Informe de incidentes de IA de julio y agosto de 2022
La base de datos de incidentes de inteligencia artificial (AIID) contiene una gran cantidad de instancias históricas de daños causados por la IA que brindan información para el futuro desarrollo e implementación de la inteligencia artificial. Esta compilación de incidentes de IA publicada en julio y agosto de 2022 destaca los incidentes emergentes y proporciona un resumen de las adiciones recientes a la base de datos.
Desde preocupantes prejuicios y censura hasta lesiones físicas, julio y agosto vieron una variedad de titulares de incidentes emergentes en AI. Los chatbots hicieron declaraciones ofensivas y acusadoras, mientras que un robot infamemente le rompió el dedo a un niño durante un partido frente a la cámara. El software de traducción de idiomas y representación de imágenes hizo olas con contenido ofensivo.
Una inmersión más profunda en la base de datos del feed de julio y agosto revela una variedad más amplia de incidentes. Con el aumento de los despliegues y la competencia en el espacio de los vehículos autónomos, la cantidad de informes recibidos de las carreteras de los Estados Unidos ha aumentado. Las grandes empresas tecnológicas como Google y las redes sociales continúan siendo noticia con algoritmos que fallan. Además de esto, hay una variedad de nuevos informes sobre vigilancia y cumplimiento en el mundo real que muestran una variedad de prejuicios y daños a la seguridad.
Incidentes ocurridos en julio/agosto.
07/18/2022: Incidente 285: Google Lens’s Camera Translation Feature Provided an Offensive Mistranslation of a Book Title in Korean
07/21/2022: Incidente 241: Chess Robot Broke Child's Finger in Russia (5 reports)
07/24/2022: Incidente 271: Tesla Model 3 Sedan on Autopilot Killed Motorcyclist in a Rear Collision in Utah
08/07/2022: Incidente 278: Meta’s BlenderBot 3 Chatbot Demo Made Offensive Antisemitic Comments, Acknowledged by Developers as Occasional Errors
08/17/2022: Incidente 314: Stable Diffusion Abused by 4chan Users to Deepfake Celebrity Porn
08/21/2022: Incidente 303: Google’s Automated Child Abuse Detection Wrongfully Flagged a Parent’s Naked Photo of His Child
08/25/2022: Incidente 313: BlenderBot 3 Cited Dutch Politician as a Terrorist
Cómo encontrar esta colección en la base de datos:
Vaya a la "Vista Tabular" en el menú del AIID y escriba 2022-07 para los incidentes ocurridos en julio y 2022-08 para los incidentes ocurridos en agosto.
Nuevos incidentes en la base de datos
Un resumen de todos los incidentes de IA agregados a la base de datos en julio/agosto.
Los siguientes incidentes son nuevos para el AIID, pero son envíos de incidentes que ocurrieron antes de los últimos dos meses.
Lesiones riesgo de lesiones a humanos de IA
Incidente 241: Chess Robot Broke Child's Finger in Russia
Incidente 242: Manufacturing Robot Failure Caused Factory Worker's Death in India
Incidente 252: Remotely Operated Taser Drones Proposed by Taser Manufacturer as Defense for School Shootings in the US
Incidente 261: Robot Deployed by Animal Shelter to Patrol Sidewalks outside Its Office, Warding off Homeless People in San Francisco
Incidente 279: TikTok’s “For You” Algorithm Exposed Young Users to Pro Disorder Content
Incidente 281: YouTube's Algorithms Failed to Remove Violating Content Related to Suicide and Self
Incidente 286: TikTok’s Recommendation System Alleged in a Lawsuit to Have Pushed and Incentivized Two Young Girls in Participating in the “Blackout” Challenge, Resulting in Their Deaths
Incidente 290: False Negatives for Water Quality Beach Closures
Problemas de vehículos autónomos
Incidente 253: Cruise's Self Cars Allegedly Lost Connection to Their Server, Causing Traffic Blockages in San Francisco
Incidente 271: Tesla Model 3 Sedan on Autopilot Killed Motorcyclist in a Rear Collision in Utah
Incidente 289: Starship Delivery Robot Scuffed Bumper of a Resident’s Car in Texas, Allegedly Refusing to Release Footage of the Accident
Incidente 291: Tesla Allegedly Misled Customers about Autopilot and FSD Capabilities
Incidente 292: Apple’s AVs Reportedly Struggled to Navigate Streets in Silicon Valley Test Drives
