Digital democracy|Dystopian new tech
A third of all AI incidents since 2010 linked to 5 companies
As artificial intelligence is rapidly adopted by companies for automation, data analytics, and improved customer experience, it is important to acknowledge that AI is not immune to risks. AI incidents occur where there are allegations or concerns of harm or potential harm to individuals, property, or the environment involving AI systems. This week’s chart explores the connection between major technology companies and alleged AI-related incidents from AI Incidents Database¹, as well as some notable recent cases reported by the media.
Key insights
- The top 5 companies mentioned in the most AI incidents are Facebook (involved in 48 incidents), Tesla (40), OpenAI (39), Google(33), Amazon (24). Events featuring the top 5 most mentioned companies comprise over 35% of all recorded incidents.
- Facebook saw no incidents prior to 2014, after which the number of incidents grew until they peaked at 13 in 2020. One of the recent incidents alleges that Facebook's algorithm displayed job ads in a biased manner, favoring younger men over older and female workers for blue-collar positions². Overall, Facebook is featured in 9% of all AI incidents.
- Tesla went without an incident until 2016, after which the yearly number of incidents stayed at around 7 until now. In the most recent incident, video captured a Tesla on FSD Beta 11.4.1 reportedly failing to yield to a pedestrian, despite the presence of a state law sign instructing vehicles to yield to pedestrians within crosswalks³. Overall, Tesla is featured in nearly 8% of all AI incidents.
- OpenAI, other than a single incident in 2016, was not involved in incidents until 2019. Since then, and especially in 2023, the number of OpenAI-related incidents has been increasing drastically. An infamous recent incident involves the Italian Data Protection Authority banning OpenAI for personal data collection concerns⁴. Overall, OpenAI is featured in slightly less than 8% of all AI incidents.
Methodology and sources
This chart is based on the AI Incident Database, which collects, verifies, and classifies user-reported events in which harm was either caused or almost caused by AI. An AI incident refers to an alleged occurrence or potential occurrence of harm to individuals, property, or the environment in which an AI system is involved. The data used includes incidents that are dated between 2010 and the date of collection, which was June 16th, 2023. When calculating the total number of incidents the highlighted companies were mentioned in, we excluded instances where the company was involved only through being harmed by AI. Instead, we counted incidents where the company was a developer and/or deployer of the incident-causing AI. Note that just because a company is involved in an incident, it does not mean that they directly caused the incident.
For the complete research material behind this study, visit here.
