup/
Multilingual Incident Reporting: Incidents can be submitted in 133 languages and the entire website is readable in English, French, and Spanish. https://incidentdatabase.ai/blog/multilingual-incident-reporting
Enhanced Incident Discovery Tools by Textual Similarity: Discover textually similar incidents An NLP API now powers a Spatial Visualization and “Similar Incident” suggestion panel on an incident profile. https://incidentdatabase.ai/blog/using-ai-to-connect-ai-incidents
Entities Elements: Each incident profile is now tagged with the alleged involved entities - those deploying/developing the system and those harmed - which are collected in a sortable table. https://incidentdatabase.ai/blog/user-story-spotlights#featureduserstory:entitypages
Incident Responses for Entities: Entities can now submit incident responses when they are involved in an incident. https://incidentdatabase.ai/research/5-response
Research
Defined New Categories of AI Risks: Proposed a two-tiered system for indexing AI incidents and issues plus the concept of an incident “variant”. https://arxiv.org/abs/2211.10384
Defined a New Taxonomic System: Presented a taxonomic system for failure cause analysis of open source AI incidents. https://arxiv.org/abs/2211.07280
Related Activity
NIST Standards: The AIID has been incorporated into the draft NIST AI Risk Management Framework Playbook. https://pages.nist.gov/AIRMF/
ForHumanity’s GDPR Standards: ForHumanity is pointing their community at the AIID for AI audit practices. https://www.linkedin.com/feed/update/urn:li:activity:6991076106464321536/
Monthly eNewsletter: The “AI Incident Roundup” gives a digestible recap on the latest incidents in the AIID. https://incidentdatabase.ai/blog/incident-report-2022-october
In addition to major updates, we will be sending a new Annual Report in early 2023 to discuss in depth our 2022 outcomes and plans for the future.
Best,
The Collab Staff
https://incidentdatabase.ai/contact