Incidente 293: Cruise’s Self Car Involved in a Multiple Collision at an San Francisco Intersection
Incidente 294: Tesla Autopilot Allegedly Malfunctioned in a Non Collision in Greece
Incidente 297: EasyMile Self Shuttle Unexpectedly Stopped Mid, Injuring a Passenger
Incidente 304: Tesla on FSD Reportedly Drove into the Wrong Lane in California
Incidente 306: Tesla on Autopilot TACC Crashed into Van on European Highway
Incidente 319: Tesla on Autopilot Fatally Crashed into Parked Fire Truck in Indiana
Incidente 320: Tesla on Autopilot Collided with Parked Fire Truck on California Freeway
Incidente 321: Tesla Model X on Autopilot Crashed into California Highway Barrier, Killing Driver
Incidente 322: Tesla Model 3 Crashed into Police Patrol Car on Connecticut Highway
Incidente 323: Tesla on Autopilot Crashed into Parked Police Car in California
Prácticas policiales con IA
Incidente 244: Colorado Police’s Automated License Plate Reader (ALPR) Matched a Family’s Minivan’s Plate to That of a Stolen Vehicle Allegedly, Resulting in Detainment at Gunpoint
Incidente 245: Unverified Misreading by Automated Plate Reader Led to Traffic Stop and Restraint of an Innocent Person at Gunpoint in California
Incidente 246: Misreading of an Automated License Plate Reader (ALPR) Unverified by Police, Resulting in Traffic Stop in Missouri
Incidente 248: Automated License Plate Camera Notified Police about a Previously Stolen Rental Car that was Returned, Causing an Innocent Person to be Detained at Gunpoint in California
Incidente 255: Gunshot Detection Audio Picked up by ShotSpotter’s Algorithms Previously Admitted as Evidence of Conviction in a Murder Case in Chicago Now Dismissed due to Reports of Unreliability
Incidente 256: DUI Arrest Case in Chicago Alleged in a Lawsuit to Have Based Solely on an Alert by ShotSpotter, Citing Concerns of Unreliability
Incidente 257: Police Departments Reportedly Deployed Gunshot Detection Sensors Disproportionately in Black and Brown Neighborhoods
Incidente 260: US Department of Homeland Security (DHS)’s Opaque Immigration Software Criticized by Advocacy Groups for Allegedly Relying on Poor Data and Targeting Immigrants Via Sensitive Information
Incidente 264: AI Vehicle Speed Estimation App Denounced by UK Drivers as Surveillance Technology
Incidente 268: Permanent Removal of Social Media Content via Automated Tools Allegedly Prevented Investigative Efforts
Incidente 288: New Jersey Police Misidentified an Innocent Black Man via Facial Recognition, Resulting in Wrongful Arrest and Alleged Violation of Civil Rights
Incidente 295: Wrongful Attempted Arrest for Apple Store Thefts Due to NYPD’s Facial Misidentification
Incidente 309: UK Police’s FRT Trial Performed Poorly at Notting Hill Carnival
Incidente 310: High False Positive Rate by SWP's Facial Recognition Use at Champion's League Final
Sesgo de IA
Incidente 259: YouTuber Built, Made Publicly Available, and Released Model Trained on Toxic 4chan Posts as Prank
Incidente 262: DALL Mini Reportedly Reinforced or Exacerbated Societal Biases in Its Outputs as Gender and Racial Stereotypes
Incidente 265: Black Delivery Driver Allegedly Discriminated by Uber Eats for Excessive Facial Photo Verification Checks and Dismissed Based on False Facial Recognition Results
Incidente 275: Facebook’s Content Moderation Algorithm Incorrectly Banned Users for a Photo of Aboriginal Men in Chains as Historical Evidence of Slavery
Incidente 278: Meta’s BlenderBot 3 Chatbot Demo Made Offensive Antisemitic Comments
Incidente 280: Coffee Meets Bagel’s Algorithm Reported by Users Disproportionately Showing Them Matches of Their Own Ethnicities Despite Selecting “No Preference”
Incidente 300: TikTok's "For You" Algorithm Allegedly Abused by Online Personality to Promote Anti Hate
Incidente 312: Startup's Accent Translation AI Denounced as Reinforcing Racial Bias
Problemas de reconocimiento facial
Incidente 254: Google’s Face Grouping Feature Allegedly Collected and Analyzed Users’ Facial Structure without Consent or Notice, in Direct Violation of Illinois Biometric Information Privacy Act
Incidente 258: Australian Retailers Reported to Have Captured Face Prints of Their Customers without Consent, Potentially in Violation of Australian Privacy Act
Incidente 267: Clearview AI’s Algorithm Built on Photos Scraped from Social Media Profiles without Consent
Incidente 273: FaceApp Detected Different Genders for Similar User Photos with Slight Variation in Eyebrow Thickness
Incidente 276: Local South Korean Government’s Use of CCTV Footage Analysis via Facial Recognition to Track COVID Cases Raised Concerns about Privacy, Retention, and Potential Misuse
Incidente 298: Student Facial Recognition App Raised Ethical Concerns
Incidente 307: iPhone Face ID Failed to Recognize Users’ Morning Faces
Incidente 315: Facial Recognition Service Abused to Target Russian Porn Actresses
Algoritmos dañinos
Incidente 250: Dutch City Court Defended Home Value Generated by Black Algorithm
Incidente 251: Amazon Allegedly Tweaked Search Algorithm to Boost Its Own Products
Incidente 263: YouTube’s Content Recommendation Systems Implicated in One User’s Extreme Political Radicalization
Incidente 270: Apple Tweaked App Store Ranking Algorithms, Allegedly Resulted in Demotion of Local Apps in China
Incidente 272: Grab Tweaked Matchmaking Algorithm, Providing Preferential Treatment to Drivers Registered with Affiliated Car Rental Service
Incidente 274: Virginia Courts’ Use of Algorithmic Nonviolent Risk Assessment Failed to Result in Lower Incarceration Rates or Sentencing and Neither Increased or Decreased Historical Racial Bias in Sentencing
Incidente 282: Facebook’s Algorithm Mistook an Advertisement of Onions as Sexual Suggestive Content
Incidente 296: Twitter Recommender System Amplified Right Tweets
Incidente 302: Students Allegedly Wrongfully Accused of Cheating via Medical School's Internal Software
Incidente 305: YouTube’s Recommendation Algorithm Allegedly Promoted Climate Misinformation Content
Incidente 316: Facebook Ad Algorithm Allegedly Missed Fraudulent Ads via Simple URL Checks
Incidente 318: Facebook Recommended Military Gear Ads Despite Pause on Weapons Accessories Ads
Problemas de censura
Incidente 283: Facebook’s Automated Content Moderation Tool Flagged a Post Containing Parts of the Declaration of Independence as Hate Speech by Mistake
Incidente 284: Facebook’s Removal of Content Containing Artworks and Their Advertisements That Featured Nudity via Automated and Human Moderation Denounced as Censorship
Incidente 303: Google’s Automated Child Abuse Detection Wrongfully Flagged a Parent’s Naked Photo of His Child
Incidente 311: YouTube Auto Mistakenly Banned Women of Sex Tech Conference
Incidente 317: Bug in Facebook’s Anti Filter Allegedly Blocked Legitimate Posts about COVID
Problemas de Deepfakes
Incidente 299: Japanese Porn Depixelated by Man using Deepfake
Incidente 314: Stable Diffusion Abused by 4chan Users to Deepfake Celebrity Porn
Otros incidentes
Incidente 243: Bots Allegedly Made up Roughly Half of Twitter Accounts in Discussions Surrounding COVID Related Issues
Incidente 249: Government Deployed Extreme Surveillance Technologies to Monitor and Target Muslim Minorities in Xinjiang
Incidente 266: AI Partners Created by Chatbot App Replika Abused by Its Users
Incidente 277: Voices Created via Publicly Available AI Generation App 15.ai Stolen and Resold as NFT without Acknowledgement
Incidente 285: Google Lens’s Camera Translation Feature Provided an Offensive Mistranslation of a Book Title in Korean
Incidente 287: OpenAI’s GPT Reported by Healthcare Company as Unviable in Medical Tasks Due to Its Erratic and General Nature
Incidente 308: Atlas Robot Fell off Stage at Conference
Incidente 313: BlenderBot 3 Cited Dutch Politician as a Terrorist
Cómo encontrar esta colección en la base de datos:
Vaya a "Vista de tabla" en el menú del AIID y vea los ID de incidentes después del #241.
Apoye nuestros esfuerzos:
¿Sigues leyendo? ¡Ayúdanos a cambiar el mundo para mejor! Enviar nuevos incidentes a la base de datos, contribuir para desarrollar la funcionalidad de la base de datos e iniciar una conversación con nosotros en LinkedIn, Facebook, o Twitter.